THE ROLE OF THE COMPUTER (Sep, 1952)
This is the third in a series of 5 articles I’ve scanned from an amazing 1952 issue of Scientific American about Automatic Control. Discussing automatic machine tools, feedback loops, and the role of computers in manufacturing and information theory, these are really astounding articles considering the time in which they were written.
- Part 1 – Automatic Control
- Part 2 – Feedback
- Part 3 – The Role of the Computer
- Part 4 – Automatic Machine Tools
- Part 5 – Information
THE ROLE OF THE COMPUTER
The multifarious control loops of a fully automatic factory must be gathered into one big loop. This can best be done by means of a digital computing machine
by Louis N. Ridenour
IF THE thermostat is a prime elementary example of the principle of automatic control, the computer is its most sophisticated expression. The thermostat and other simple control mechanisms, such as the automatic pilot and engine-governor, are specialized devices limited to a single function. An automatic pilot can control an airplane but would be helpless if faced with the problem of driving a car. Obviously for fully automatic control we must have mechanisms that simulate the generalized abilities of a human being, who can operate the damper on a furnace, drive a car or fly a plane, set a rheostat to control a voltage, work the throttle of an engine, and do many other things besides. The modern computer is the first machine to approach such general abilities.
Computer is really an inadequate name for these machines. They are called computers simply because computation is the only significant job that has so far been given to them. The name has somewhat obscured the fact that they are capable of much greater generality. When these machines are applied to automatic control, they will permit a vast extension of the control artâ€” an extension from the use of rather simple specialized control mechanisms, which merely assist a human operator in doing a complicated task, to over-all controllers which will supervise a whole job. They will be able to do so more rapidly, more reliably, more cheaply and with just as much ingenuity as a human operator.
To describe its potentialities the computer needs a new name. Perhaps as good a name as any is “information machine.” This term is intended to distinguish its function from that of a power machine, such as a loom. A loom performs the physical work of weaving a fabric; the information machine controls the pattern being woven. Its purpose is not the performance of work but the ordering and supervision of the way in which the work is done.
There are in current use two different kinds of information machine: the analogue computer and the digital computer. Several excellent popular articles have discussed the characteristics of these two types of computer; here we shall briefly recall their leading properties and then consider their respective possibilities as control mechanisms.
THE ANALOGUE machine is just what its name implies: a physical analogy to the type of problem its designer wishes it to solve. It is modeled on the simple, specialized type of controller, such as a steam-engine governor. Information is supplied to the machine in terms of the value of some physical quantityâ€”an electrical voltage or current, the degree of angular rotation of a shaft or the amount of compression of a spring. The machine transforms this physical quantity into another physical quantity in accordance with the rules of its construction. And since these rules have been chosen to simulate the rules governing the problem, the resulting physical quantity is the answer desired. If the analogue machine is being used as a control device, the final physical quantity is applied to exercise the desired control.
Consider, as an example, the flyball-governor pictured on the cover, whose purpose is to hold a steam engine to a constant speed. We notice, first, that information on the engine speed reaches the governor in the form of the speed of rotation of a shaft, while the output of the governor is expressed as the motion of a throttle which is closed or opened as the whirling balls rise or fall. Second, we notice that the relation between these two physical quantities is determined by the actual construction of the governor. The design of the controller has been dictated by its function. In contrast to the analogue machine, a digital machine works by counting. Data on the problem must be supplied in the form of numbers; the machine processes this information in accordance with the rules of arithmetic or other formal logic, and expresses the final result in numerical form. There are two major consequences of this manner of working. First, input and output equipment must be designed to make an appropriate connection between the logical world of the digital machine and the physical world of the problem being solved or the process being controlled. Second, the problem to be solved must be formulated explicitly for the digital machine. In the case of the analogue machine, the problem is implicit in the construction of the machine itself; construction of a digital machine is determined not by any particular problem or class of problems but by the logical rules which the machine must follow in the solution of any problem presented. Thus far the need for specialized input and output equipment, more than any other factor, has restricted the role of digital information machines to computing. In a computation, both the input and the output quantities are numbers, so the most rudimentary equipment will suffice to introduce the problem and register the result. There is no need (as there would be in a control application) to transform various physical quantities into numerical form before submitting them to the machine, or to transform the results of the calculation into a control action, such as moving a throttle. To use a digital information machine as a computer it is necessary only to provide (1) an input device such as a teletypewriter, which with the help of a human operator can translate printed numbers into signals intelligible to the machine, and (2) an output device such as a page-printer or electric
typewriter, which can translate the signals generated by the machine into the printed numbers intelligible to men. Even this simple requirement, however, has not always been well met by the designers of information machines.
When a digital information machine is to be used as an instrument of controlâ€”and we can confidently expect that this will eventually be its major roleâ€” the design of input and output equipment becames a more formidable task. While it is true that the structure of the machine itself depends on principles of logic rather than on the nature of its application, this is by no means true of the input and output elements. The input devices, or receptors, can use standard elements for receiving the program of instructions, but they must also receive data specifying the state of the particular process being controlled, and
for this the detailed design will vary widely from one application to another. Similarly the effectors, which exercise the machine’s control, must be designed in terms of the nature of the process or device being controlled.
In comparing digital and analogue machines as instruments for automatic control, we observe, first, that for simple control applications the analogue machine is almost always less elaborate than a digital machine would be. Even the most elementary digital machine requires an arithmetical (or logical) unit, a storage unit, a control unit, receptors and effectors. For simple problems, this array of equipment is wastefully elaborate. In contrast, an analogue machine need be no more complicated than the problem demands. A slide rule, for example, is a perfectly respectable information machine of the analogue type. The analogue machine’s ability to do simple work by simple means explains its current predominance in the field of automatic control. The whole control art is so new and so little developed that most of the problems thus far tackled have been of a rather elementary nature.
As the control task becomes more complex, however, the analogue machine loses its advantage, and we begin to see a second fundamental difference between the two types of machine. The analogue machine is a physical analogy to the problem, and therefore the more complicated the problem, the more complicated the machine must be. If it is mechanical, longer and ever-longer trains of gears, ball-and-disk integrators and other devices must be connected together; if it is electrical, more and more amplifiers must be cascaded. In the mechanical case, the inevitable looseness in the gears and linkages, though tolerable in simple setups, will eventually add up to the point where the total “play” in the machine is bigger than the significant output quantities, and the device becomes useless. In the electrical case, the random electrical disturbances called “noise,” which always occur in electrical circuits, will similarly build up until they overwhelm the desired signals. Since “noise” is far less obtrusive than “play,” electrical analogue machines can be more complicated than their mechanical equivalents, but there is a limit. The great machine called Typhoon, built by the Radio Corporation of America for the simulation of flight performance in guided missiles, closely approaches that limit. It is perhaps the most complicated analogue device ever built, and very possibly the most complicated that it will ever be rewarding to build.
The digital machine, on the other hand, is entirely free of the hazards of “play” and “noise.” There is no intrinsic limit to the complexity of the problem or process that a digital machine can handle or control. The switching system of our national telephone network, which enables any one of 50 million phones to be connected to any other, is a digital machine of almost unimaginable complexity.
THE THIRD important difference between analogue and digital machines is in their accuracy potential. The precision of the analogue machine is restricted by the accuracy with which physical quantities can be handled and measured. In practice, the best such a machine can achieve is an accuracy of about one part in 10,000; many give results accurate to only one or two parts in 100. For some applications this range of precision is adequate; for others it is not. On the other hand, a digital machine, which deals only with numbers, can be as precise as we wish to make it. To increase accuracy we need only increase the number of significant figures carried by the machine to represent each quantity being handled. Of course in a control operation the machine’s over-all precision is limited by possible errors in translating physical quantities into numbers and vice versa, but this does not alter the fact that where high precision is required, a digital machine is usually preferable to the analogue type.
There is a fourth respect in which the two machines differ. An analogue machine works in what is called “real time.” That is, it continuously offers a solution of the problem it is solving, and this solution is appropriate at every instant to all the input information which has so far entered the machine. If the machine is doing a mathematical problem, for example, it need not formulate explicitly the equations to be solved and then go through the steps of solving them, as a digital machine would have to do. The equations are inherent in the very structure of the machine, and it solves them by doing just what it was built to do. It can thus respond promptly to changing input data, and offer an up-to-date solution at every moment. This property of working in “real time” is very important in most problems of automatic control. An autopilot flying a plane must respond at once to an attitude change resulting from a gust of wind; the most precise information on how to adjust the flight controls will be worthless if it comes 30 seconds too late. Since a digital machine works by formulating and solving an explicit logical model of the problem, it can work in “real time” only if the time it requires to obtain a solution, given new input data, is short compared with the period in which significant changes can take place in the system being controlled. Present-day digital machines can achieve this speed for many important problems-flight control of aircraft, for exampleâ€” but they are not yet fast enough to handle all the “real-time” problems that we should like to turn over to them. It has been estimated that the fastest existing digital machines are some 20 times too slow to deal with the problem of simulating the complete flight performance of a high-speed guided missileâ€”the problem that Typhoon was built to handle. As development proceeds, the operating rates of digital machines can be expected to increase rapidly.
WE SEE, then, that both analogue and digital machines can be used for automatic control, and each has advantages in its own sphere. For simple applications in which no great precision is required, an analogue controller will usually be preferable. For complex problems, or problems in which high precision is required, a digital controller will be superior. Where “real-time” computations must be made, analogue machines are almost always used now, though digital machines are beginning to achieve speeds that fit them for this type of application.
All this refers to the present state of the art of automatic control. What can we guess about the developments to come?
The simple specialized analogue controllers already in use will surely be extended to wider application. But the most significant and exciting prospects reside in the digital machine. We can expect that it will soon open up a new dimension of control. The meaning of this prediction can be admirably illustrated in terms of the highly instrumented catalytic cracking plant which Eugene Ayres has described in a preceding article.
Mr. Ayres tells us of a plant in which there are some 150 different analogue controllers, each governing some aspect of the continuous process that the plant performs. Several hundred indicators on a central control panel offer the most detailed information on system performance. Many of these indicating instruments also provide continuous recordings. Manual controls which can override any automatic controller are present for use in emergency. The instruments and controls have been arranged on a flow diagram which simulates the organization of the plant and helps the human operator to find his way through the complexities of instrumentation. And the most important process-controls are adjusted manually according to the results of a periodic product analysis.
Clearly the human operator is still the master of this “automatic” plant. However elaborate the instrumentation, the readings of the instruments are still presented to men; however competent the automatic controllers, provision for human veto of their action is built into every one of them. Men are expected to meet emergencies, and to take control under “conditions of unstable equilibrium such as starting up or shutting down.” The cracking plant is automatic only when the unexpected is not happening; in times of stress it falls back on human control, and its whole design is dictated by this necessity.
To this scheme there will soon be added end-point controlâ€”continuous adjustment of the main process-controls on the basis of a continuous product analysis within the system itself. This modification will improve performance, but it will leave the situation essentially as it was before: more routine responsibility will be given to machinery, but the human supervisor will still be vital to proper operation.
THE DIGITAL information machine, employed as an instrument of supervisory automatic control, can change this picture radically. Since such a machine can be instructed to perform any set of logical operations, however complicated, it can be programmed at the outset to react in emergencies precisely as would a well-instructed human operatorâ€”and it can react at least a thousand t