Conversely, should we give greater consideration to special purpose equipments designed to do a particular job more efficiently? The se need not be single-purpose machines which cannot survive changing business conditions, but rather might handle a continuing common task like Inventory Control or Production Scheduling effectively and economically.

This may give us a different viewpoint on the evolution of computer design. We can look at our own applications not in terms of how they might be done on a universal or generalized machine, but rather what is the inherent structure of the manufacturing problem, what are the true needs, its real requirements? With this approach, we are more inclined to look at the problem from a systems analysis standpoint, rather than from the stereotyped let's-mechanize-what we-have-today point of view.

One simple analogy is that of railroad freight cars. Years ago, there were just a few kinds of freight cars: like the flat car and the box car. But then some thoughtful transportation men started analyzing true function and actual needs. So what has been the result? Instead of offering only the standard cars into which all freight had to be fitted, a number of special purpose equipments were developed to perform a certain Job in the best possible way.

Another interesting lesson can be learned from the freight cars analogy; all these cars, though designed to do a specific function, are planned for integration, they can be joined by standard couplings in any sequence of reasonable length to make up a whole freight train, together they may travel throughout the country on standard gauge track.

23·

Next let's look at specific phases of the over-all manufacturing system. Here is a list of some of our manufacturing functions:

1. Master scheduling

2. Requisition processing

3. Parts explosion

4. Inventory control

5. Purchasing

6. Motion time standards

7. Dispatch rules determination

8. Automatic dispatching

9. Quality analysis

10. Manufacturing operations and feedback

Next, look at the other side of the coin - at the other dimension of this concept. We feel that for manufacturing processes there are four different types of computer equipment which might be used to solve these particular problems:

1. Batch Computation - some processes by their very nature have to be done in a group. You have to accumulate a certain number of transactions or activities before going through the computational process. The weekly payroll or quarterly financial statement are typical examples, although even here the "real time" advocates point out the prospects for doing a substantial part of these jobs as the data become available.

2. Real Time Data Processing implies that the computer will keep in step with the actual events going on in the factory. Like the famous Norman Rockwell painting "The Gossips", where the direct reporting is carried on as rapidly as practical. In fact the last picture shows the feedback to the originator.

24·

4. In Simulation a computer is used to act like an actual operating process. For example, it might be a true imitation of the teal system simply speeded up or directly scaled down like a model train.

Let's put these two sets of ideas together, manufacturing functions and types of computers, and then look at the whole picture. For example, the double check on master scheduling indicates it might best be performed on a batch basis, though it could also be done on a real-time computer. Simulation and machine control seem to have little application to waster scheduling, s0 on through the chart.

Let's start with the first of these areas, the one with which we are most familiar, Batch Computation. This can be applied either to planning the system or actual calculation of operating statistics. For example, in systems planning linear programming could help determine the best traffic patterns, optimum warehouse locations and most profitable product mix; it might even be used for directing over-all inventory policy or manufacturing cycle time determination. Similar computational areas are: ABC inventory analysis, collation chart preparation, and long-range load analysis.

One of the interesting operating problems which has been ingeniously solved on a batch basis is that of inventory control - Otis Elevator has purchased a special inventory control computer from the Nuclear Development Corporation of America (NDA). This machine is designed to operate each night to incorporate all the transactions and activities which occurred during the day. Processing some 20,000 transactions against 35,000 stock accounts each day would have required an extremely large random access memory. The solution was a newly developed tape processing machine which first sorts all the transactions into sequence and then runs the new transaction tape and waster perpetual inventory tape in parallel to recalculate the current status.

Another process which might lend itself to batch computation is that of setting motion time standards. I can conceive of a machine tool wired to do its own time study, keeping track of its own movements and the elapsed time for each. But there might be a better way of semi-automatically setting time standards, using predetermined time standards like M. T. S. (motion time survey). We could have the time study man use a special keyboard to record the various basic motions which need to be performed to accomplish a particular task. This device could prepare a punched paper tape which would then be fed into a small special purpose computer having in its memory the standard times for each work element; then with merely a summary operation these times could automatically be added together and another time study would be ready for review.

25.

While this is fine for certain departments' problems, it is of course too small to handle many others. The answer may be the PRODUCTRON, Sr. which would be designed particularly for the multitime period problem. One of the dangers here is that for each additional period covered, the problem becomes more complex as the square of the number of time periods. For example, if we use an index of one for the complexity of the single period problem, a five-time-period situation would be five squared or 25 times as complex.

In discussing real time data processing, let us consider the area that has received primary attention in the past - that of machine and process operations control. We in Manufacturing feel that the large business machine companies have in the past misunderstood the true nature of the manufacturing or production control data processing function. Production is a dynamic function, one that lives and breathes in rhythm with the actual physical processes in the shop. Many of our functions must either keep pace or fall hopelessly behind. This dynamism is an important concept, for it implies that machines must frequently operate in real time, on the line; that machines must make decisions; that machines must keep track of progress; that machines must talk to machines. This dynamism plus manufacturing's central location in the business system leads to a unique problem. Since virtually all the business systems' inputs - requisitions, engineering lists, and planning records - pass through manufacturing hands, we are naturally concerned with their accuracy. We have a direct interest in the machine language used for their preparation and ease of file maintenance. In turn, we work on this data to prepare products for sale; we schedule, load, prepare purchase orders, factory paperwork, dispatch the jobs. move them into and out of stock rooms, pack and ship them, and prepare shipping papers. The output of our manufacturing processes feed to finance for their periodic reviews and analyses. We have the responsibility and the necessity for transmitting costs, quantity, and quality information.

Looking at this whole picture then we feel that unless manufacturing can keep up with the shop processes, we are inherently putting a restriction, a barrier in the path of the other functions' efforts to mechanize. If, for instance, we only provide cost data once a week, the cost system can be no better than a weekly process.

One specific "real time" idea which demonstrates clearly the kind of approach we will make in manufacturing automation is that of dispatch control. This will illustrate how progress can be realized reaching ever higher and higher levels of automation. The first step might well be a specialized computer designed to help the dispatcher by calculating instantaneously relative priorities of all

26.

be possible to build up these decision processes into a machine and hook up a device s0 that the machine operator could insert his own particular key or ,late and have it printed out for him a voucher and work instruction. This step will eliminate the dispatcher.

The final step might come with the elimination of the machine operator so that the dispatch device communicates directly with the machines themselves, picking out the Job with the best priority and issuing computer sensible instructions and actually starting, directing operations and stopping the machine.

The third area for computer equipment is that of direct machine control, which will actual1y operate machines, giving them their instructions, measuring their performance and correcting their adjustment.

An example with which most of you are familiar is the Giddings and Lewis tape controlled milling machine. GE built the tape control device for this machine, which is used for contouring of airplane wings whose shapes are very complicated. TO be added in the future is some sort of a control feedback of actual performance to insure that the machine is following the tape recorded plan.

Another area is that of quality control. We visualize testing machines more elaborate than those being built today which are programmed to perform all the tests required and to reject those falling outside of control limits. In addition, it will analyze these test results in terms of causative factors and prepare necessary control charts, signal the faulty input areas, adjust those

27.

There will also be superimposed en the factory a comprehensive feedback system. This might be similar to I.B.M.'s newly announced Automatic Production Recording System, which enables certain devices, attached to machines Or placed on conveyer lines, to signal a central location on quantity counts, liquid flow, voltage variance, weights, etc. All these figures, even if measured in analog form, can be converted to digital language for computer use. We might use this same idea to feedback quality data, cost data, and process performance statistics.

The fourth and possible the most interesting area is that of simulation whereby the computer can be used to "act like" some operating process. For instance, we can visualize a computer that will perform in a manner similar to a manufacturing Job shop so that we can test dispatch rules in a fraction of the time, without disrupting normal day-to-day operations. We can simulate product performance for quality control testing and evaluation as was done for the Jet Engine Department by the General Engineering Laboratory. We can test various empirical management decision rules to see if external changes will influence the likelihood of successful operation. This same type of simulator can be used for predicting future sales trends and for determining optimal inventory balances where sales vary erratically. Production leveling may find a practical solution with simulated trials of various manufacturing plans. Another idea is to simulate the traffic arrival pattern to determine the number of truck docks and receiving employees required at various times.

Simulators, while offering many unusual opportunities, may require extensive random access memories, since we are dealing with time dependent problems. One simple example will show what I mean.

The total number of different ways to arrange a schedule of 15 Jobs of 5 operations each is more than 10⁶⁰ (a one with 60 zeros). Even if 99 per cent of these arrangements are impossible or not feasible, we still must deal with the remaining One per cent or 1058 trials. This is obviously impractical in spite of 70S's, UNIVAC's et alia. So we turn to the cut and try approach and our method is to try basic, even intuitive, formulas or rules and study whether there is a significant variation rather than trying to solve the problem analytically.

The use of simulation rests heavily on the refinement of new operations research techniques. This means that we in manufacturing must become more professional in the adaptation of these techniques

We are working on one such problem today, that of dispatch rules determination, This might ase a mixed analog and digital device with both potentiometers and magnetic cores which would enable us to run speeded up tests on shop performance, to tryout various rules, different formulas, a variety of techniques to see which would result in greater efficiency if installed in the factory, We hope to build a model on a small scale, a model of manufacturing dispatching business. This model must act enough like the shop itself to allow us to extrapolate from the test results obtained. We will try, instead of using the conventional required finish dates

For control, to use slack times and flow factors to evaluate, as best we can, the interactions of one Job on another, By proper use of statistical inference we should be able to prove or disprove certain assumptions concerning the behavior of the manufacturing system.

We are personally enthusiastic about the potentialities of automatic data processing and hope that our own new Manufacturing Control Automation Systems Project, which we are Just starting, will serve to some extent as a product planning function for the Industrial Computer Section, and conversely, we enter into this project with greater confidence of success, knowing that the need for this kind of equipment has been recognized and the Industrial Computer Section established. We feel this area of development is urgent, that present equipments do not cover the span of industrial requirements and that a mutual study of these problems will be most rewarding.

29·

An integrated systems approach, an over-all business concept must be the basis for economical, manufacturing automation. Versatility and flexibility should be inherent in the automata designed We must provide for growth; we must provide for change. We must provide for adapting computers to specific situations, and yet with all this, the total installation costs must be economically justified. The answer here may be that we shall have to think in terms of savings to the whole business; we shall think in terms of departments' profits, rather than in terms of direct out-of-pocket cost of machine applications.

To quote Dean Wooldridge, President of the Ramo-Wooldridge Corporation. "The lesson that has been learned by government at the cost of many millions of dollars is that the integrated systems approach is the proper approach to complex systems; this principle is just as applicable in business and industry."

30.