(1) FUTURE TRENDS IN FACTORY AUTOMATION (2) TECHNOLOGY FORECASTS FOR CIM
Robert U. Ayres International Institute for Applied Systems Analysis Laxenburg, Austria
RR-89-10 August 1989
Reprinted from Manufacturing Review, Volume 1, Number 2, June 1988 and Volume 2, Number 1, March 1989.
INTERNATIONAL INSTITUTE FOR APPLIED SYSTEMS ANALYSIS Laxenburg, Austria Research Reports, which record research conducted at IIASA, are independently reviewed before publi cation. However, the views and opinions they express are not necessarily those of the Institute or the National Member Organizations that support it.
Reprinted with permission from Manufacturing Review, Volume 1, Number 2, June 1988 and Volume 2, Number 1, March 1989. Copyright© 1988 and 1989 American Society of Mechanical Engineers.
All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage or retrieval system, without permission in writing from the copyright holder.
Printed by Novographic, Vienna, Austria iii
FOREWORD
These two papers form a natural package. Both appeared in earlier versions as IIASA Working Papers for the CIM Project. The second one also reflects a major thrust of the project, which culminated in the summer of 1989. (A second paper by J. Ranta reflecting on that effort will appear separately.)
F. SCHMIDT-BLEEK Leader Technology, Economy, and Society Program
MANUFACTURING PERSPECTIVE Future Trends in Factory Automation
ROBERT U. AYRES Department of Engineering and Public Policy, Carnegie-Mellon University, Pittsburgh, PA 15213
This paper is a review of contemporary manufacturing technology. from both a U.S. muf world perspective. It emphasizes the historical background of the current trends toward comp1aerized a:utomation in terms of the increasing societal dem.ands for performance. whic11 in turn genera.Les requirements for ever grea.ter coniplexity and precision. This is the root <~f the "qual.ity cri3is." The author belie·ves thlit the ne.xl industrial revolution ·1.vU.l present a fund a mental sh i}1 from the use of h1.l1nan workers as "micro" decision makers (machine controllers) in fadors lo the use of "smart sensors" for this purpose. The paper elaborates som.e of the more spec~fic implications.
DISCRETE MET AL PARTS MAl\TUF ACTURING However, in our diverse economy it is natural that some TECHNOLOGY (c. 1975) items, especially durable goods, are needed in small numbers and seldom replaced, while others are needed in larger he choice of manufacturing technology at present is numbers. The distinction most commonly made is between highly dependent on the scale of production. But batch and mass production. The value added of the U.S. some items, such as connectors, have long been manufacturing sector in 1977 was about equally divided standardized and mass produced in enormous between these two categories, as shown in Fig. 1. Batch numbersT whereas other items, such as auto engine plants or manufacturing can be further divided into one-of-a-kind space shuttles, are virtually custom made. The cost per unit (piece) or very small batches and medium to large batches, of items made in large numbers can be as little as one as indicated in Fig. 2. Unit cost differences arise from hundredth of the unit cost of the same item made individu several factors. In the first place, small volume production is ally. For example, the 600 distinct machining operations inherently much more labor intensive than large volume required for a V-8 cylinder block in 1975 cost around $25 in production because fewer functions are automated. Table 1 a mass production plant and only required 1 min. productive Metalwork ing Machin'!"-' labor time. By contrast, the same 600 machining operations TOTAL VA = $58.S,000 ,000 ( '977$ ) (2 6% of Du,abl€ Bate~.) carried out by skilled machinists in a job shop would have D required 600 min. of machinist labor and cost at least $2500 2 .1f--~ Other Botch-Produc-".". [ l, 2]. One of the ironies of this situation is that the special D of Durable aotch) ized machinery typically used in mass production, for exam -~ "'\ Other Heavy ple, the large transfer lines and multispindle drilling and Mac hi nery boring machines, are themselves customized, one-of-a-kind D (6 63 al Durable Botch) ~'lk @H u©~®~» \ (T o~ol B::: :ch "' 553 cl Durob l ~s\ investments. 1 If auto engine plants could be mass produced 26.7% as auto engines are, the capital costs would drop by as much 28.8% \ Mo ss Praduc ed Durables as 100-fold. D ( 4o! .77. o f Duroble9 ) A more recent example is instructive: helical rotors for compressors, as first produced in Sweden by hand in the ~ Non-Dura ble Batch 1950s, required up to 200 machinist hours. By 1967 this had ~ (40. 33 of No n -Durables) fallen to 6 h, by 1978 to 65 min. and in 1979 to 26 min. [3]. None of these advances utilized numerical control, which ~ Non - Du ra ble Moss Ll (59.77. of Non-Durables} entered the picture subsequently.
1The design of an auto engine plant , capable of producing 120 units per hour for 20 yea rs, requires about 60,000 engineering man-hours 12]. FIG. I. Distribution of manufacturing value added (Source: Miller, 1983 (11))
© Copyright 1988 American Society of Mechanical Engineers 93 Type o f production :
Batch Ma ss
34% - holidays, vacations
0%_j______J_ _ L______L _ L ______(a) 1003,~------.-~-----~ 44% - incomplete use of 2nd and 3rd shifts
0%
~ ;_arge ,, complex 1-10 10-300 Over 200 _§ part ~ Small "Ci. simple 1-300 300-15,000 Over 10,000 ....> part
Typical Aircraft, Mar ine engines, Autos, products large turbines . large electric fasteners, centrifuges motors, tractors small appliances 28% - plant shu td own Typical Manual , NC with auto machines stand·alonc pan.hand li ng, Transfer, NC machining cell, dedicated Hex mfg. svst. special machine
FIG. 2. Characlerislics of melal producl manufacluring (Source: (b) American Machinist, 1980 [4]) shows the progressive elimination of manual operations by automated equipment of increasing degrees of sophistication. Another re ason for the big difference in unit cost between mass production and piece production in a job shop is that machines can be utilized much more efficiently in the former case. Differences in typical machine utilization patterns as a function of scale of production are shown in Fig. 3. It is noteworthy that in a typical job shop machines a re only tended about 20 % of the time and only 6% of the
'fablf' I. Comparison of manual manufaduring steps elimination by various dP.grees of automation 27% - plant shut (Sour<~: General Ac•:ount.ing Office 197.6: p. 38) down (2 week shut down, Sundays, Production Methods 13 holidays) Stand- alone Machining Step Conventional NC center FMS (c) I. Move workpiece M M M c 22% - productive cuning to machine 2.. [.oad and affix M M M c workpie-ce on machine 3. Select an.d M M c c insert tool 4. Estab"llsh, and M c c c set speeos 5. Control cutting M c c c FI G . 3 (a) Low-volume manufacluring, (b) mid-volume manufacluring, 6. Sequenee tools M M c c (c) high-volume manufacluring (Adapled from: American Machinist, and motions 1980 [4]) 7. Unload part M M M c from mac~ine time is used for productive cutting. This contrasts to 22% M = manual op-eration C = computer-controller operation productive cutting in a mass production facilit y [4] .
94 Manufacturing Review vol I, no 2, June 1988 The key characteristic of mass production is that it macri •n.n9wste,., achieves low unit cost by extreme specialization of equip woo .-_ y~--T ment. For automobile engine or transmission production, the heart of the plant would consist of a set of giant multiple spindle machines, generally with between l 00 and 1000 tools, mainly drills, cutting simultaneously. The spindles are "" ·· ~Cns1 oe• clustered in groups (or stations). v; un•t Con$!rucnor> m~ch •n r r v 0 u The mechanical requirements are exacting. Each of the Q.> 10 • .2' spindles in each station must be permanently positioned very ;;; precisely with respect to all the others. All the spindles in 1! each group must also be exactly synchronized, so that the 1 0 . resulting holes are not only parallel but also drilled to the Sm• ll t•uc l.. s exact same depth. Drill speeds must be precisely predeter t> •c vr l ~ mined for the same reason. The necessary simultaneity can 01 - f -t· 1···· 1 be achieved by mechanically linking all the spindles at each 10 1 10:? 103 104 105 106 107 station, via elaborate gear trains, to a single drive shaft. Or, number of products per month batch production mass production separate drive motors can be subject to a common controller. Workheads are either ON or OFF. Machines are designed to FIG. 4. Costs and automation versus volume (Source: Author, operate at a fixed speed over a fixed cycle that is optimum adapted from various sources) for the design application. Large groups of machines (sections) are also synchron ously linked together mechanically via indexing transfer lines. solid-state monolithic integrated circuits and large-scale They are not individually controllable, hence not easily integration (LSI, VLSI) modern computers are of the order adaptable to other design specifications. If the product being of 100,000 times less error prone than human workers [8]. In manufactured becomes obsolete the custom-built manufactur effect, the direction of technological change (in the industri ing equipment is likely to be scrapped, since adaptation is alized countries, at least) is inexorably toward the substitu difficult or impossible. This rigidity explains the otherwise tion of computers and smart sensors for humans in all puzzling fact that U.S. automobile manufacturers in the phases of the manufacturing process. 1970s were not able to convert plants from eight-cylinder engine production to six-cylinder-engines. For the same :\llCHOELECTHO'.\llC' THE'.\/ BS reason, a plant dedicated to making conventional transmis It is fairly obvious that computers and smart sensors, in the sions and drive shafts for large rear-wheel-drive vehicles sense used above, must be based on the technology of could not be converted to manufacturing transaxles for microelectronics. The same is also true, incidentally, of front-wheel-drive cars. Programmable Controllers (PCs), which are another key The economics of such special-purpose automation, as ingredient of advanced forms of automation. compared with other modes of manufacturing, is indicated Unit costs (i .e., costs per gate or bit of memory) have schematically in Fig. 4. The curve represents the cost moved down essentially in step with the number of elements minimizing choice as a function of scale of production. per chip. Chips are made by a complex, but highly auto Evidently, fixed costs are very high but variable costs (mostly mated and capital-intensive process in which direct (i.e., labor) can be minimized. Thus hard automation pays off "hands-on") human labor plays almost no role. Jn fact, in when production volumes become large enough. modern plants humans must be rigorously kept away from While the mechanization of parts manufacturing has not the actual manufacturing steps because of the danger of yet reached any physical limits, its contributions to gains in contamination. The major elements of cost for electronic manufacturing productivity were diminishing by the 1970s. devices are now the design and the specialized capital equip Even within the manufacturing arm of a big "systems ment used in manufacturing. 2 integrator," logistics, assembly, and quality co_ntrol3 now The marginal cost of production is virtually the cost account for, by far, the biggest share of the real costs of of materials only, which is negligible. The relative ease of manufacturing-quite apart from indirect costs of finance, copying successful designs explains why chipmakers try to marketing, personnel management, and the like. To reduce amortize each new product in a very short time and why costs significantly-below present levels-a completely new vicious price cutting tends to rapidly follow the initial technology of production permitting substitution of "smart introduction. The 256K RAM chip, first introduced to the sensors" for "hands-on" labor, and coordination of all market in 1983, is now selling at $4 or $.00156 per bit. Price activities by computer, seems to be needed. This will become trends for logical functions are shown in Fig. 5 and impacts increasingly manifest over the next two decades. on system costs are summarized in Table 2. Jn relative terms, The long-range imperative, of course, is to design the costs have declined by a factor of about one million since human worker out of the production system . Thanks to the era of vacuum tubes. It scarcely needs to be said that rapid technological 2Thc co st of " logistics" including materials handling, storage, in ventory control, and shipping, accounts fo r over 27 % o f manufacturing value added improvements and corresponding cost reductions seem in Sweden 151. A British study concluded !hat 19 .5% of industrial labor co sts virtually assured for the next decade, at least, by the are attributable to materials handling alone. For the U.S . logistics accounts enormous research and development resources currently for 22. 5% of manufacturing value added 161 . being invested in these areas. A number of major new l lncluding in spection, moni1oring, rewo rk, el c. One survey showed 1ha1 technologies, including optical devices and organic chemical qualit y con1 rol averaged 5.8% of sales, or roughl y 11 - 12% of value added [71. molecular (molecutronics) devices, now appear to be feasible and perhaps immanent.
Ayres: Future Trends in Factory Automation 95 50 different tools. They are thus ideal for small batch production of very complex metal shapes, for example, for 1---V~~uurn tubes \ 10 ------the aerospace industry. ~--\ Application of iarge-scale integration (LSI) technology in ~ the early 1970s, brought the costs down while simultaneously (l) 1 iii SSl /MSI Ol providing for vastly increased capability. A minicomputer (!) costing $30,000 in 1974 is vastly outperformed today by a ::- .10 I\ "'0 microcomputer costing $1500. Moreover, the increased avail (.) L~ (!) > ability of computer power in the early 1970s also permitted "iii .01 ~\ a; the introduction of far more flexible machine controls, a: I VL~' known as computer numerical control (CNC.) Moreover, .001 \. \ modular program packages were becoming available which cut programming time for CNC systems by a factor of 3 I from 1971 to 1974 alone. This corresponds to increased use .0001 I I I 1950 1955 1960 1965 1970 1975 1980 ~1985 of CNC in larger-scale production applications (requiring bigger machines) and, especially, a growth in use of machin ing centers. The first generation of adaptive controls, featur FIG. 5. Cost reduction for logical functions (Source: NIRA, 1985) ing force feedback sensors in the workload to detect early signs of tool wear or misalignment, also appeared at that time. The advent of CNC also permitted another develop Nl''.\IEHICi\L CONTHOL OF l\IACHJNE TOOL8 ment: simultaneous control of a .number of NC machines by The first step toward computer integration is the numerical a single computer, known as Direct Numerical Control (analog or digital) control of machines, especially metal (DNC). By the year 2000 comparable cost/ performance cutting and forming machines. The first experiments were reductions can be expected. The plain implication is that the conducted in the period 1948- 1953 under the sponsorship of electronic " hardware" costs are becoming negligible. In the the U.S. Air Force. Numerical controls (NC) were first 1990s and beyond, software wili be the only cost factor offered commercially in 1955. A sequence of tool positions affecting the choice between manual and CNC machine tools and feed rates was specified via a punched paper on or other programmable devices. magnetic tape. The early controllers were expensive and The trend toward user-friendliness in controls has (by modern standards) difficult to program. continued. So-called fourth generation languages of the An early outgrowth of the NC technology was the 1980s exemplified by FOCUS, MARK Y, RA!\11S , IDEAL are development of the so-called machining center (MC) first far more user-friendly than COBOL or FORTRAN, the assem introduced in 1958. These are multiaxis NC milling machines bly languages of the 1960s. At this time, turnkey CAD with the addition of automatic tool-changing capability. systems were successfully introduced to the market givi ng Machining centers are therefore capable of carrying out a rise to euphoric expectations of "intelligent factories" by the sequence of cutting operations on a single part , using up to end of the decade [10] . By 1983 NC and CNC machines accounted for one-third of all new machine-tool purchases in the U.S., and over 103,000 NC and CNC machines were in Table 2. Cost impaets of major microelectronic service. Although this represents only about 5% of all developmc>nts (Sour(~e: NIRA, 1985) machine tools in the U.S., it accounts for a much higher (but not accurately known) percent of output. Bearing in Evolutionary slep Components Component and Cost to assemble assembly costs* ratio mind that many smaller and older machine tools are not I . Discrete-component used for production, and that many production machines are systems (transistors, specialized and automatic, it is likely that NC/ CNC has resistvrs, capacitors, etc.) DISCRETE 20,000- 30,000 $6,000-$9,000 already achieved at least 25% penetration of its maximum potential, given the present emphasis on mass production in 2. Integrated circuits (small-scale the U.S. integration-less than IO gates or bits of memory per HO ROTS device) SS! 350-500 $600-$900 10:1 3. Medium-scale Industrial robots with point-to-point controls for simple integration (adders, material handling tasks were first introduced commercially in counters, etc.-100 gates or bits of 1959 and the first robot with path control capability memory per device) appeared in 1961 (the Unimate). These robots were suitable l\1SI 125-150 $250-$450 20:1 for a number of purposes, including spray painting, spot 4. Large-scale integra welding, arc welding, and investment casting. Demand tion (micro picked up somewhat in the early 1970s. By 1974, when CNC processors and custom LSI circuits capabilities became available, there were about 1100 robots more than 100 gates in service, and unrealistic expectations exploded, only to the or bits of memory disappointed. The number in service probably reached 25,000 per device) LSI 7-10 $100-$200 50:1 sometime in 1986. 5. Single-chip micro- The slow pace of robot introduction in the U.S. prior to computer VLSI $5-$10 1,000:1 1983 is essentially explained by the relative crudeness of the *Excluding backplanes, cables, cabinetry, etc . technology and the high cost of application engineering. The
96 Manufacturing Review vol 1, no 2, June 1988 first practical assembly robots appeared only after l 980, and have not yet been widely accepted. It is much more difficult to find useful tasks for robots in older plants than it is to embed robots in newly designed factories. Even CNC robots are inherently difficult to control precisely because of the stand-alone machine tools relatively large number of degrees of freedom involved (up \ to 7). Most robot manufacturers make it hard to integrate I 32% their robots with other machines under higher-level computer control by retaining secret proprietary operating systems. However, robots of the 1980s are substantially more accurate and better coordinated (e.g., two-hand control) than robots of the 1960s. Programming languages for robots remain diverse and still relatively clumsy. Thus engineering costs for new appli cations tend to be quite high-up to two times the cost of the robot itself, which is a major impediment to small and tools & fixtures 25% first-time users [l I]. Nevertheless, these difficulties are gradually being reduced as experience is accumulated. U.S. based robot manufacturers produced 3060 robots in 1983, \ tunng worth $330 million (they also lost money). Robot capabilities are progressing, primarily because of system improvements in controls and ease of programmability. A control recent breakthrough in gripper design promises to reduce the 8% amount of specialized engineering needed for each applica ~--'------tion. Electric motor drives are replacing pneumatic and FIG. 6. Manufacturing system (FMS) hardware cost (1984) (Source: Data from Kearney & Trecker, Inc.) hydraulic systems for robots requiring greater precision, such as assembly. Operating speeds are increasing, but not dramatically. Robots, in general, work at about the same control (Table 3). Numerical control (NC) capability adds rate as humans. Their economic advantage is greater reli about one-third to the per-spindle cost of a typical machine ability and timelessness. In principle robots can operate 24 tool, and the provisions for integrating CNC into an FMS hours a day, although this capability is seldom fully adds a not her 20% roughly. exploited. However, the major technical breakthrough of the This cost comparison is only meaningful if we compare 1980s is the addition of vision and/ or tactile sensors and equipment manufactured on the same scale of outputs. Rela adaptive (feedback) control to robots. tive costs, too, will change over time. Control-related com ponents of flexible manufacturing systems are rapidly dropping in price, as pointed out earlier. As the price of FLEXIBLE (BATCH) l\IANUFACTUHING: these components decreases, so will the overall cost of the Fl\IS AND LS/FMS FMS. So-called flexible manufacturing systems (FMS) have An obvious implication of the foregoing discussion is attracted much attention since the first attempt to combine that the hardware cost of flexible factory automation can be several NC machine tools with an automated materials cut sharply (perhaps three-fold or more) by deliberately using handling system under computer control (ca. 1967). Applica more standardized equipment modules that could themselves tions have focused on mid-volume batch production of be manufactured in much larger batches. Rapid Japanese moderately complex parts at volumes of 2000 to 50,000 penetration of the U.S. CNC machine tool market since 1980 units/ year. seems to be based on this strategy. These modules will neces In a modern sophisticated FMS, palletized workpieces of sarily be quite generalized in capability, that is, with variable different types randomly travel between and are processed at speeds and cycles and an exogenous system of electronic various programmable, multipurpose machine tools and controls. (Determination of the appropriate control settings other workstations. Parts flow through the system according is done off-line, with the assistance of simulation models.) to individual processing and production requirements, under Here the essential difference between small batch automatic computer control. manufacturing in a multiproduct plant and large-scale or The flexibility of an FMS is only relative (e.g., to a mass production of a single product becomes apparent. In special purpose machine). It is also not achieved without small batch production (job shops) there is really no need to cost. A transfer line and an FMS both need basic machine synchronize the operations of different cells. Coordination drive workheads, materials handling system, and tools. But can be rough, since no run is very long and workpieces in an FMS requires variable speeds and cycles, numerical (i.e., process can normally wait until a suitable machine becomes digital) controls and a supervisory computer to coordinate available for the next operation. Machine utilization can be cell operation (see Fig. 6). In addition to the added hardware cost of an FMS is the cost of the systems software and the Table 3. Cost of machine tool controls ($ x ioa) in FMS specialized programs need to implement a particular task. In Fixed. sequence 100±25 a more sophisticated FMS with automated inspection or Variable sequence 110±25 adaptive control capabilities the cost of sensors and vision NC (Tape) 125 ±.25 (or tactile) information processing must also be included. It CNC ISO± 25 is clear that the implemented cost increases as the level of Adaptive, with sensing 175±25
Ayres: Fu!ure Trends in Factory Automation 97 increased at the expense of work-in-progress inventory, and many respects, the control problems are similar to those vice versa. The optimum balance is determined by encountered in a traffic flow network or continuous process experience, or with the help of scheduling models. But plant, that is, the buildup of nonlinear transients resulting machine utilization is likely to be quite low and inventory of from feedbacks in the system. The analogy between traffic work-in-progress is likely to be high even in a well-managed flow and parts flow and phenomena collisions and conges job shop. Idle machines or exceptional delays are the major tion is quite close. clues to shop schedulers to modify normal processing A recent report by the Economic Commission for sequences. When such problems are persistent, the remedy Europe (ECE) shows extremely rapid growth in the number may be to add an additional stand-alone machine, or possi of first-generation FMS installations since 1975. At the bly to eliminate one that is unnecessary. beginning of 1985 there were 46 FMS's in the U.S. (com In a hard-automated large batch (mass) production pared to four at the beginning of 1975) Clnd around 250 in environment, however, only one product is being made at a the world [12]. As shown in Fig. 7 the rate of growth time and the sequence of operations is fixed. In this situation appears to be accelerating. (As of 1985 the ECE counted the ideal situation is one where the inventory of work in 100 FMS's in Japan, 60 in the USSR, and 36 in West Germany progress is, essentially, one workpiece per workhead. In prin [9].) The technology now appears to be reasonably well ciple, machine utilization is very nearly I 00% when the plant established. A recent forecast by the Yankee Group (cited in is operating except for setup periods and tool changes or reference [13]) puts the likely number of FMS's in the U.S. other scheduled maintenance. Of course, a breakdown at any by 1990 as 280. (Many of these are already planned or on point in the fixed sequence causes the whole line to stop. In order). The U.S. market for FMS is expected to increase an imperfect world this limits the number of machine opera from about $262 million in 1984 to $1.8 billion by 1990. tions that can be linked safely in sequence without a buffer. The first generation FMS systems are largely custom Such a linked set of machines constitutes a "cell" in the designed to produce a "family" of parts in small-to-medium mass production environment. batch sizes. Once built, they are not particularly adaptable to The generic large-scale FMS (LS/ FMS) will therefore other sizes or shapes. However, as adaptive machine control consist of a number of "cells" buffered by intermediate technology becomes increasingly practical in the 1990s and storage, but operating synchronously on the average. The machine control software packages become more powerful target operating mode would be such that the number of and easier to use, more and more new and virtually workpieces stored in each buffer unit fluctuates around half unmanned (second generation) plants will be built to make of its maximum storage capacity. products that are less standardized and subject to more It can be assumed that each machine is controlled by a frequent design change. microprocessor which, in turn, communicates with a mini computer at the cell level. The machine microprocessor con unto tains a stored program of instructions for the machine, 100 ,----~~~~~~~~~~~~~~~~~~~~~ downloaded from the cell controller. Sensory automation monitors performance in real time. Any deviation from the Japao 90 expected status of the machine/ workshop during processing Lh'led States would trigger a slowdown or stop which is signaled to the F.R .G . (High) cell controller. ao The cell controller coordinates materials handling func F . A.G . (Low) tions within the cell and provides the "beat" that synchro U. S . S . R. nizes the individual machine programs (as a conductor 70 synchronizes the musicians in an orchestra). Again, sensory feedback data monitors cell performance in real time, and deviations from the norm can result in a programmed shut 60 down of the cell, and an automatic maintenance call. The cell controller, in turn, communicates directly with neighbor 60 ing cells in a "distributed control" scheme, or with a higher level supervisory computer that coordinates other cells and buffers, as well as overall materials handling functions. If 40 one cell is down the supervisory computer may instruct neighboring cells to continue to function temporarily, taking workpieces from buffer storage or feeding them into buffer 30 storage. In a very sophisticated LS/ FMS there may also be several cells, in parallel, carrying out the same sequence of operations. In this case the supervising computer might 20 bypass one cell and temporarily speed up the others to compensate. This would increase the rate of tool wear and result in earlier tool changes in the affected cells, but this 10 would often be cheaper than simply reducing production for the plant as a whole.
Evidently, the computerized operating system for a 1970 1975 mac 1985 LS/ FMS in large batch production mode would be quite complex, though qualitatively different from the operating FIG. 7. Growth of FMS in the Federal Republic of Germany, Japan, the USSR, and the United States (Source: ECE system for a multiproduct "parts-on-demand" plant. In 1986, Fig. III.I [131)
98 Manufacturing Review vol 1, no 2, June 1988 CAD/CAM CAD systems sold will be on 16-bit personal computers by 1990 [16]. Computer-aided design/ computer-aided manufacturing There is · much less information on the CAM market, (CAD/ CAM) is very nearly self-explanatory, except perhaps since it is extremely diverse and most of the work in this that it is unclear where NC, CNC, or DNC become CAM. field is undoubtedly in-house software development for Roughly speaking, CAM systems are high-level supervisory specific applications. It is likely that the expansion of CAM systems that may carry out planning and scheduling func applications is keeping pace with CAD. However, until CAD tions for a plant and generate programs for individual and CAM are truly linked into one system, the dream of machine tools and/ or cells. Under present conditions CAD "industrial boutiques" producing parts on demand will not and CAM are largely separate, but it is clear that as designs be realizable. (and design changes) are increasingly digitized the blueprint stage will eventually be bypassed. Moreover, the detailed MACHINE VISION AND TACTILE SENSING planning of a manufacturing process (e.g., a sequence of steps), starting from a set of design drawings and specifica Machine vision systems became commercially available in the tions will increasingly be automated. Table 4 illustrates the late 1970s and a large number of new startup ventures progressive complexity of CAD applications with increasing entered the field after 1980. Vision technology is currently emphasis on expert systems. "hot" and the apparent rate of technical progress is very CAD had its beginnings in proprietary systems devel high , as suggested by Fig. 8. The first generation of vision oped in-house by large aerospace manufacturers such as systems required a fairly powerful minicomputer, which McDonnell-Douglas and Boeing. These early systems used specialized software to process visual information (pixels/ s) mainframe computers. However, CAD reached the market and discriminate patterns of shapes by "neighborhood." place around 1970 when the first turnkey systems became These early systems were both crude and very slow. Vision available. The industry grew rapidly, passing the $25 million technology of the mid-1970s was binary. It detected and clas mark in 1977 and the $350 million level in 19?9. At that sified "blobs" based on their shapes, using statistical pattern time virtually all CAD system producers were in the U.S. recognition . A second generation of vision systems capable Worldwide demand continued to grow rapidly, from $592 of discriminating gray scales and more sophisticated syntactic million in 1980 to an estimated $2.8 billion in 1982 and $3.5 pattern recognition began to be available to commercial users billion in 1985 (of which $2.8 billion was supplied by U.S. in the early 1980s. Future systems will eventually add color, firms). At least a $10 billion market is expected by 1995 stereo, shading, texture, motion, shadows, and so on. How [15] . ever, it is not at all clear how soon these capabilities will Unit prices are dropping rapidly. The average CAD appear in affordable commercial systems. Nevertheless, adap system installed in 1980 cost close to $500,000 million when tive systems employing sensory feedback, primarily vision 1500 systems were installed. In 1985, 11,000 were installed at and/ or touch, are going to be the key to truly computer an average cost of just under $400,000. Most of these integrated fifth generation automation, as summarized in systems use 32-bit minicomputers. There were about 18 ,000 Table 5. CAD installations in the U.S. in 1985, and probably 25,000 The key to improve performance of vision systems is worldwide, with an average of four workstations per system. parallel processing and the key to reduce costs is customized It is expected that unit prices of systems sold in 1995 VLSI chips. Such chips began to be produced in quantity by will be about 20% of 1987 prices, with 70% of the per
formance. This is due to the increasing use of CAD adapted ------· ··------, for 16-bit personal computers. It is estimated that 90% of
Factor Improveme nt in Speed Table 4. CAD technology (Source: Chorafa.-;, 1987) {p i xe ls/sec ) Year Capability Cost • c. 1961 CAD, 20, Single Terminal 10.000 tvtainly Drafting <.'. 1963 CAD, 2Y2D, Multiple Terminals
c. 1966 CAD, 30, Full Scale Industrial Applications 1000 Emphasis on Design c. 1968 CAD, Finite Element Analysis VHSIC------+ c. 1970 Simulation Capability Experimentation Capability 100 c. 1972 Integrated Engineering D8 Custom .__ __ _ VLSI - ~ c. 1974 CAD/ CAM, BUI of Materials, Integrated Engineering and Manufacturing D8 Semi-Custom 10 LSI -----+ c. 1978 CAD/CAM Networks, Online Integrated Engineering and Manufacturing 08 with Dynamic Configuration
c. 1982 Integration of Engineering and Manufacturing D8 ~----1- - - _ __L______I _ __~ ~ with MIS 1980 1982 1984 1986 1988 1990 Year c. 1984 30 Geometric Modeling FIG. 8. Estimated improvements in speed (pixels/ sec)/ cost/ratio c. 1986 Integration with DSS for neighborhood pr?cessing (Source: Funk, 1984 (17))
Ayres: Future Trends in Factl'ry Automation 99 1985. Tactile sensors will require parallel processing very color, orientation, reflectivity (shine), and so on. Automated similar to that needed for vision systems. It thus seems quite inspection may become far more sophisticated in a few safe to project that adaptive contro' for both machine tools years, however, as judgment capabilities using artificial intel and robots using vision and/or tactile sensors will become a ligence are built into the vision systems. practical reality by 1990 and will be fairly widespread by At present, most applications of vision (or taction) 2000, as shown in the last column of Table 5. require substantial front-end investments in applications Current applications of vision systems are primarily for engineering. Moreover, they are still quite limited in their the control of manipulation tasks (such as drilling, routing, capabilities, primarily because of difficulties in interpreting a riveting, spot welding, soldering, sorting, palletizing, and visual scene. However, rapid technological improvements in assembly) and for inspection. Examples of both types of the area of sensor sensitivity, software programmability and applications (ca. 1985) are listed in the appendix. In the case user friendliness, together with expected rapid cost reductions, of inspection, the simplest use of machine vision is to check will make automated 100% inspection a practical reality for part dimensions against a stored template. Other types of most kinds of large volume production by the year 2000 (if inspection already exemplified include checking for integrity, not sooner).
Table 5. Five ge1wrations of automation First (I 300): Second (1800) : Third (1950): Fourth (1975): Fifth (1990?) : Fixed mechanical Variable sequence variable sequence variable sequence Adaptive intelligent Premanual s1ored program mechanical program electron1echanical digital (CNC) (AC) A.I. control (clockwork) (punched card/ tape) (analog/digital) (computer control) (systems integration) Source of Human operator Machine designer/ Off-line pro- On-line Off-line Generated by Generated by instructions fo!· builder gram mer I operator operator pro- computer, based computer, based machine (How is records sequences "teaches" grammer on machine level on high-level message sent?) of instructions machine prepares stored program language instruc manually manually instruc- instructions modi tions, modified by tions fied by feedback feedback
Mode of storage NA Built-in (e .g. Serial: patterns as Serial : Serial: In computer In computer (How is message as patterns of coded, holes in as mech . as purely memory as pro memory as stored") cams, gears) cards/ tape or as (analog) electrical gram, with program with pre record impulses branching interpretive/ (e .g., (e .g., on possibilities adaptive on wax magnetic capability vinyl tape) disc)
lnte1 face with Mechanical Mechanical: Mechanical : Electro- Electronic: Electronic: (as in controller (How linkage to power machine Is self- machine is con- mechanical: machine CNC) machine is message source controlled by trolled by mech. controlled by reproduces adjusts to cumula received?) direct mech. linkage actuated valves, switches, motions computed tive changes in link:, to drive by cards via peg- etc. that are by program, state shaft or power in-hole mechanism activated by based on feedback source transducers-in info. turn, controlled by playback of recording
Sensors NA NA NA NA Narrow Spectrum Analog or digital, providing analog digital wide-spectrum, feedback? (con- (e .g., complete descrip verted to optical tions visual, digital) encoders) tactile, requiring (e.g., computer voltm./ processing strain gage)
Communication NA NA NA NA NA Optional Essential, High with higher-level primary because level cot:troller? program micro- controller down processor has loaded at learning from machine ability higher level level must pass visual and tactile info to higher levels to coordinate
100 Manufacturing Review vol l, no 2, June 1988 WHAT NEXT? It is very difficult to estimate the maximum level of penetra tion of robots, FMS, CAD/ CAM, and vision systems. In the case of robots, a simplistic calculation based on the substitu tion of one robot for every two workers in the semiskilled machine operative category (excluding transport operatives) suggests an ultimate potential of 3 to 4 million robots in the U.S. manufacturing sector. This is much too high a number, if the potential for 24 h/ day operation is realized. On the Conventlonal other hand, robots will not replace all operatives, especially Plant & EQ.ipment in smaller firms, for at least 30 to 40 years. Any such massive replacement also presupposes dramatic improvements in robot programmability and performance. In fact, the full potential of robots (and, for that matter, computers) will not be realized until interactive verbal communication in natural language becomes feasible. This has been an objective of research in computer science for many years, but a break through is still very remote. It appears quite safe to assert that this capability will not be a practical reality until well beyond the year 2000. All things considered, the present level of penetration of robots, FMS, CAD/CAM, and vision is probably not more than 1 DJo of the maximum potential, and possibly less. This implies, among other things, that despite a considerable history, nothing much can be inferred about future rates of growth of the sectors involved. The technology is still too primitive and unpredictable for either technology innovators or their customers to make reliable projections as to future price/ performance ratios. Experience from the past does suggest, however, that the difficulties are easily underesti 1950 60 70 80 90 2000 10 mated. In the field of automation, market forecasts have been consistently overoptimistic. Several fairly strong conclusions can be drawn, neverthe FIG. 9. Composition of capital stock (value) less . One is that human labor, especially in the operative category will continue to be eliminated from manufacturing, functions for a world-class manufacturing firm . Security will primarily to increase product quality and reliability while become a far more complex problem in view of the ease of cutting costs. This trend is well under way. It seems quite transferability of software. clear that direct manufacturing labor will decline to an insig A more speculative conclusion concerns the "North-South" nificant level before the second or third decade of the next economic competition. Recent trends indicate a fairly rapid century. This has obvious implications for unions, educa movement of manufacturing away from the high-wage indus tional institutions, and government at all levels. trialized countries, especially to the perimeter of Asia. This A second conclusion that also seems equally robust is has been particularly noteworthy in the area of electronics that the software component of capital will continue to grow assembly and garment manufacturing. It would seem, how in importance vis-a-vis the "hardware" component (Fig. 9) . ever, that as the direct manufacturing component of total The electronic hardware component (computers and elec cost declines, large firms will be increasingly disinclined to tronic controls), which grew rapidly in the 1960s. and 1970s, fragment their operations in this way, with the accompany will not continue to grow so fast, because of declining ing penalties in terms of more complicated logistics, inven prices. In fact, by the year 2000 software is likely to be so tory controls and so on. The logic of the situation would important that it will have to be explicitly measured. While seem to indicate a future trend toward the co-location of no such measures presently exist in the national accounting production with major markets. Flexible automation seems system or the SIC, some indicators are available. It is now a to reduce the benefits of extremely large-scale production widely accepted rule of thumb that the ratio of software to facilities (dictated, in the past, by the costs of "hard" auto hardware costs average around 3: I for any newly computer mation). This, in turn, suggests a more dispersed, decentral ized system. This is roughly the reverse of the rule of thumb ized production system with many more small plants, located in the early 1960s. Issues of software in flexibility software near markets. In effect, production for the U.S. market will compatibility and software productivity are now becoming be increasingly located in the U.S., and similarly in the dominant considerations in designing major systems. An developing countries. International trade in standard increasingly important objective of research will be the manufactured goods will not grow as rapidly in the future as development of intelligent (i.e., adaptive) programs and soft it has in the recent past. ware to generate software. The competitive advantage of low-wage countries may A third and related conclusion is that competitiveness in also be diminished to the extent that by depending more on manufacturing industry will increasingly depend on the qual human labor than the developed countries, they may find ity of a firm 's production software. Software engineering themselves unable to produce goods of the requisite interna (and software security) will become increasingly important tional quality standards. Thus, it seems likely that increas-
Ayres: Future Trends in Factory Automation 101 ingly after the 1990s low-wage countries will have only APPENDIX: EXAMPLES OF APPLICATIONS OF limited access to the markets for manufactured goods in the VISION SYSTEMS IN INDUSTRY wealthier countries, primarily at the low end of the quality User Sensor-controlled manipulation Vendor spectrum. applications Westinghouse Robot-vision system to pick In-house ACKNOWLEDGMENT Winston-Salem, & place and inspect turbine with NC blades. C-MU This paper borrows quite heavily from earlier collaborative GM Consight 1 Vision-Robot System In-house work. In particular, I wish to acknowledge significant Picks randomly placed parts off intellectual contributions by Steve Miller and Jeff Funk, of moving conveyor. who wrote Ph.D. dissertations under my direction on General Motors Light-stripe sensor or Robot RYS Janesville, wrist (Robo-Sensor) for economic impacts of robot machine operation and assembly, Wis. welding of J-cars. respectively. I also want to acknowledge the contribution of Lockheed-GA Robot-based assembly of cargo RVS Susan Bereiter, who did some serious thinking on the impli aircraft using the Robo-Sensor. cations of large-scale flexible manufacturing system (LS Includes: light projector, FMS). She is now completing her Ph.D. under Steve Miller's wrist-mounted camera, computer, software. Hardware direction. l\R cost: $35-$70,000. Lockheed-GA Assembly of internal part for RYS C-130 Hercules Cargo aircraft. REFERENCES Kawasaki Laser-based vision system used t. Cross, R E (1982) . Automation. In: Handbook of indus1rial engineering, for path correction in arc Salvendy and Gavriel (eds.), Wiley, New York, pp. 7.5.1 - 7.5.4. welding of motorcycle parts. 2. Cook, N H (1975). Computer-managed parts manufacture, Scienlific Matushita Robot-vision system for American 232 (2): 22-29. Electric Co., vacuum cleaner. 3. Georghiou, L, Metcalfe, JS, Gibbons, M, Ray, T, and Evans, J (1986). Japan Posl-innovalion performance, Macmillan, London. Texas Calculator assembly lines with 4. Machine Tool Technology (special report No. 726) (1980). American Instruments robots. Machinisl. Lubbock, TX 5. Agren, 8, and Wandel, S (1983). Costs and efficiency of transportation, United Drilling and riveting for inventory, and materials handling activities in Sweden. Unpublished. Technologies, aircraft assembly. Includes: 6. Kearney, A T (1984). Measuring and improving productivity in physical Sikorsky ASEA, I Rb-60 robot mounted distribution. National Council of Physical Distribution Management, Aircraft on track, DEC LSI 11123 as Chicago, IL. system controller, various 7. Quality (June 1977). Qualily Cos/ Survey: 20. contact and vision sensors. 8. McKenney, J L, and McFarlan, F W (Sept.-Oct. 1982). The information Hitachi Robot-vision system which archipelago: Maps and bridges. Harvard Business Review. detects holes for assembly. 9. Comprehensive sludy of microlec/ronics (1985). National Institute for Includes: solid state optical Research Advancement, Tokyo. sensors, CCD-type TV camera to. Quantum Science Corp. (1974). The inlelligenl faclory: Cos/ jus1ifica1ion mounted on robot arm. breaklhrough. Quantum Industry Report, New York. Western Color-sorting of telephone 11 . Miller, S (1983). Potential impacts of robotics on manufacturing costs Electric receiver caps into bins. within the metalworking industries, Ph.D. dissertation, Carnegie-Mellon Atlantic (6500/h). Uses photo diodes University, Pittsburgh, PA. Plant and color filters. 99.90Jo accuracy 12. Sheinin R L, and Tchijov, I A (1987). Flexible manufacturing systems 1 GM Stacks random mix of pre- (FMS): Stale of art and development (Working Paper WP-87-17) llASA, Warren, MI taught parts. Uses light stripe, Laxenburg, Austria. PUMA robot system, 3 DEC 13. ECE (1986). Recenl /rends in flexible manufac/Uring. Economic LSI 11 's, video camera, and Commission of Europe, United Nations, Geneva. VAL programming language. 14. Chorafas, D (1987). Engineering produc1ivi1y through CAD/CAM. But Inspection Applications terworths, London. 15. Office of Technology Assessment (1984). Compulerized manufac/Uring: Unknown Automatic inspection of welded MIC Employment, education, and lhe workplace (OTA-235). U.S. Congress, automobile wheel hubs. Checks Washington, D.C. for integrity of structure. 16. Ebel, K-H, and Ulrich, E (1987). Social and labor effec/s of CAD/ CAM. Unknown Off-line floppy-disk jacket MIC ILO, Geneva. inspection, manually operated. 17 . Funk, J (1984). The potential socie1al benefits from developing flexi ble Checks dimensions. assembly technologies, Ph.D. dissertation, Carnegie-Mellon Universily, Unknown Automatic identification of MIC Pittsburgh, PA. various models of electrical 18 . GAO (1976). Manufacruring rechnology: A changing challenge 10 circuit breakers on a conveyor improve productivi1y. General Accounting Office, Washington, D.C. belt. Checks product type. Unknown Automatic inspection of MIC ceramic supports for cathode ray tubes. Checks for dimensions. Unknown Automatic inspection of ray MIC tube displays. Checks for integrity of features.
102 Manufacturing Review vol 1, no 2, June 1988 w ·tMI rr __.
APPENDIX continued Mr. Ayres is Professor of Engineering and Public Policy at User Sensor-controlled manipulation Vendor Carnegie-Mellon University and the Deputy Program Leader applications of the Technology, Economy and Society Program at the Unknown Automatic inspection of spark MIC International Institute for Applied Systems Analysis (IIASA) plugs on a moving conveyor in Laxemburg, Austria. He is also the Project Leader belt. Checks dimensions. (currently on leave) for the Computer Integrated Manufac Unknown Automatic fluoroscopic MIC turing Program at IIASA. Mr. Ayres is the author of eight inspection of cut and welded parts for stress cracks. Checks books on technology and economics, of which the latest is integrity of internal structure The Next Industrial Revolution, published in 1984. and dye is used to make flaws fluoresce. Unknown Automated inspection of glass MIC CRT necks, uses a UV-light source to image internal defects. Checks intergrity of internal structure. Unknown Automatic inspection of plastic MIC sutures. Checks integrity and dimensions. Unknown Automatic inspection of MIC automotive wheel hubs for conformance to forged dimensions prior to subsequent machining operations. Checks integrity. Unknown Inspection of valve bodies for MIC automatic transmission. Vision is interfaced with robot. Soft ware mask examines internal details. Exact positioning is required. Checks dimensions of a single type of product. Unknown Automatic inspection systems MIC for precision components. Vision is interfaced with a robot. Checks dimensions. Unknown Gray-scale imaging system for Octek, Inc. paper-cup packaging. Checks for number of cup lips. Cummins Inspection of engine blocks. RVS Uses light striping. Hitachi Automatic Reticle System (ARI) Japan which uses semiconductor photomask inspection for products. Delco Determines chip position and In-house Electronics orientation, inspects chip Kokomo, IN structurally, allows for proper alignment of test probes with chip contacts. Honeywell Robot vision station for solder In-house joint inspection of circuit boards. Uses TV camera for 2-D image, PUMA 560 robot, Autovision II, plus micro-computers. Combined sensor-controlled manipulation and inspection application Automatix Robot-vision system for Automatix Corp. assembly and inspection of Billerice, MA keyboard arrays. Uses the Cybervision Assembly Station and the Autovision JI processor, with the AID 600 robot and AI 32 controller.
Key lO Vendor Abbreviations CMU: Carnegie-Mellon University, GM: General Motors, MIC: Machine Intelligence Corp., RYS : Robot Vision System , ?: Vendor of system not specified in literature
Ayres: Future Trends in Factory Automation 103
l\'IANUF ACTURING PERSPECTIVE Technology Forecast For CIM
ROBERT U. AYRES International Institute for Applied Systems Analysis, Laxenburg, Austria and Department of Engineering and Public Policy, Carnegie-Mellon University, Pittsburgh PA. 15213
Tllis paper identifies a n11111ber of aggregated ('"top dow11'') meCis11res of trcl111ologieal eaJJ INTRODUCTION MANUFACTURING FROM AN INFORMATION PERSPECTIVE here are two main approaches from which any fore It has been argued previously that manufacturing can fruitfully casting problem may be viewed. These may be be viewed as a process of embodying useful information in termed top down (or macro) and bottom up (or mi materials [I]. To be precise, the "value added" to a material Tcro). The top-down approach begins with a broad or workpiece corresponds to the addition of some kind of in characterization of the technology in terms of its function, formation. The term "information" is used, here, in its tech and proceeds to attempt to identify a suitable quantitative nical (Shannonian) sense, meaning, roughly, distinguishability measure of performance or, failing that, a set of discrete from the environment or surroundings. Where the term en stages or milestones, together with estimates of when they tropy can be defined, information can be defined as its nega will be achieved, assuming current rates of investment in tive (negentropy). An exposition of this way of looking at R&D continue. The bottom up approach, by contrast, seeks economics has been given elsewhere and need not be dis to decompose the overall technology into its systems, subsys cussed at length here [2]. tems, components, and so forth. It then proceeds to identify In fact , it is helpful to categorize embodied information suitable quantitative measures of performance for each of into two types: (I) thermodynamic information and (2) mor these elements, and finally to re-?.ggregate them. In this paper phological information. The first is a function of physical and we will address the forecasting problem from the first of these chemical composition and micro-structure. It can be com directions. puted quantitatively by methods described in standard text An inevitable source of some confusion must be ac books, but which need not be discussed further here. Suffice knowledged. In principle we would like to draw a clear dis it to s?..y that every stage of materials processing, from extrac tinction between changing technological capabilities and adop tion and crude separation through smelting, refining, alloying tion/diffusion/penetration of existing technologies into new or chemical synthesis can be described thermodynamically as areas of application. In the case of CIM these two processes the creation of entropy in the environment and negentropy (or proceed simultaneously and to some extent indistinguishably. information) in the materials being separated, refined, etc. The technology of computer integration cannot really be said The second type of information is essentially a function to exist for a given manufacturing sector and product until it of geometrical shapes and forms. It depends, of course, on has been demonstrated in practice at least once. The problems the precision with which shapes or forms are specified. Since in different segments are sufficiently different so that it is the more conventional usage of the term information-that sometimes very difficult to extrapolate from one case to an which is conveyed by words or pictures-refers to symbols other. Because of this difficulty, however, some measures and patterns, it is evident that the kind of information found which appear to be measures of diffusion are, at least indi in a library or a computer program is entirely of the morpho rectly, also surrogate measures of capability (and vice versa). logical kind. Morphological information also presumably cor- © Copyright 1989 American Society of Mechanical Engineers 43 responds to negentropy, but there is no advantage in adopting prices of useful information and garbage information (errors), thermodynamic language, except to the extent that it facili respectively embodied in products. The utility U refers to the tates conceptual clarity. For our purposes, the morphological given unit of time. The expression (1) above can obviously be information embodied in a material as a result of some metal expanded into summations over specific types of manufac working process can be thought of as the minimum amount of tured products. information needed to describe the process or, in effect, to in Of course, the unit price of garbage information is nega struct a numerically controlled machine tool to carry it out. It tive. That is to say, misinformation (errors and defects) cre must be emphasized that the minimum, which is the amount ates a cost, and-in many circumstances-the cost of detect embodied, may be considerably less than the actual amount of ing and correcting one error is significantly larger in information in a practical computer program. magnitude than the unit price of "virgin" useful information. The output of various industrial sectors can also be clas This reflects the well-known fact that a single defective part sified in terms of the kind of information embodied. Evi embodied in a larger system may suffice to cause that system dently the metal-working sectors of interest to us (SIC 33-37) to fail, or necessitate major additional expenditures for rework are exclusively concerned with shaping, forming and assem and repair. In any case, the fact that some errors and defects bly, i.e., with morphological information. In traditional man are inevitable introduces the need for pervasive and continu ufacturing the actual shape-changing operations are done by ous error detection and correction activities. Inspection and machines, but all of the control decisions are made by human repair/rework becomes a major element of cost in complex workers. The rate at which human workers can process sen production systems. sory inputs and generate information outputs (e.g. instructions The problem is that failures tend to cascade. According for machines) is biologically limited [ l]. Moreover, ergono to a nursery tale: "For want of a nail, the shoe was lost; for mists have observed that the human error-rate, or the fraction want of a shoe, the horse was lost; for want of a horse, the of information outputs that is garbage, tends to rise sharply as rider was lost; for want of a rider the message was lost; and the worker approaches maximum output, as indicated sche for want of the message, the war was lost. " The modern ver matically in Fig. 1. Since the economic value of output infor sion (unfortunately) might be: "For want of an 0-ring the mation-expressed as patterns or shapes-is likely to decline Challenger and its crew of astronauts was lost; for want of the as a high power of the error or garbage fraction, there is a Challenger the U.S. manned space program has lost billions high incentive for increasing accuracy (decreasing defects) . of dollars and several years. " It may, indeed, be permanently In view of the above points, the function of manufactur crippled. ing, taken as a whole, can arguably be described in terms of The severity of the error-defect problem is essentially maximizing the useful (i.e. correct) morphological informa proportional to the complexity of the final product. If one is tion embodied in materials by the manufacturing process, manufacturing nails, the cost to the user of a defective nail is while minimizing the incorrect information, or defects. Since essentially the cost of the time to recognize it and discard it. two objectives cannot simultaneously be optimized, however, The cost of a defective watchspring is comparable, as long as we must devise a single composite objective function. Such a the defective spring is recognized before it is put into a function might be of the form watch. If the defect causes the watch to fail later, the much greater cost of dismantling the watch and putting it back to gether must be added. In larger systems, the cost multiplica U = PuH + PgG (1) tion continues. A defective chip in a P.C. board will necessi tate only some manual rework if it is detected before the where H is the quantity of useful (i.e. correct) morphological board is installed in a navigational computer. If the computer information embodied in the products of the metal-working fails in flight, however, the result could be catastrophic. It sectors during a unit of time, G is the quantity of incorrect follows , therefore, that P 8 is some function of the complexity (garbage) information embodied, and Pu and Pg are the unit of what is being produced. Evidently, complexity is an im portant attribute of technology. See Fig. 2. Further discussion can be found in [3] . As yet, we have no good measure of complexity for use Information in estimating the cost of defects and garbage information. Output Rote Complexity, itself, is a measure of structural (morphological) bits/sec information embodied in the completed product. For a simple product like nails, it is essentially the same as H in the above formula. For multi-component products, of course, the com plexity of the product as a whole corresponds roughly to the inforn1ation embodied in all the components individually, plus Maximum Useful Output the structural information needed to assemble them. ... __ __ Can complexity be measured in practice? A complete and consistent set of rules for adding up the information contents Useful Output of the parts to make the whole involves a number of special considerations, such as symmetries [!]. As a first approxima tion, however, one might measure complexity of products in Information terms of the number of discrete parts, times some measure of "'------C> Input Rote bits/sec the complexity of the average part. In the case of a purely mechanical system, the complexity of a rotational or prismatic part can be estimated in terms of the number of cylindrical or Fig. 1. Information Overload Phenomena flat surfaces and the precision with which each of them is 44 Manufacturing Review vol 2, no 1, March 1989 (e .g. welding) machines. By far, the largest number of ma Number of Ports chines in use belongs in the first category and the majority of (Powers of 10) 8~------1 them are turning machines, drilling machines, milling ma Space Shuttle chines, grinding machines and sawing machines. Among Moss Produced Products Smell Botch Production • metal-forming types, by far, the largest is presses, followed Un ique Products 74 7 Harvard Mork I •.. • B by bending machines and punching machines. Turning ma •.•.... ·· chines generate cylindrical sections (external or internal) or Ca lculator (25,000 ports) .•. OC3 profiled surfaces with rotational symmetry, such as screws . • Milling machines generate flat surfaces, as well as almost any ··· ······························ • Au tomobile type of curved surface. Grinders are also used for many sur face geometries. Gears-a specialized subset-are made by a Bicycle variety of specialized machines that could technically be Springfield Rifle classed as millers, planers, shapers, hobbing machines and (1 40 ports) 'M u sket ( 5 1 ports. generating machines. Saws (along with flame cutting torches) · 10,000 units) 1 are mainly for crude cuts subject to subsequent finishing . It is argued here that each machine operation embodies a 1800 1820 1840 1860 1880 1900 1920 1940 1960 1980 2000 certain amount of morphological information in the resulting Yeor surface, depending on its geometry [ l]. An average informa tion output for each operation, for each type of machine, Fig. 2. Complexity Trends should be possible to determine by a combination of empirical research and theory. Thus, the information required to specify a simple flat surface may be computed from the information specified in the design [I]. In the case of higher-order curved required to specify a plane intersecting a solid. Depending on surfaces, such as turbine blades or propellers, a more sophis the required accuracy, this can be determined in bits [ l]. ticated mathematical analysis is needed. For microelectronic Roughly, the amount of information per operation (i.e. parts (chips) similar considerations appear to be applicable to per surface generated) is 10 bits per dimensional control pa the individual circuit elements themselves. Further research is rameter or coordinate that must be specified plus a contribu obviously needed to develop such measures into useful analyt tion from the information embodied in the machine's design ical tools. that enables it to "know" that a given parameter refers to a particular geometrical family of shapes (e.g. bevel gears). The MORPHOLOGICAL INFORMATION EMBODIED amount of information implicit in this knowledge is, at the BY CUTTING AND FORMING OPERATIO NS moment, difficult to estimate. Several approaches, however, In principle, it would seem, all the variables in the simple are currently being explored. utility function for manufacturing (equation 1) can be esti Milling operations might require as few as 4 parameters mated from technical data, except for the two price coeffi to specify (for a flat surface), but a more general case would cients. It is illuminating to focus specifically on the shaping be to specify 13 parameters, consisting of 3 initial position and forming component of manufacturing. 1 Let us note the parameters and 3 feed rates for the cutting tool and the same time dependence explicitly and write again for the table or pallet, plus spindle speed. As in the case of turning machines, the feed rates can be functions of time for cutting very complex asymmetrical, curved surfaces H(t) = A(t)M(t) (2) such as a turbine blade. This, with the grinding machine, is the most flexible type of machine tool. It is capable of pro where A(t) is the amount of morphological information em ducing virtually any surface, interior or exterior. bodied in finished materials per unit mass (e.g. ton) by the A typical lathe operation might require 7 parameters, cor metal-working sectors and M(t) is the tonnage of materials responding to initial tool position (X-Y-Z) and feed rates plus processed, in year t. Obviously A(t) constitutes a rather gen spindle speed, which again determines surface roughness/ eral measure of technological capability. Again, the expres smoothness. The tool contour can also be selected (e.g. for a sion (2) can be expanded into summations over types of mate screw-cutting machine). For more complex contours the feed rials and types of products. However, so far, we have no rates can be functions of time. Turning machines (lathes) can plausible independent measures of information embodied in produce any shape with a cylindrical symmetry. materials. To obtain such a measure, it is necessary to expand A typical drilling operation would require 5 parameters: the magnification of our field of vision by examining the an X-Y coordinate, a depth and radius of the hole, and a sur manufacturing process in more detail. face roughness/smoothness parameter related to spindle speed Within the metal-working sectors, it is reasonable to sub and feed rate. The hole radius is, of course, the drill radius divide all processes except assembly and inspection into unit which is an attribute of the tool. Boring, broaching and hon operations, each of which is performed by a machine. The ing machines produce successively more precise and smooth major categories are metal-cutting, metal-forming and joining interior cylindrical surfaces, but 5 parameters would suffice for each. Tapping and threading machines are specialized to generating interior threaded surfaces. The thread contour is 1A reviewer has noted that this approach could also be applied to the in formation content of input materials themselves, which constitute an increas determined by the choice of die, in the case of a tap, and by ingly important component of total costs. Composition and micro-structure can the tool contour in the case of the threading machine. Saws, also be considered to be forms of embodied information. This is one of the shears, planing, lapping and polishing machines all produce "hotter" areas of research and innovation today. For a more detailed discus sion of some methods of estimating compositional and micro-structural infor successively more precise and smoother flat surfaces (4 pa mation, see also [l]. rameters). Shaping machines produce grooves in flat surfaces, Ayres: Technology Forecast For CIM 45 with the contour of the groove determined by the tool con tour. Gearcutting machines, by definition, produce gear shapes (7 parameters, not including the tooth contour). Again, Ball bearings the tooth contour is typically determined by the cutting tool. Assuming the number of control parameters indicated CNC above, the information embodied in products by each machine Piston rings 46 Manufacturing Review vol 2, no 1, March 1989 associated with metal-working. Even among metals there are great differences between the problems of cutting, shaping rt ~ Pcec;s;on oeoc;ngs and forming soft or malleable metals such as copper vis a vis ~ 1 ···n;es. hard metals such as tool steel or superalloys. For the present, ·r Ga.s t urbine ports Di ese l engine Gasoline engine if not necessarily for the indefinite future, the major structural block, crank block. crank Secuiit.Y keys metal is carbon steel, with gray (cast) iron, stainless steel, MC,CNC FMS aluminum alloys and other alloys following in rough order of Hydraulic\ Gear wheels turbine pa'r.~s (machined) Machining, importance. I shall not , in general, discuss processes used Investment costing only for special materials (such as the investment casting pro cess used for superalloys). Size is an important factor in its own right, at least on the micro end of the spectrum, because very small parts can . not readily be handled hy human fingers or even discerned by .3 Housings human eyes. Thus increasingly specialized techniques have had to be developed over the decades to manufacture watches Lsm_ o_ll _____;_ __ .:-- _nn_e_,d;-u m----~,o~, 0~e ----- Botch size (for instance) efficiently. It is probably no accident that a large-scale Japanese watchmaker (Seiko) pioneered in robotic assembly. At the other end of the spectrum, large or heavy Fig. 5. Very Complex Shapes workpieces are a special problem because they cannot be eas ily manipulated by unaided human arms and hands. There is a tradeoff, which applies to machines as well as humans, be tween dexterity and strength [4]. Very large pieces must be """"'°"' moved and assembled with the aid of massive and corre spondingly crude cranes or hoists. For this reason assembly merges with construction. The other choice factor, as noted, is scale of production. Unfortunately, scale has several possible meanings. In the case of mass production, the relevant measure is just the length of the production run , or the number of pieces over which the capital equipment must be amortized. However, as soon as some capital sharing becomes possible (economies of scope) there are several, perhaps many, joint products. One of the relevant variables is batch size, because each batch is preceded by a setup which takes anywhere from a few min utes to a couple of days, depending on the product and the equipment design. Another relevant variable is the number of different members of the product family that can be made on S ouTce R. a.y BatchSil:e the equipment. Still another is the frequency of change. All of these are obviously related. In practice, we must consider Fig. 6. Complexity-Batch Size situations ranging from very large batches (long runs) of a few variants of the same basic design, to customized produc tion of many different parts made once and only once. The cost-reduction potential of computerized control tech balanced load "walk" across the floor can imagine the prob nologies varies with the type of manufacturing. There is no lem). Thus, very high speed rotating machinery, such as gas impact in situations where there is no variation either of the turbines and centrifuges, constitute the most challenging of all product or the production level. This might apply, for in manufacturing problems. stance, to fully automatic production of completely standard 4 Closely related to precision is the control of variance. ized products such as nuts and bolts, sparkplugs, or light Manufacturers in the 1990s can be expected to demand signif bulbs. On the other end of the spectrum, there is also little or icantly tighter control over the width of the band of allowed no impact in situations where the product is always different variance in order to increase overall output quality and reduce and made to measure, as in some tool and die or industrial re-work. This is one of the primary strategies to reduce the pattern shops or machine shops building experimental design cost of garbage information, noted earlier. prototypes. This domain may be accessible in the future, The other major factors involved in the choice of manu however, to truly intelligent machines. (The nature and degree facturing technology are materials, physical size, and scale of of intelligence needed will be discussed later). production. Needless to say, the problems of manufacturing However, most production lies between these extremes. things from glass, ceramics, stone, plastics, wood, paper or Starting from the low volume job shop level, a reasonable textiles are quite different in many respects from the problems amount of design work might well justify at least a small mi cro-computer based CAD system to simplify design changes and record-keeping. With a still larger volume of business, 4 Lack of control over variance. rather than precision per se was the bar especially in the area of high precision and complexity, a rier to interchangeability of parts. which was first recognized and addressed stand-alone NC milling machine or lathe (or two) might make historically in the manufacture of small arms. For a good discussion of how this problem influenced the evolution of manufacturing, see, for example, sense to reduce machining time and increase accuracy. For Hounshell [Hounshell 84] . higher volumes still, machining-centers with CNC controls Ayres: Technology Forecast For CIM 47 and automatic tool-changing capability begin to make eco nomic sense for complex parts. # of Ports/Year The next step, as one moves up the volume scale, is to 10 7 divide the sequence of operations among several more special ized machines. At first this might involve control of the indi 106 \ vidual machines from a single mini-computer which down 5 loads the instructions for each part to each machine in the 10 l'loiliO sequence required. Loading and unloading and transfer can 15,000~ Transfer 4 still be manual. Of course various degrees of automation can 10 Lile 2000 FMS be applied to the material-handling problem, from simple 1000 roller or belt conveyors to automatic transfer machines, auto 500 FMC matic guided vehicles (AGV's) or robots with special-purpose 100 grippers. A technology that is not yet available, but which is 2S CNO gradually being developed, is the general purpose gripper [4]. 10 When a number of machines is linked together and run by a • O 80 0 common computer, the system is called a flexible manufactur l O 100 1000 ing system or FMS. Numb er o t Differ e nt Ports As it happens, the FMS solution can also be reached, as it were, by evolution from the mass production end of the spectrum, by adding some flexibility to a conventional hard Fig. 7. Volume/Variety Tradeoff automation system. In point of fact, the realm of batch pro duction is beginning to subdivide into distinct sub-regions, de others of comparable scope and cost, are going to be success pending on the volume-variety tradeoff. This tradeoff is illus ful enough to stimulate many imitators, or whether they will trated in Fig. 7. In effect, two kinds of FMS systems appear be judged a few years hence as over-ambitious failures. to exist, namely small systems , which are designed for small batch production with considerable variability and cost less WHEN WILL MACHINES BE "INTELLIGENT"? than $5 million, and large systems which are designed for Before proceeding to forecasts of CIM performance, it is large batch production and much less product variation, and worthwhile to review the current status of CIM technology which cost closer to $20 million [5]. The first category uti from the perspective of limits, i.e. , what it cannot yet accom lizes a minicomputer for scheduling and downloading operat plish. Given the widespread use of terms like intelligent ro ing programs to individual machines, but most setups, loading bots and expert systems, not to mention the rather optimistic and unloading, and inspection is manual. The second category specifications of some of the advanced systems recently built of system does not necessarily have more machine-tools than or currently planned, it is not unreasonable to wonder why the first, but the machines are designed for long life and high CIM technology is not being adopted faster than it is? The production rates with minimum human supervision. Loading plain fact is that in the realm of eye-hand coordination and and unloading, parts transfer, storage and inspection are likely decision-making, machines are not yet able to substitute for to be automated. The software systems required to control the human skills beyond the most elementary ones. According to associated materials handling and inspection functions tend to two researchers at the leading edge in the application of artifi be correspondingly more elaborate and expensive. cial intelligence to manufacturing technology: Another type of large system, also quite expensive, is de signed not for large volumes but for very large part families In the global view of the world . . . represented in glossy (in the thousands). In such cases, the software is even more trade magazines, the machine level of the factory hierarchy elaborate and may become a substantial fraction (half, or is normally represented as a black box with its own con more) of the total investment costs. Several of these systems troller. From a senior management point of view, there have been installed by aircraft manufacturers, such as General might be the temptation to believe that the machine tool is Dynamics, Vought, British Aerospace (U.K.), and M-B-B already an unattended autonomous unit. However, more (West Germany). Machine tool companies have also built a day-to-day experiences with machine tools quickly show number of such systems for their own use. Examples include that this is not the case . . . [4]. Fanuc and Yamazaki in Japan, and Cincinnati Milacron and The authors go on to point out that, while there are some Ingersoll Milling Machine Co. in the U.S . examples of NC machine tools being run unattended for batch The $20 million Ingersoll system, for instance, has been production (such as the Fanuc plant at Fuji, Japan), this is under development for more than a decade (6) . It was origi currently possible only in very restricted conditions. For ex nally planned to manufacture 25 ,000 different parts annually, ample, part accuracies cannot be held to less than 0.002 of which 70% were in lots of one, and 50% would be unique inches, programs must be debugged and run many times un and never repeated (it is doubtful that this ambitious goal has der manned conditions before being run in unmanned condi yet been fully achieved) . It consists of 3 cells which replace tions, part geometries involving critical cutting conditions are 40 stand-alone machine tools. The system links order entry, avoided; feeds and speeds are set very conservatively to make pwduction scheduling, bill-of-materials, geometric modelling cutting as predictable as possible, fixtures, tooling and maga and billing. This involves an integrated, company-wide com zining are all done manually before hand, etc. Putting it an puterized management and business information system (MIS/ other way , "the production of a good part that is right the BIS). This task alone involved replacing 1300 applications first time in a rapid prototyping environment is extremely dif programs and 225 separate files , and took 2 years to com ficult. The full automation of such a process in which only a plete. It is too early to judge whether the Ingersoll system, or handful of parts are to be made is currently impossible" [4] . 48 Manufacturing Review vol 2, no 1, March 1989 When newly designed parts first go into production, there sequences of operations, setting feed rates and cutting speeds is always a learning curve. As a general rule, the first few and avoiding trouble. Trouble can arise from a variety of parts are carefully compared with the desired standard. Devia sources, such as vibration , excessive tool wear, heating, jam tions require adjustments to be made to the programs, the ming, accumulation of chips and shavings, etc. Problems can tools and the fixtures. In a typical case, there is still a 20% not always be anticipated, especially in cases where it is nec probability of an out of tolerance defect in parts numbered 6- essary to operate near the margins of capability (as in dealing 15, and a 10% probability in parts numbered 16-25 [Kegg, with very hard metals, or metals extremely subject to work cited in 4]. Most of the problems that arise here are resolved hardening. The machinist, relying on years of experience, re on the spot by a skilled machinist. The principal source of de lies on a variety of signals, ranging from the color and length fects (as might be expected, in view of what was said earlier) of the chips to the sound of the cutter and even the smell of is human errors in programming or setting up the NC tools. the hot oil. He interprets the data and modifies the settings The errors only show up when the tools are actually operated, on-line if possible; or he stops the machine and changes the and then the problems must be diagnosed and fixed based by fixtures or the program. He occasionally diagnoses incorrectly the expert machinist. and makes the wrong decision, which must be corrected later. What are the types of knowledge involved, and how long Incorrect diagnosis is the inevitable consequence of uncer does it take to acquire? As to the first question, Wright & tainty. The human machinist learns from his mistakes and be Bourne [4] list the subsystems that will surround the central comes more skilled as time goes on, but he can never hope to processing unit (CPU), or cortex, of the future intelligent ma avoid all errors. chine-tool controller (the black box in the standard hierarchi In this context, it is relevant to note that the higher levels cal control diagram), along with the estimated CPU usage of of human skill in this field (like others) requires many year:s each subsystem: of study and effort to achieve. Given that computers are still unable to beat the best human chess players- despite an in CAD description (5%) tensive effort spanning decades- it is really not plausible to Operator interface (7 .5%) imagine that intelligent machine tools will supplant skilled Plan formation (12.5%) machinists in a shorter time. Progress in computer hardware Machine tool control (12.5%) can be assumed, but hardware is not the problem. The diffi Robot control ( 10%) culty in the case of chess is that even the best and most ana Fixture lab (5%) lytic chess players are unable to describe how they evaluate Gripper lab (5%) positions and choose among possible moves in a way that can Tool supply (2.5%) be reproduced by a computer. Stock supply (2.5%) Yet the rules of the game of chess are relatively simple Chip removal (2.5%) and explicit and there are no uncertainties to contend with, Sensor lab (5%) except the intentions of the other player. In the case of ma Unified interpretation (7.5%) chining (or other manufacturing processes) the rules and con Inspection (I 0%) straints are far more complex. Moreover, they vary from case Vision monitoring (12.5%) to case, depending on the material , geometry and precision Even a top-of-the-line machine controller of today lacks most desired. Finally, there are significant inherent uncertainties to of these functions. Since programmable grippers and fixtures contend with. (For example, tools from the same batch may (hands) do not yet exist on commercial machines no cortex differ in useful life by a factor of 4). Moreover, skilled ma functions are yet allocated to them. Similarly, vision systems chinists are not likely to be good at articulating and analyzing have not yet been integrated with machine tools. Finally, ex their own thought processes. Thus, although a start has now isting systems have no ability to plan or make decisions, so been made at codifying this kind of knowledge in expert sys these cortex functions are also lacking at present. Moreover, tems , the level of expertise that can be embodied in such a the operator interface is currently very clumsy and inefficient, system is not likely to be really effective for many years to since the machine cannot hear, interpret , or reply to spoken come. commands. In fact, the most rapid progress in automating small When, if ever, will machine tools become intelligent batch production is likely to come from a completely different enough to dispense with a skilled machinist in the debugging source, namely computer-aided-design. Until very recently, mode described above? Sensory feedback-primarily vision designers typically had one objective, namely to maximize the and taction (from the flexible grippers and fixtures)-is a pre functionality of the product. Manufacturability was a con requisite [7] . But the ability to correctly interpret visual and straint, of course, but design changes to make a part easier to tactile information in terms of internal models of the machine produce were usually afterthoughts resulting from wrestling and its appendages, the workpiece, and the cutting tools is the with actual difficulties in the shop. In recent years, efforts essential missing ingredient. have been focussed in several laboratories on codifying princi Wright & Bourne see significant progress over the next ples of design for manufacturability , especially assembly [8, 20 years, with unattended machines capable of error recovery 9]. Dramatic savings in production costs have been demon diagnostics by 2010 [4 p. 290]. In view of a history of over strated in a number of cases by simplifying and rationalizing optimistic forecasts in the field of artificial intelligence (Al), the design. It is likely that designers in the future aided by in it would not be surprising if the suggested 20 year time-frame telligent CAD systems, will be increasingly able to avoid call also turns out to be too short. ing for combinations of part geometries, tolerances and mate It is not appropriate here to summarize the content of the rials that create difficulties and uncertainties for machinists. machinists skills or other manufacturing skills that need to be This is, in fact, by far the most promising route to the goal of captured. For the most part they are concerned with choosing a good part the first time. Ayres: Technology Forecast For CIM 49 MEASURES OF TECHNOLOGY to centralized, company-wide databases, the software fraction of total investment may well reach the 80% level already Intelligence apart, a list of the important attributes potentially characteristic of large military systems. The projected growing distinguishing CIM from conventional manufacturing tech fraction of software investments vis a vis hardware invest nologies is a good starting point. The following have been ments over time is shown in Fig. 9. suggested: Several of the other attributes listed above also reflect as 1. Ratio of unmanned/ manned time for machines or cells pects of flexibility, of course. For instance, a high degree of 2. Maximum continuous unmanned time in practice flexibility corresponds to the smallest possible order-to-deliv 3. Fraction of unproductive machine time ery time, the smallest possible inventory of work-in-progress, 4. Workpiece cycle (throughput) time and the maximum ratio of product variety to batch size. 5. Work-in-progress/output ratio These data may be available for scattered cases, perhaps with 6. Order-to-delivery time some before and after comparisons, but industry-wide data are 7. Reject fraction not available. The same holds for reject rates (a quality indi 8. Product variety (family size) cator) and tolerances. In fact , it is not clear that CIM will 9. Average batch size have any impact either way on the latter. 10. Direct/Indirect labor ratio Another kind of technology measure is suggested by the 11. Software fraction in invested capital foregoing discussion of manufacturing in terms of machines Bearing in mind the issues noted in the previous section embodying information in materials, one discrete operation at above, one of the best surrogate measures of the level of CIM a time. The time required for each operation is a function of technology is probably the fraction of software cost to total the rate of cutting or forming. This has been increasing rap manufacturing system cost, or the fraction of software in cap idly for a century or more (Fig. 10), but it is not clear that ital investment. It can be argued that this is a rather direct past rates of improvement can be sustained. The reason is that measure of flexibility , since software-driven systems are pre improvements have mostly resulted from the development of sumably reprogrammable, in contrast at least to hard automa harder cutting materials-from tungsten steel to nitride and tion. A schematic tradeoff curve showing the increase of soft boride ceramic coatings. The latter approach the hardness of ware costs relative to span of control is given in Fig . 8. diamond, which is the hardest known material. Diamond Some confirmation of the figures in Fig. 8 can be found coated cutting tools from Asahi Diamond Industrial Co (Ja in the data we have gathered so far. Stand-alone Numerically pan) reached the market-place early in 1988 [N.Y. Times 10/ Controlled Machine Toll (NCMT) costs, today, are roughly 25 /88] . At least one expert has remarked that cutting technol I /3 attributable to operating software [ 1OJ . When a number ogy has already reached 90% of its theoretical maximum of NCMT's are linked together in an FMS , the NCMT's ac potential [13]. count for 50-55% of total costs, (of which software accounts On the other hand, the utilization of the available time of for a third, as noted) and systems control , communications cutting tools is still quite low, on the average. Data for the and interfacing software adds another 20-30% or 37% to U.S. were compiled by the Machine Tool Task Force [14). A 48% in all [I 0-12). These figures seem to be fairly stable for summary of the time budget for U.S. machine tools in small both small and large FMS's. Software is said to account for batch, large batch and mass production is shown in Fig. 11. 50% of the cost of one recent showcase Japanese automated One of the major savings offered by NC is to increase the uti factory , and may reach 60% for the massive effort currently lization rate of expensive capital equipment. This can be re under way at General Motors. (This was the reason GM spent garded as an indicator of system efficiency. $2.6 billion to acquire Electronic Data Systems Inc. a few There is indirect evidence (statistics are lacking) that years ago). stand-alone NCMT machines are utilized between 2 and 3 For a fully computer-integrated factory of the future with times as efficiently as conventional manually operated ma all major functions assisted or carried out by computers linked chines. One early study in Germany concluded that one NC Fraction Software Milit ary Fraction 0 .9 New Manufacturing of Total Span of Computer Control Ad ded to Prior Level Average Manuf acturing Inve stment 08 - 0 .7 0.02 S t o n d~ o l one Mocn1ne Instruct ions for machine control 0.03 Machining Center Instructions for changing tools 06 0.04 Moching Cell Multiple machine control 05 0.06 FMS(l ) Scheduling 0. 10 FMS( 2) Load1ng/Unlooding, Storag e 0 4 0. 15 FMS(3) Inspection. Sorting 0.3 .Ci 0.20 Autom ated Production Li ne Assembly, Pa11et 1zing . Kitting 0.40 Au tomated Factory( 1) Computer1zot1on ot Functional Modl1les 0 - viz . MIS. MRP , CAD. CAPP, CAM 0 1 0.50 Au tomated Foctory(2) Linkage of MIS. MRP, Order Processing. 0 ScnedlJling. Cost Analysis 0.70 Au tomated Foctory(3) Linkoge of CAO , CA.E , CAPP & CAM 1950 1960 1970 1980 1990 2000 Yeor Fig. 8. Fig. 9. Software as Fraction of Total Capital Investment 50 Manufacturing Review vol 2, no 1, March 1989 data that would permit reliable quantification of this trend on an aggregate level. Form'"'b'• c~tt•n~ Sp ••d. lut pe< m'nut• (slpm ) Uo Fig. 11. Fraction milling machine displaced 3 conventional drillng machines one NC milling machine displaces 2-3 conventional milling 0 9 I- machines, and that one NC machining center displaces 2 c a ~ Avercge Ma chine Utilization: Sma ll Batch drills, 1 milling machine and I boring machine [15]. Similar Average Machine Ut ilization: Lorge Botch substitution ratios appear to be applicable to recent FMS in 0 .7 L- Ave rage Machine Ut il izat ion : Moss Production stallations. Since NC machines operate no faster than conven Average Utilization Rote: St and-Alone 0.6 I- Average Ut ilizatfon Rote: NC in FMS tional machines, all of the increased output can be attributed to more efficient use of working time, especially faster set 0 5 ~ ups. In short, unproductive time is sharply reduced. 0 4 I- When the NC machines are linked by programmable ma terials handling equipment and controlled by a central com ::: ______-~ ~ ~ ___ £r- ___ -8- - --- A- ----~ - puter to manufacture a family of parts with closely related de signs (in an FMS), the machine utilization rate tends to be significantly higher still. In part, this can be explained as a 1960 1965 N70 1975 1980 1985 1990 1995 2000 reduction in scheduled idle time, a straightforward economic Year response to the increased cost of capital equipment. But the utilization rate of machine tools is also a useful measure of the capability of CIM. A schematic illustration of the trend is Fig. 12. The Impact of Computer Control on Machine Utiliza shown in Fig. 12. Regrettably, there are very few published tion Ayres: Technology Forecast For CIM 51 In regard to all of the above measures, it must be empha REFERENCES sized again that they are hybrids, reflecting elements of both I. Ayres, R U (November 1987). Manufacturing and human labor as infor pure technological performance and market penetration. This mation processes. Research Repon RR-87- 19 of the Intl Institute for Ap plied Systems Analysis. Laxemburg, Austria. seems to be unavoidable. 2. Ayres RU (Fall 1988). Self organization in biology and economics. Inter national Journal on the Unity of the Sciences I (3). 3. Ayres, R U (March I 988). Complexity, reliability and design: Manufac CONCLUSIONS turing implications I (!),Manufacturing Review NJ. p. 26-35. The projections displayed already in Figs. 8, 9, 12 can 4. Wright, P K and Bourne, D A (l 988). Manufacturing intelligence, Addi son-Wesley, New York. be regarded as forecasts . Together they present a picture of 5. Tchijov, I and Sheinin, R (1989). Flexible manufacturing Systems (FMS): the future of CIM technology from a fairly high level per Current diffusion and main advantages. Journal of Technological Fore spective, as promised at the start. However, some important casting and Social Change. In-press. 6. Hess, G. Ingersoll Milling Machine Co, Rockford IL, 1986. Mimeo. topics have not yet been addressed in specific terms. In this 7. Ayres , RU and Funk, J L (1988). The role of machine sensing in CIM. section we will attempt to fill some of these gaps. We also Robotics and Computer Integrated Manufacturing 5 (1), p. 55- 71 address the important issue of "push" vs "pull." In the latter 8. Boothroyd, G (1980). Design for assembly-A, Research Repon of the Dept of Mechanical Engineering, Univ of Massachusetts, Amherst MA. context it is also important to not the existence of drag ef 9. Boothroyd, F and Dewhurst, P (1983). Design for assembly: A designer's fects, or bottlenecks. Of these, the greatest is the difficulty of handbook 2nd edition. Mech Eng Dept. Univ of Mass. Amherst MA. capturing human manufacturing skills in intelligent machine IO. Ranta, J, Koskinen, K, and Ollus, M (1988). Flexible production automa tion and computer integrated manufacturing: Recent trends in Finland. controllers. Computers in Industry IO. One topic that has been neglected thus far is logistics. 11. Shah, R (l 987). Flexible Festigungssysteme in Europa: Erfahrungen der We have noted that one of the most important expected bene Anwender. VDI-Z. 129 (10). p. I3-21. 12 . Bose, PP (April I988). FMS automates F-16 production. American Ma fits of CIM is a reduction in order-to-delivery time. This will chinist. p. 43- 47. result, in part, in a reduction of actual manufacturing cycle 13 . Kozar, Z (1988). Personal communication. time. Its real benefit, however, is to make possible a dramatic 14 . Machine tool technology (Special repon 726) (October 1980). American reduction in inventories, which represent idle capital, and to Machinist. 15 . Gebhardt, A, and Hatzold, 0 (1974). Numerically controlled machine serve customers better. tools, in Nabseth, L, and Ray. G F (eds), The diffusion of new industrial There are two distinct aspects to this problem represent processes: An international study. Cambridge Univ Press. Cambridge UK. ing two different situations. The first concerns standard prod p. 22- 57. 16. Merchant, M (1980). The factory of the future: Technological aspects. ucts. At present, when a customer orders an existing standard Proceedings of the Winter Annual Meeting of ASME, Nov I- 6. product it must be taken from an inventory of finished goods 17. Advanced manufacturing equipment situation and outlook (1985). Com ("off the shelf"), most likely sitting in a warehouse in the mission of the European Communities. p. 1I2-ff. wholesale trade sector. The second situation has two subcases. If the product is an existing one it must be ordered from the factory where it must await the next batch of that particular item, or it must be treated as a special order (batch-size of one) and jump to the head of the queue. In the absence of flexible manufactur ing technologies, special-ordering of spare parts (e .g. steam turbine blades) is a very expensive procedure, since the rou tine of the entire production facility is interrupted, and no economies of scale can be realized. Yet, in the case of spare parts for major items of capital equipment, like steam tur bines, these high costs are still preferable to the opportunity costs of idle generators. In fact, there are quite a lot of exam ples where a spare part must be produced as fast as possible, Robert Ayres is Professor of Engineering and Public Policy at almost any cost, to keep a large system operating. Spare at Carnegie Mellon University and the Deputy Program parts inventories ameliorate the problem, to a degree, but in Leader of the Technology, Economy and Society Program at some cases there is simply no possibility of having a spare the International Institute for Applied Systems Analysis available on site for every part that might fail. Thus, the po (IIASA) in Laxemburg, Austria. He is also the Project Leader tential benefits of a quantum leap forward in flexibility are (currently on leave) for the Computer Integrated Manufactur unquestionably very large. From this point of view alone, ing Program at IIASA. R. Ayres is the author of eight books there is ample economic justification for the large investments on technology and economics, of which the latest is The Next that are now being made in CIM technology. 1\-R Industrial Revolution, published in 1984. 52 Manufacturing Review vol 2, no 1, March 1989