Figure 11. Parameter model and algorithm for the self-maintenance copier. From Ref 19; reprinted with permission from ASME. Comment Qualitative models appear to have many advantages, among them being the ability to represent important behavior patterns with relatively simple models. However, it remains to be seen whether they can ever attain the sophistication and complexity of welldeveloped quantitative models, such as those exhibited by the automobile companies for simulating crashes, skids, and air flow. Tomiyama feels that QP not only can coexist with quantitative models but has the potential to model types of phenomena that cannot be represented in traditional models. My feeling is that augmenting quantitative models with logical state variables could increase their generality and a meeting ground in the middle might be found. HITACHI-SEIKI (H-S), 21 August 1991 Background Our hosts were Mr. Kubota, R&D Center Director, Mr. Ishihara, Controls Design Manager, and Mr. Takeshita, General Manager of the Technology Development Center. Other interesting participants were Mr. Otani, General Manager of the Design Department, and Mr. Katoh, Manager of the Technical Research Department. H-S makes machine tools very similar to Yamazaki's, but the company is about one-fourth Yamazaki's size. It is in a similar state regarding CAD and is less modern in its manufacturing facilities. However, H-S uses its own FMS for making almost every part it uses. No other kind of machine tool or system was observed. In fact, H-S uses not only FMS but also several temperaturecontrolled rooms for machining and assembly for the purposes of improving quality and of impressing customers. The company has 1,700 employees and Unlike Yamazaki or any of the H-S's main business problem is managing fluctuations in demand. Five years ago the transfer line business fell apart and many people had to be relocated. Today they have 50% more design business than they can handle. "It's a big panic." Their CAD does not help them much. They work nights and weekends and take on marginally capable subcontractors whose work they must often redo. H-S is the only company to admit to me that this type of overwork hurts product quality, including design quality. However, you can't say no to a sale, especially from a car company. They know when to buy, that is, when the suppliers are hungry. H-S notes, in a fascinating aside, that they used to be able to count on a lot of design help from the car companies' own engineers but that's not true any more. Those engineers now have only 2 or 3 years' engineers now have only 2 or 3 years' experience and H-S ends up teaching them! of part names, machine models, part size and material, and year the part was designed. It is a memory-jogger system that behaves like a library card catalog. Any way the designer can think of to help him remember a part is available to help him narrow the search. The output is a part number. He must then go to the CADAM system and punch in the drawing number in order to get a picture of the part. They have brought in outside consultants in the hope of making their factory floor more efficient. Right now a few hours of actual cutting time takes 10 days. All they got from the consultants over 2 years was advice to group all the needed machines around the parts, which is a variety of flow shop. Such an approach is bound to be costly and inappropriate for a job shop environment. Analytical techniques exist to address this problem, but it is a tough one. The FMS was invented in the first place to address just such situations, and nothing much better has come along since. Only Yamazaki has learned how to design its parts to make FMS particularly efficient. H-S's president has challenged his people to become more standardized in their design methods. They are still in the process of interpreting this challenge and may again utilize outside professional consultants. Right now their only response is to consider more reuse of existing designs and more modularization of their subassemblies. They do not have a strong tradition of the "series method" described by Yamazaki. In general, they do not have a clear view of the importance of design methods, designed characteristics of their machines, and design technology to improving their company. A major reason why H-S is having H-S relies on CADAM. Ninety-five mechanical engineers access 28 terminals and about 20 drafting boards. They would buy more terminals if they were not so costly (average ¥400,000/ month including the host IBM 4341). They think workstations will be their next move. All the facilities of CADAM are used, including generation of NC data. This is sent via fiber optic cable to the factory control room. Eight months ago they got CAEDS, the IBM version of SDRC's I-DEAS solid modeler, and they are just learning how to use it for stress and deformation analysis of machine tool beds. Even if they get a terminal for every designer they will not abandon paper because the CAD screen is too small. Other companies have said the same thing. Product Design Methodology H-S has two product lines, medium size NC lathes and machining centers and transfer machines. The NC machines take about 2 years to design: 1 year for concept design, 6 months for detailed design, and 6 months for preproduction tryouts and performance documentation. This leaves out fabrication of the prototypes, but I assume it is included in the first 18 months. The transfer lines present a totally different kind of design task, since they are made up of a linear string of standard modules. Each module drills a set of holes, machines a flat, taps a hole, or does some other simple operation on an engine block or other similar item. Figuring out what each module should do and wiring up the controls are the major challenges. Such a machine takes about a year to design and build. A set of four almost ready for delivery to Nissan were scheduled for customer buyoff the day we visited. Our hosts and I together estimated that about 112 man-months of design effort were involved in this system, which had over 75 modules. Half these man-months were obtained from subcontractors, as were many of the machines' more standard elements like conveyors. As far as I could tell, there were no computerized methods in use for determining line balance or job assignment in the design of these transfer machines. One experienced person accomplished these essential conceptual steps in about 3 months. the aim of obtaining a final accuracy requirement. This is much more sophisticated than their current method of just using higher accuracy components. Similarly, cost analysis is based on past data, especially since 80% of the parts are identical to existing ones or nearly so. New parts are hard to predict. The main components of cost are materials, vendor costs, and machining time, expressed in standard hours. To support all of this design activity, H-S has 95 mechanical engineers and 35 electrical/electronic engineers. Another 20 engineers are in R&D. About half of the engineers are in production and manufacturing. They R&D Activities comment extensively on the product designs but apparently do not take a large part in product design itself. About half the product designers are skilled in production techniques already. Even so, H-S agrees that there is too much delay and correction to the design drawings; about half are returned for changes or questions. Engineers gain their breadth of experience from a short 4 months initial training plus continuous rotation until age 45 or so. After that, an experienced designer stays put. Those that do not understand production methods and new technology are "a headache." H-S is having trouble attracting new employees. North of Tokyo is not considered "in" territory. Other Hitachi companies located in the area said the same thing. Otani, the Design Manager, operates with a number of rules of thumb. For example, to do cost estimating, his first test is to count the parts and multiply by ¥10,000. To determine if his designers will be over- or under-loaded, he multiplies the predicted sales by 10% and compares to his shop's salaries. He can support ¥6,000M of design effort per year, so annual sales over ¥60,000M will force him to go to outside contractors or overtime. There are few systematic design techniques in use. Tolerances are determined by reference to previous “experienced parts." Only recently have they begun to adopt the idea of the "error budget" for allocating tolerances with H-S has originated a number of interesting concepts. From their own experience using their products, they realize that setup time is a major headache for customers. So they have developed novel techniques for easy setup and programming of their machines and easy in-process measuring of parts. This is based on programmable logic controllers (PLCs) that they build themselves and link up to Fanuc controllers. The PLCs do the work while the Fanuc controllers provide a familiar name and an easy user interface. Other R&D activities include direct drive spindles (like Mazak's, I presume) and two-component casting of machine ways. These castings are cast iron underneath plus a layer an inch or so thick of steel on top, where the ways are machined. My hosts could not explain how these parts are cast. New products include laser heat treating and machines that grind ceramics, especially for bearings. Kyocera is a big customer. Their product has half the market because it is priced aggressively, the result of its being just a modified NC machining center with some novel tool attachments. Factory Floor Tour This factory makes more specials and fewer ordinary machines. The “mass production" work is at another factory where five machines a day come out "right on schedule." Assembly line repetitive working conditions drive many of the employees away. They can easily get jobs nearby since the whole area's manufacturing economy is growing. priority in this unstructured environment and of determining the overall behavior of the system by choosing which parts to load into it. This system resembles in both physical layout and human involvement a system build by Sunstrand for Ingersoll-Rand almost 20 years ago. It was very efficient and easy to schedule. Regardless of the kind of work, the Abiko plant is fully automated in machining and partly so in material handling. No automatic guided vehicles (AGVs) are visible as at Yamazaki, however. There are several FMS lines, Future CAD Needs most about 8 years old. Rescheduling them is a problem, since it takes half an I got in English only part of the long hour. It appeared that the larger sys-list Otani gave in Japanese in response tems were not very efficient and not well utilized. At least one suffered from an old design error, namely, a single straight line parts transfer vehicle with one-part carrying capacity and no buffer positions at the machines. Moving parts around is like solving the cannibals and missionaries problem. This error was discovered at Caterpillar in the mid-1970s. The more successful H-S FMS have more buffers at each machine and a continuous flow conveyor system with parts circulating past the machines all the time. The advantage of this setup was also recognized in the mid-1970s. to this question. He sees the potential of solid modeling as an input to CAE such as static and dynamic analysis. The main use for this is to shorten the 6 months of performance validation that is currently required of new designs. He also wants to be able to design oneof-a-kind machines, such as transfer lines, more efficiently than he can now. These are not money-makers in the current environment and use up a disproportionate share of his engineers. More extensive use of standard elements would be a likely outcome. He calls this "automated design." FUJITSU TECHNICAL CENTER AND FUJITSU LABORATORIES, KAWASAKI Significantly, H-S plans to junk the oldest and least efficient FMS and replace them with stand-alone computerized numerical control (CNC) machines. It may arrange some of these 22 August 1991 into part-type-specific cells. One L-shaped cell for spindles has already been tried successfully. The most interesting system would interest Prof. Okino of Kyoto University. It is called Holonic and won a prize for H-S in 1988. This system consists of a number of machine cells and continuously circulating parts. There is no overall schedule. Instead, parts are marked with bar codes that permit each machine to recognize the part and its required remaining work. A part can go to the next machine that is capable of doing the work it needs next. The operators have "human involvement" consisting of giving certain parts Background been slowing for years) because mainframes are used in large communication systems and commercial systems like those for banks, automatic teller machines (ATMs), and airlines. Fujitsu's introductory video stressed use of computers to do things "the way people do." Applications under study include neural net machines for machine translation and robot controls. Other work is on x-ray lithography at 0.2 micron line width, coherent light digital optical communication, optical memories, ceramic circuit boards with approximately 50 layers (probably for use with multichip modules although that terminology was not used), and real time image processing. Fujii's people mostly do electronic design, so most of their CAD/CAM is in that area. He showed only mild interest in our group's work in feature-based design and modeling of mechanical assembly processes. He may have felt that I was proposing to replace the designer with a computer program, whereas I was proposing a tool to help the designer sort through many possibilities. His final judgment was diplomatic: when solid modelers become easier to use, then our work will be of great interest. However, he stressed that design is very complex and the designer considers comprehensively all the factors at once. Such a process cannot be given to a computer. Our hosts were Mr. Fujii, General Design for Assembly (DFA) Manager of Corporate Manufacturing Systems Development, and his assistant, Mr. Asada. Fujitsu is a large diversified company focussing mostly on computers and electronics. Its major products are mainframe and supercomputers, as well as laptops, high performance disk drives, array processors, low noise super transistors for use in space telescopes, and so on. Unlike IBM and Hitachi, Fujitsu has not suffered a drop in mainframe computer sales (although sales have Mr. Miyazawa gave a short presentation about Fujitsu's home-grown DFA methodology. It employs a PC to take in data from the designer about the assembly, including part descriptions, part mate types, and so on. I remarked that all of this data would be in the feature-based solid modelers of the future. The DFA method has four characteristics: to provide a quantitative estimate of ease of assembly, to give a rough estimate of (manual) assembly time, to improve the design, and to be easy to use. Its objectives are to reduce the cost of the product, to shorten the development time, and to provide an easy way to communicate to factory personnel about assembly problems. In these ways the method remarkably resembles Hitachi's, which they know about but have not bought. My report is sketchy since, like Hitachi, they do not want to give out all the details. his opinion is merely to make the product easier to make, shorten development time, reduce costs, and reduce the number of prototypes needed. He specifically feels that CE does not (cannot?) address product quality and function. The design for assembly/ manufacturing (DFA/DFM) aspect of CE makes it hard enough already for the designers, and he worries that they cannot remember everything. So he is in favor of computer tools to help them. To use the method, the designer Aside makes an assembly flow chart (presumably an assembly sequence) and types in data about the parts and mates. The computer scores 30 different characteristics, such as direction of assembly motion, part size in relation to direction, the method of attachment, whether the part is flexible, how many moves or actions are required to affix the part, and so on. Unlike Hitachi, Fujitsu's method includes some interesting twists on part characteristics. Parts are designated as “main,” “subsidiary,” and “fastener," among others. Fastening methods are designated as single action, snap action, and so on. The scoring is done separately for each class of part and a profile is made up. A perfect score would be a product with one side, one direction, single action assembly of rigid, little parts. The score is presented either as a spider chart or a profile and is accompanied by the scores of recent similar products. From these charts, obvious ways the product is worse stick out clearly: too many awkward fastening methods, too much effort to put in subsidiary parts, not enough simple actions, and so on. Using this method, the estimated manual final assembly time of a laptop computer was reduced from 6 minutes 20 seconds to 5 minutes 10 seconds. After this presentation, Fujii remarked that DFA is part of Concurrent Engineering (CE), whose job in Several companies have their own DFA. Each shares basic properties but each has interesting additional elements that reflect characteristics of the products or of the underlying philosophy of what DFA ought to be able to accomplish. Nippondenso is the broadest, but Fujitsu is the most careful in identifying different kinds of parts and thinking that they should be scored differently. No basis is given by any company for the different emphases; they just appear obvious to the developers. New Method for Mold Design Linked to Rapid Prototyping The brief presentation was in Japanese with interpretation. The overheads were also in Japanese, so I could not follow too well. The laboratory tour afterwards was better. The old method of mold design was a series of steps, each with an approval before the next one could start. The steps are, approximately: initial part design, design verification, mold design, and mold feature evaluation. Each step was conducted by an expert. Too many people had veto power, too many steps were done sequentially, etc. The new method uses the same experts but launches them simultaneously. Significantly, it was stated that merely doing this does not save any time. In addition, the team needs some computer tools. The ones in use at Fujitsu are 3D structural design using SDRC's I-DEAS solid modeler, mold flow software, and stereolithography. The latter is implemented on a machine called SOUP (Solid Object U-V Laser Plotter) made by Mitsubishi. Input to this machine is via I-DEAS. (Stereolithography is a technique for making rapid prototypes of solid parts. It functions by forming a part layer by layer, either from a liquid or a powder. Laser scanning is commonly used to form each layer. In the SOUP and similar machines, the laser scans the top free surface of an ultraviolet (UV) curing liquid plastic. When the layer has formed, the object is lowered about half a mm and the process is repeated. Parts with re-entrant features can be made as long as a hole is left to remove the uncured plastic. The layering gives the part a stepped outer surface finish just like "jaggies" on computer drawings. These are usually handsanded off. Stereolithography is capable of creating the shape of a part, and methods exist for using it to make metal molds from which parts of the correct material and strength can also be made. The method permits people to quickly "get their hands on the part," including communicating to factory or subcontractor personnel in ways that drawings cannot.) Fujitsu uses the SOUP output to make a mold from silicone rubber. Accurately made silicone inserts are used to create some mating surfaces, with the result that mating pairs of molded ABS parts assemble with surprisingly close-fitting mates. A companion rapid prototyping facility is a Fanuc NC machine, which is programmed from a ProEngineer system. (ProEngineer was chosen specifically for its easy user interface, suitable for the technician they have assigned to operate this facility.) This machine makes smooth profile molds |