The figures in Table 1 accurately reflect the total manpower employed. Few or no assistants such as draftsmen

are used.* “Engineers make their own drawings." Technicians and test engineers for laboratory evaluations of designs are not included, however.

Part count in these products (except cars) is in the range of 100 to 1,000. As a rough average, one designer may be responsible for 20 to 50 parts. These statistics are remarkably consistent, as are the times quoted for converting market requirements into a final product: 1 to 2.5 years for all of the above except cars (4 years).

Teams of 20 engineers are unlikely to have serious communication problems, indicating that face-to-face communication and phone calls will be sufficient and computerized methods will be unnecessary. My Japanese hosts agree with this.

The teams are small for two reasons. First, they work feverishly and accomplish a lot together. Second, large products like cars are subdivided and many common components like alternators and air conditioners are bought from suppliers. Competition among suppliers keeps quality high and permits the final assembler to focus its design staff on the core items that determine performance, namely the body, suspension, and power train. Both Toyota and Mazak go outside for many of the higher technology items like controllers, high speed bearings, and integral machine tool spindle-motor assemblies, even if the specifications for these are drawn up in detail by the buyer.

The computer issues this practice raises are long distance communication, compatibility of design data, and compatibility of design software. Suppliers with many customers for the same product line face a serious problem since they cannot be compatible with everyone. Suppliers and customers both complain about this.

Living With Change. Many U.S. companies structure their design processes by sequencing the tasks in the hope of maintaining control and avoiding change. This usually requires many formal design reviews and formal transfer of information packages from one stage of the design to the next. Yet changes routinely occur and have to be absorbed. When it comes to change, Japanese companies put their heads in the lion's mouth by adopting the overlapping tasks methodology. I asked repeatedly if this did not risk many design changes. In every case I was told that external pressures from the marthat external pressures from the marketplace force even bigger changes on the process. The choice is to resist change or to learn how to live with, or even profit from, it. These companies have chosen the latter. Right now, as with everything, they rely on internal communication and experience, plus long hours from their staff and fast action by

top management, to mechanize change absorption. Many also have elaborate document release and design review mechanisms, including attendance at reviews by designers from other projects.

No better use for computer design tools can be imagined than coping with change. Data and documents must be revised, notifications must be sent out, and new test data must be compared with old. Databases must be designed and search methods created so that people and data affected by changes can be identified easily and automatically. Design tools that permit redesigns to be made and simulated quickly are also necessary. Finally, any use of such tools to avoid extra and timeconsuming prototypes in the first place makes time available for absorbing change. Many companies use the words "virtual design" or "virtual manufacturing" (the latter coined by Prof. Kimura) to describe their aim.

Standardization of Design Tasks. This topic, while mentioned by several companies, means something different to each. Basically, companies do not want their engineers to grope, but instead to know what to do, when to do it, and how. They want to reduce the detail that must be communicated as well as the need for lubrication from personal friendships and past design efforts. Global companies want all their overseas engineers to act like domestic ones so that their designs will be predictable and uniform. They want computer tools that contain the design process steps and have the necessary data ready for the engineer. In some cases, such as electromagnetic design (motors, alternators), companies have developed spreadsheet-like design interfaces that take in specification data and output performance curves. Only manual tradeoff analyses are available so far, but design optimization is being sought. One company wants a computer system that will literally orchestrate the actions of many designers on a network, requesting parameters from them one by one, performing some portions of the design automatically, and distributing the results back along with the next round of requests. It would base this system on its existing "design standard books."

Feature-based design and constraintbased design are current research topics that have a potential bearing on standardization. These attempt to provide a designer with the ability to deal with geometry on a CAD screen that is linked to a data file of attributes which give the geometry an engineering identity. Thus a cylinder becomes a feature called a tapped hole, complete with process plan, tolerances, and assembly insertion direction. A rule or constraint can say that the hole must be at least one diameter away from the edge of the part. The data files and constraint rules underlying the geometry provide

standardization and save the designer time. The rules warn the engineer if a violation occurs.

No companies have such software, and none is commercially available. However, a few companies have impleHowever, a few companies have implemented their own primitive featurebased design for some machined items and linked it to semi-automatic, knowledge-based process planners. These computerized process plans are more consistent than those produced by human planners. Everyone asks for rule-based systems that not only warn of violations but also recommend how to change or improve the design. The difference between these two capabilities is vast.

Bottom-Up Computerization. A researcher at the IBM Tokyo Research Laboratory said to me, "The United States is too top-down oriented and Japan is too bottom-up oriented." He was referring to the two countries' different tendencies regarding use of computers. The United States, in his opinion, rushes in to computerize something without really understanding it first. Often this means converting an existing manual process, in a factory or office, step for step into a computerized one. Grave inefficiencies are often converted at the same time. A careful analysis of the requirements on the process combined with the new capabilities of computers to meet those requirements would likely produce a rather different and more efficient process. The Japanese tend to study and improve a process manually for years (“kaizen” or continuous improvement) before computerizing it. Ironically, this often produces a process that runs very well without computers. This must hurt IBM Japan's sales, but the contrast is a real and instructive one.

The attention paid to design process organization in Japan is impressive. Where computers have been

applied by buyers of CAD, the applications reflect availability of software. But where it has been applied by userdevelopers like Nissan, Toyota, and Nippondenso, it reflects their own assessment of priorities. All companies have basic two-dimensional drafting on computer. Only a few have supercomputers for detailed fluid dynamics analyses. An informal survey of visit data indicates that product analysis has priority over process analysis and that, among process analyses, forming (cutting, molding, bending) has priority over assembly.

At Nissan, the historical sequence of computer applications over the past 30 years has been

⚫ Data processing and coordination of test data

• Design specification control, document control, "parts trees” (what we call material requirements planning or MRP)

• Efficiency of engineering

Higher quality (better testing, smoother outer body panels)

Simultaneous engineering, including fusion of the above data and software

• Worldwide communication

No company visited has a common database in the computer science sense, but many use that term to describe what they have. This is usually a database that many designers can access and many software modules can read from and write to. The necessary data conversions are often error prone and time consuming, especially when manual intervention is needed, such as in meshing for finite element applications.

Description of Some of the Research Issues

Rationalization of the Design Process Itself. Many Japanese companies see study and improvement of the design process as a crucial element of corporate development. They take different approaches and emphasize different things. Few have systematic approaches. Apparently they conduct post-mortems, although no direct evidence of this was observed. The main topics I identified before and during this project are:

⚫ task sequencing

• prioritizing design improvements ⚫ identifying tradeoffs ⚫ mustering experience

Task Sequencing Understand ing the design process requires seeing it in enough detail that opportunities for improvement can be identified. Saving time by resequencing has been recognized by one company, IHI, and an attempt to implement it with CPM is starting. Our own research (Ref 4) has identified an approach developed originally to help solve systems of simultaneous equations. At General Motors (GM), a task modeling technique developed by the U.S. Air Force called IDEF has been used to obtain and structure information about existing design procedures. No comparable methods appear to be in use in Japanese companies. Formal process representation, analysis and, especially, improvement methodologies are in their infancy.

Prioritizing Design Improvements. Setting the priorities for which areas of the process need improvement usually involves the expected factors of time and cost. Car manufacturers early identified body engineering as a long pole in terms of time and cost and have focused rationalization and CAD/CAE/ CAM on it. Nippondenso studied the

cost structure of automation as a function of how many models the machine was supposed to handle. They found that the cost of handling and feeding the many different parts grew faster than any other cost component. Avariety of approaches is now being pursued: low cost bin-picking (a goal for a new project with no results yet) and designing so that the same part can be used in many models or so that one gripper can grasp many kinds of parts. Nippondenso's highest priority is designing product and process together so that a change from one model or version to another can be accomplished essentially without stopping the production line.

Identifying Tradeoffs. Functional and process designers have conflicting needs. When design begins, "fights start almost immediately." Successful negotiation of these conflicts often benefits from finding a win-win solution. In a complex design, this can be difficult to do, and zero-sum solutions often appear to be the only ones. (We need to make the product lighter, so make the walls thinner. If too much heat flows through the walls, you can just find a way to dissipate the heat.) Hitachi and Seiko-Epson have adopted the slogans "user first” and “common goal," respectively, meaning that the design team should do what will benefit the customer the most. This is a good spirit but it is not quantitative. At present, engineering models of most products and processes are too weak to permit modeling a product completely in mathematical terms, preventing use of formal analyses of mathematical structures, for example, as a way of finding tradeoff opportunities.

Mustering Experience. Every Japanese company is proud of its accumulated experience and how it is used to make better products. This is both a source of strength and of weakness.

Two potential approaches to capturing this experience have been taken: knowledge capture in expert systems and data archiving. Several companies have internally developed expert systems for specific tasks (shop floor scheduling, design of turn signal lamps, layout of car trunks and exhaust pipes, machining process planning), but everyone complains that there are few knowledge engineers, and methods of knowledge capture that engineers themselves can use easily are scarce. Several companies maintain data files of test results not only for comparing new and old designs but for direct transfer into design software. None has a good way to search such databases.

Group technology (a way of coding items so that "similar" ones are in the same group) has been applied to classification of features as input to process planning systems, but companies want to be able to identify, classify, and retrieve experience in the form of past designs and process plans, which is much more ambitious.

Mazak has deployed past information about designs rather efficiently by using the "series design” method. They put considerable effort into designing a new machine and then rapidly design variants of it over the next several years by reusing existing CAD and machine performance data. The variants include larger tool storage systems, faster spindles, longer base, and so on. Six engineers can design a variant of a lathe in 2 years.

Managing Data - Integrating, Sorting, Classifying. Car, ship, and airplane designs involve huge amounts of data. More than one company said their use of solid modeling has been held back by its inability to handle the huge amounts of data efficiently. Instead, the companies must use simplified solid models or wireframes. Simplification omits some crucial details that are necessary for interference checking or

mold filling analyses, for example. Wireframes are impossible for factory personnel to interpret, impeding communication and product-process integration. Managing all this data is a serious problem. Managing it in an environment of constant change and overlapping tasks is even worse. However, one company said that it did not use a data management and control system, even though they are commercially available. This is a paradox.

to

Converting Experience Algorithms. Expert systems have been used in limited ways to capture expertise at some companies. Rules, “knowledge," and formulas are combined to create a machining process planner: the rules include METCUT data and the type of tool to use in certain circumstances, the knowledge includes how big radii should be or how feeds and speeds should be chosen, and formulas calculate wear rates and tool heating.

Such applications are relatively straightforward. On the other hand, no one has a way to convert the experience of a process or industrial engineer who judges whether something is easy or difficult to make or assemble. At an auto company, we saw an assembly engineer studying a 3D wireframe model of wrench access to tighten screws on several engine compartment parts: headlight assembly, washer tank, battery bracket. All are near each other and some interfere with the tool during fastening of the others. His priority was to use the same length wrench extension for all, adjusting the assembly sequence to make it possible. This meant using a short extension, which made one deep vertical insertion almost impossible. He accepted this solution, saying that the assembler could ram the wrench down the hole faster than the acceleration of gravity, thus keeping the screw from falling out of the socket! He was not interested in my suggestion that a magnetic or gripping

socket be used to hold the screw. I was told that he confers with the line foremen all the time and knows what he is talking about.

How does one evaluate whether this kind of experience and judgment is worth capturing and, if so, how to capture it and make it applicable to new tasks? One company said that all such decisions should be made based on cost. Laughter greeted my asking if they have a cost model for this.

Improving CAD for Process Engineers. Product designers have all the toys, it seems: finite element methods (FEM), supercomputers, etc. This helps them win a lot of arguments with the process engineers, who agreed heartily when I pointed this out. The first priority of the companies after supporting functional design and analysis is to make product design data available to the process engineers. Then they can at least simulate tool motions, robot actions, and cutter paths. No company I visited had fully accomplished this. The assembly engineer mentioned above had to position the wrench on each screw himself, using his mouse, but tons, and database information on the coordinates of the screw's axis. The screw head did not exist as a feature with an easily retrieved location, and no command "put the wrench on screw 22" existed. In fact, only one company showed me any assembly simulation.

No company has thought about assembly sequence analysis, much less disassembly (for repair) analysis. Nissan claimed that sequences can be worked out on the factory floor; once learned, they are not worth changing since model changeover time is too short for the necessary retraining. Sony says its engineers "know" how to plan assembly while doing functional analysis. Yet some redesign is still necessary when a product is switched from manual to robot assembly. Another company says the same thing but also remarks, "Please

don't show this (design) to Prof. Boothroyd," a well-known advocate of design for assembly.

Understanding What DFM and DFA Really Mean. DFM (design for manufacture) and DFA (design for assembly) are well-known terms. They typically mean adjusting the design to make fabrication or assembly easier or less costly. I was told that our group's work and that of Boothroyd have been very influential in Japan in simplifying designs. Boothroyd & Dewhurst, Inc. and Hitachi's DFA evaluation software are popular. Sony, IBM, and Fujitsu have developed their own DFA methodologies and software.

At Nippondenso, these commercial DFA systems are not used. The explanation goes beyond the fact that Nippondenso's products apparently are a little too big in its opinion or that the rules in those methods do not seem to improve Nippondenso's current designs. Rather, Nippondenso has raised DFM and DFA to a higher level, meaning the creation of a design that permits a new kind of manufacturing strategy to be pursued.

Nippondenso has classified product flexibility into increasingly difficult accomplishments and reached each level after about a decade's work on each. The goal is to make different models of a product on the same equipment with essentially no changeover time penalty. The simplest level permits combinations of different versions of an item's parts to be assembled. The next permits different numbers and kinds of parts to be included in a housing that is always the same size. The hardest and most recently achieved permits different sizes of the same product to be made on the same equipment. Each step required increasingly radical innovations in how parts are designed, fabricated, and assembled. Nippondenso has identified increasing flexibility (or "managing diversity") as a corporate

research topic and is seeking ties with universities in order to pursue it. A collection of good internal examples is also being compiled.

Nippondenso has its own DFA evaluation method, which is consistent with the above approach. Appropriately, it spans much more than the act of mating the parts, which is the focus of the Hitachi and Boothroyd methods. Instead, Nippondenso evaluates 65 factors covering such high-leverage items as ease of switching from one model to another.

University researchers seldom have the depth of contact with manufacturing necessary to identify a problem of this type and focus on possible solutions. It has taken Nippondenso several decades to work out a long-range plan with specific steps.

Nonetheless, it is difficult to generalize from the design innovations the company came up with, except for the simplest. These are part substitution methods developed 15 years ago for dashboard panel meters. Some of the recent ones for making alternators are based on converting flat strips of raw material into fully developed nonflat, nonstraight shapes in a multistep continuous flow process almost like paper making. Prior processes were stop and go or formed pieces directly in final shape with consequent waste of material and need for lengthy die changes. A few years earlier, Nippondenso applied similar techniques to improving flexible manufacture of radiators.

Fujitsu's DFA method stands between Boothroyd's and Nippondenso's in sophistication. It classifies parts in several ways (main, subsidiary, rigid, flexible) and scores the assembleability of each class separately. Assembly time and cost are estimated. An assembly score profile results and is compared to the scores of other products. Priority in redesign is given to eliminating nonrigid, nonmain parts and to simplifying the assembly of the remainder.

Sony has a DFA method very similar to Hitachi's. A difference in emphasis is that Sony requires its designers to use it while sketching possible designs. The DFA score is one important way that alternate concepts are prioritized during this conceptual stage.

Toyota uses no formal DFA and asks quite seriously why anyone would need such a tool. Regarding wellpublicized DFA activities at GM and Ford, Toyota designers ask if communication between designers and manufacturing engineers is really that weak at those companies.

These differences in approach and attitude indicate that the role of assembly analysis in product design is still evolving and capable of considerable improvement.

TYPICAL APPLICATIONS OF COMPUTERS IN DESIGN

In general, U.S. computers, both mainframes and workstations, and U.S. software dominate in Japan. Due to space limitations in offices, Japanese laptop computers are seen everywhere. Except for a few programs, nearly all commercial software is from the United States.

Specific applications of computers in design were much as one would expect. What is sometimes surprising is the depth of penetration of networked computerization at some companies (3,000 workstations, 1,000 Macintoshes, etc.), the degree of integration of many design steps in one computer system, and the commitment to growing their own capability internally and through joint ventures with software houses. In design, most companies visited are paperless or nearly so. However, paper is still valuable: no screen is as big as Esize paper, and huge drawings are commonly seen covering tables surrounded by conferring engineers. The factory floor people still want paper because it survives, can be marked up, and can be met over.

Figure 1 is a summary of design and product realizations actually observed at 13 companies visited.

The data in Figure 1 were sorted and cross plotted and appear in Figure 2. Companies are arranged across the top, sorted left to right by decreasing number of the computer technologies observed at each company during my visits. The technologies listed in Figure 1 are arranged down the left side, sorted top to bottom in decreasing number of how many companies they were observed at. An entry of "1" means that the technology was observed at the company. A "O" means it was not observed.

The "data" behind this plot are not particularly sound statistically since they represent what was observed. Especially at a large company, something not observed is just that and is not necessarily missing. Nonetheless, the "data" are interesting and suggestive. Sorting the technologies by their commonness across the companies shows that some technologies are very common, and the ones that are do not surprise us. Sorting the companies by how many technologies they have undertaken shows which companies are the most aggressive and advanced. Comparing these two kinds of data by cross plotting the sorted lists allows us to determine if companies have built up their computer design capabilities from the common to the rare or whether companies can jump in at any level in the hierarchy.

The ability to draw the diagonal line and contain most of the “1”s above and "O"s below indicates that computer technologies in design are accumulated and represent a long-term company effort to build capability, understanding, and infrastructure. It argues against computers being commodities. If it were possible just to buy computers and be able to "play with the big boys," then one would see "1"s all over Figure 2.