the computer's suggestions being real is enhanced if they are built up from verified elements.

Thus his model is that the designer can manipulate such elements in what he calls a "virtual factory" that contains models of machining, molding, and assembly processes. These models may be approximate but that would be good enough for preliminary design. Aside on Model Accuracy

The idea that a rough model with imprecise or incomplete data should be sought for early design is consistent with Toyota and Nissan being willing to launch downstream design tasks when only partial information is available. Body styling data accurate within 1 mm are available months before the last mm is pinned down. Yet nearly all of the design of dies can be done within 1 mm and finalized at the end. None of the stamping simulations is likely to be affected by a 1-mm change.

Thus a major component of research in support of early design should focus on determining just what information is of real value in launching any given step in the design and the level of accuracy at which this information must be provided. Such a study will likely reveal that much of the demanded perfection of data is not needed right away and that a preoccupation with perfection and finality is a time waster in many design processes. It also may be a time waster in many searches for good computer models of products and processes.

He cited Toyota's method of evolving design tools: make a simple tool that helps a specific design step; observe how the designers use it and get their suggestions for improving and broadening it to cover more of the steps. This way the tools grow organically and no useless tool is pursued or imposed on the designers. It is an example of a larger difference identified to me last year by Mr. Hazeki of IBM Tokyo

Research Laboratory: Americans jump too quickly into system software (equivalently are too top-down oriented) without thinking the process through first, whereas Japanese are too slow and spend too much time figuring out their processes and perfecting them manually before daring to put in any software (too bottom-up oriented).

I am also reminded of the difference in approach between Toyota and GM regarding die design software. Toyota has a series of elementary analyses that the designer can ask the computer to perform on a die shape. These are quite approximate but can avoid every known disaster. The totality of the programs is an end-to-end system for turning designs into finished dies. GM critiques this system (privately to me and to others) as not having any real engineering analyses or accurate simulations of the metal deformation process. GM has delayed implementing a similar system pending completion of accurate models, a step Toyota gave up on immediately as requiring too much computer power. Thus Toyota can be said to value integration over accuracy, while GM favors accuracy over integration. Toyota appears to be ahead, although Kimura jokes that CAD-designed cars seem to look alike.

Design and Manufacturing Problems. In this area, there is little work so far. It covers concurrent engineering, parametric design of routine objects, machining simulation, design for machining, assembly process planning, and standardization of CIM data. The group is active in STEP and other

standards activities.

The Relative Importance of CAD/CAM/CAE and Management in Product Development Strategies

As a researcher in CAD/CAM/CAE, Kimura feels that these topics are central to any successful product development

strategy. Thus it was interesting to hear this topic discussed by him and two researchers from Hitachi, Mr. Ohashi and Dr. Taniguchi. Both have advanced degrees from U.S. universities and are active in PERL developing software to aid manufacturing engineering, assembly evaluation, and concurrent engineering.

The Hitachi people reflect the opinion of the Nissan people, namely, that "90% of problems in product design methods can be solved by management, and only 10% can be solved by computer tools and computerized knowledge bases." They base this opinion on observing their company designing really new products for which there are no established procedures or tools. Especially important is the lack of corresponding manufacturing processes. Experiments, communication, redesign, and feedback are the essential elements, and the success of these is mostly influenced by management.

The main problem in complex design is that different groups or tasks have conflicting goals and there is no standard way to work these out. The team design method brings the issues out sooner but that only causes embarrassment. Hitachi's big point of pride is the "user first" slogan, which supposedly focusses the team on the customer and makes them forget their internal conflicts. What solution to conflict X will benefit the user the most? The fact that design tools could explore many possible answers faster is not appreciated. The reason may be that it does not seem to be a real option. It is just a promise by researchers.

In this context, what is concurrent engineering? Is it any different from simultaneous engineering? Are they just the team design method with computer support? Isn't concurrent engineering just something that Japanese companies have done for a long time?

Discussion on these points reveals little consensus except that all is not rosy inside Japanese companies. Some

people refer to "big business syndrome" in which top managers want to interfere and decisions take too long to make. Ohashi cites a recent Business Week article on concurrent engineering as a clever way to convince managers that communication is really important.

Kimura feels that computers cannot by themselves shorten the design cycle although they can improve quality and help improve the technological level of products. But shortening the cycle is almost exclusively a management factor. Thus concurrent engineering can help shorten the cycle.

The Hitachi people feel that concurrent engineering could help improve quality because they see the current push to shorten the cycle as threatening quality. There is too little time to do the necessary analyses and tests. Thus providing knowledge bases, integrating design and test data, and improving information flow are all ways that concurrent engineering could help the design process move faster.

Now that the United States seems to be catching on to the big secret of communication and team design, will Japan have to invent something better? They secretly worry about this but have no answer. It is the big debate inside Hitachi right now.

Closing Comments

Kimura's view of design research is both impressively broad and very pragmatic. His topics seem to be aimed directly at the main needs of mechanical design: higher levels of modeling that include symbolic components such as constraints, maintaining the link between design and manufacturing processes, and striving to support design in fuzzy situations like early design where answers are needed but data are limited.

The IMS idea was born in this environment and thus it has a very sophisticated intellectual base. Kimura surprised me by asking me a "very general

question:" How can the IMS project achieve its main goal, which is for Japan to relieve many international trade problems by giving its advanced manufacturing technology to the United States and Europe? Why are there so many political problems over IMS? Surrounding this question are his wonderment that the United States can develop advanced ideas like CAD and concurrent engineering with computers and still not spread them around adequately.

In one sense Kimura answered the question himself but I also offered an answer. The United States is behind Japan in its sophistication regarding manufacturing. It is not clear if we could adequately absorb Japanese technology if it were "given" to us. Thus I said that successful international projects must have well-defined technology transfer mechanisms built into them. Such mechanisms must include longterm hands-on contact with the technology and cannot be confined to delivernology and cannot be confined to delivering reports or presenting papers. Such conditions should apply to any kind of "manufacturing" technology, including management techniques, computer software, robots, sensors, and so on.

NISSAN TECHNICAL CENTER (NTC)

8 July 1991

Background

Our hosts were Mr. Jun-ichi Kobayashi (Product Development Section, planning Nissan's future CAD capabilities), Mr. Yoshida (in CAE user support), Mr. Katoh (Yoshida's boss), and Mr. Ono (Production Engineering CAE). Kobayashi was the most informative and interesting of these. He has been with Nissan since 1974 and spent 10 years in Manufacturing Engineering before coming to NTC to work on CAD/CAM/CAE.

Background of Nissan and the Nissan Technical Center (NTC)

Nissan is Japan's second largest car company, behind Toyota, and fourth in the world. In the last several years there has been considerable progress upgrading CAE and an effort is underway to link styling, design, engineering, and test data into one system world wide. Kobayashi acknowledges that this will be difficult for two main reasons: software incompatibility and people incompatibility. So far they have linked design and engineering offices in Japan, the United States, the United Kingdom, and Spain and can share data and software because (luckily, says Kobayashi) all these offices had IBM or IBMcompatible mainframes already when the integration effort began.

The Nissan Technical Center is in a beautiful location and comprises many modern buildings. "Amenities are becoming more and more important." The main design/engineering building, where we visited, is new since April and has an open, airy, atrium design. There are no offices, only wide open spaces with many large tables where meetings can be held. All the engineering disciplines for car design and prototype production are housed in this building except the engine people, who will move in next year. Thus, colocation has been adopted as a design efficiency strategy.

All of Nissan's car design in Japan is centered at NTC or at the Central Laboratory in Yokosuka, where fundamental studies are carried out in materials, power trains, electronics, and vehicle technology (body, chassis, user interface). In the styling and design. area (perhaps half of all vehicle design and engineering?) the engineers are working on several vehicle design projects at any one time. I estimate that 100 to 200 stylists and designers are at work on one car project (plus probably another 200 doing more detailed engineering).

Unless I misunderstand badly, this is a remarkably small number of engineers for the number of cars they design. Uncounted in this total are manufacturing engineers at NTC and at the factories.

I estimate that the typical new car development process takes about 4 years. Styling and design occur during the first 1.5 years or so. "The rest is trial, error, and design changes." In the first prototype, new components are often tested by attaching them to existing cars. Later prototypes use completely new equipment. Die design in particular takes about 1 year, but changes keep coming in during the last year before production starts. During the first 1.5 years or so, or whenever prototypes are being built at NTC, the factory people make frequent visits to NTC. As prototype work shifts to the factories, the NTC people travel there. CAD data do not travel well to the factories (see below).

Design begins while styling is still in progress. This is typical of overlapped jobs in Nissan and other Japanese car companies. At some point during styling, the exterior shape is frozen and this master data package is then passed on to downstream activities. Interior styling continues to be worked on after the exterior is frozen.

Apparently all Japanese companies feel pressure to reduce the design cycle but there is no specific program at Nissan to do so. "We already design cars in 4 years, compared to 6 in the U.S. and 8 in Europe. If we shortened any more, we would just get a lot of negative reactions from overseas. So we will use the ability to design faster in other ways." I was not told what ways, but obviously two are to reduce the time pressure on their engineers (who often stay until 9 or 10 p.m.) and to do more analyses on each design.

changes and in any case they come amid a stream of other changes anyway, so the important thing is to be able to respond quickly to changes. The CAD system contains no explicit change management system, according to Kobayashi.

Note that this explanation agrees to the letter with that of Prof. Fujimoto (see the Fujimoto visit report.) Even though Kobayashi is responsible for planning Nissan's future CAD needs, he refused to ascribe much importance to the use of CAD/CAM/CAE in speeding up the design process. In view of later examples (crash test simulations, search for assembly problems before the first prototype, etc.), this claim is disingenuous. But I did not try too hard to fight it.

[Via a separate communication, I obtained the following opinions from Prof. Hiroshi Sakurai of Colorado State University, a graduate of Tokyo University and former employee of Nissan. The main reasons why Nissan can design a car in 4 years are: (1) average 2,700 working hours per year for engineers versus 1,800 per year in the United States (Toyota people work even longer); (2) overlapping of jobs, for example, Nissan's body engineering begins when only rough exterior data are available whereas U.S. designers wait for smoothed data even though the difference is often less than 1 mm (Fujimoto cited precisely the same example!); (3) outsourcing of many components to small companies who work their engineers even harder and pay them even less. He notes that even in 1980 Nissan could take a car from exterior freeze to volume production in 2 years even though they had only a few CAD

terminals. But he admits that CAD/ CAM/CAE may have kept the lead time from becoming even longer and the number of engineers from growing. I note that cars are very much more complex now than in 1980.]

It is worth noting, when considering (see below) that Nissan has distributed its design software worldwide to its overseas design offices, that exporting the fuzzy design method has been a failure so far. Real problems thus remain.

Product Development Methods and Use of Computers

Nissan's CAD/CAM/CAE consists of many computers, programs, and databases. Nissan is in the midst of trying to merge all these into one seamless process but it is a hard job. What they have is like "islands of automation" in the factory world.

Nissan is in the process of converting from styling using clay models to direct styling on the computer. In the old clay model method, stylists carve quarter and full scale models and paint them to look quite real. These models are digitized and computer data created. In the new method, stylists draw shaded sketches which "technicians" (high school graduates) put into the styling computer to create boundary patch models whose shape resembles that indicated by the shading in the sketches. These models can be shaded and rendered very realistically, including showing light reflecting off the surface. The stylists sit with the technicians and adjust the shapes of the surfaces. Apparently the technicians make a very strong contribution to this process and are not mere data input drones. They must have both artistic and computer skills as well as the ability to make the stylists confident and comfortable in the presence of the computer. At some point in this process, the model is used to drive a large gantry NC machine which carves out a clay model which is then painted and judged as in the old method.

When the stylists are satisfied, the data become the master data for the car's exterior and are never changed thereafter.

For example, a torque wrench must have 60° swing room. Less requires redesign or an OK from the factory. Nissan tried using Hitachi's assemble

The boundary model data are passed, perhaps with some pain, to the design and engineering computer system. This system is separate and builds a new database starting from the master exterior data. Most of the work is done in 3D wireframe form using Nissan's proprietary CADII system (see history below). This system is good for the designers, who can put in as much detail as they want without swamping the data storage and manipulation capabilities of the software. No solid modeler can hold the detail and manipulate the data efficiently, they say. However, the wireframe models are useless when communicating with the factory people or the tooling designers. So there are problems.

The master data or the design version of them are also passed to supercomputers for various analyses (see laboratory tour below). Converting data for use by these programs takes a long time, another problem that adds time, effort, or number of people required.

CADII data are also used by industrial engineers for doing manufacturability and assembleability analyses. The assembly path of a part (we saw a windshield washer tank) can be visualized as the engineer moves a view of it around the screen using knobs. He can see interferences, view from several angles, try different paths, and so on. There is no analytical support for this. He can also attach images of tools to the heads of screws or nuts to evaluate access for tools. These issues are of the most importance in designing the engine compartment, which is the most crowded area. The success of this process depends entirely on the experience of the engineer.

While there is no analytical support for assembly analysis, there are criteria.

concluded that it is intended for small parts assembly of large production rate items for consumers. Neither Nissan nor Hitachi can do real designperformance-manufacturability tradeoffs, they say. Cost should be the criterion but no one has a method.

(Ten years ago, engine compartment design problems were found by making Xerox copies of drawings of parts and sliding them around on a table, or by making transparencies of parts and doing the same.)

In this way, hundreds of assembly problems are found before any metal is cut, comprising the majority of all such problems. There isn't enough detail in problems. There isn't enough detail in the CAD models to permit the rest to be found. Many of these involve flexible items like pipes and wires. It is typical that thousands of design changes to improve assembleability will be made after prototypes are built.

There is some indication that Nissan is beginning to adopt modular assembly, similar to what shipbuilders did 20 years ago. Examples given were entire dashboards built up in advance and then inserted into the car with a robot. This is, in fact, not new, since (1) you can't install efficiently things in a dashboard once it is in the car and (2) GM and VW have used robots to put in dashboards for several years. However, the module approach has a bright future, although it may require some redesign of the vehicle's structure. Sticking one part after another into the engine compartment is getting harder and harder to do. However, more success at getting around this problem with modules will mean that maintenance and repair will become harder and harder! The modules are likely to be big and heavy and require special tools and hoists to remove later.

In any event, if Nissan continues along this track, it has the CAD tools in hand to do the necessary studies in advance of building prototypes, an essential ingredient. However, the long term effort may not be directed at effecting robot assembly. Data in a handout show that Nissan ramped up from 540 robots in 1980 to 2,000 in 1985 but has only 2,462 in 1990, indicating real saturation at this level. "In 1980 we thought robots would replace people. We know now that it won't be that easy. Yet the goal is still 100% automatic final assembly."

Stamping dies and plastic injection molds for body and interior parts can be designed in about a year. All the important exterior dies are made inhouse. They are just beginning to use new nonlinear deformation software recently developed at the machine tool division. Up to now many problems had to be solved during die tryout. "Dedicated technicians are a big Japanese advantage," said Kobayashi. Stamping plant people visit NTC often during early styling design because (see above) the master data are frozen early. While some of this visiting may be done by passing CAD files around the network, most of it is done face to face.

A big problem, other than die design errors, is die design changes forced by other design changes. These cause a stream of changes throughout the final year before launch. Thus the environment of change Kobayashi referred to above makes the use of overlapped jobs less of a strain than might be supposed.

The use of CAD data to find problems and do designs prior to making any prototypes obviously saves time and money. However, it appears that this process is most successful during the three prototypes that NTC builds. When the plants start getting involved in prototypes, the use of CAD data wanes because (1) the prototypes are much more compelling, (2) they are easier for people who did not do the

original design to see and understand, and (3) those CAD tools and data are NTC's, not the factory's. This is a typical problem of "ownership!" The only data easily shared by all are the exterior shell descriptions.

Tour of CAE Facility

CAE facilities are used to do exterior styling and many complex analyses, plus actual design engineering. The facilities comprise two Cray supercomputers, two IBM 3090s, two Unisys 2200s plus several more large mainframes for supporting 3D CAD drawings, solid modeling for casting and forging design, and routine data management. The Yokosuka engine design facility has its own Unisys 2200 for designing castings and forgings using solid models. The IBM and Unisys computers support 1,400 graphics terminals. Many of these terminals use SDRC's I-DEAS package. The mainframes also support I-DEAS in IBM's form (CAEDS) but all agree that performance is poor and workstation implementations are better. In addition, there are over 3,000 PCs of various types, including 1,000 Macs. The penetration ratio for PCs is about 0.5 (one PC for two engineers). The computers at NTC are networked worldwide with similar computers in the United Kingdom, the United States, Spain, and Australia. The same software is supported at all sites.

Several demonstrations of CAE were given:

• Very realistic rendering of car exterior shape, including reflection of light off the body. This is a surfaced wireframe model. It becomes the master for all later design.

Stereo display of the car's interior, showing dashboard, seats, and shift handle. This was not too detailed. Using special glasses, one could get a pretty good stereo impression.

engineers design things that only senior engineers could before. One case is design of trunk, brake, and axle systems.

Another nonartificial intelligence application is use of 3D modeling to study how to package parts in pallets and shipping containers and to see how to use last year's jigs and fixtures for this year's parts.

Considerable effort has been made to export this design software system to Nissan's suppliers. Nissan and IBM have a joint venture that supports the software and sells it. A low cost workstation version is also for sale to small suppliers. At many suppliers there are now hundreds of terminals running CADII on both host type computers and engineering workstations. Kobayashi said that Toyota and Unisys have a similar joint venture for selling Toyota's CAD to its suppliers. Problems arise trying to sell to a large supplier like Hitachi, which has its own CAD software and hardware, and to overseas suppliers who have different data standards. U.S. suppliers have a data format standard, for example, that forced Yoshida to write a data translator.

History and Goals of Nissan's CAD Efforts

CAD tool development and user support, including support of business data systems, is provided by 220 people in the R&D Systems Department. In 1985 it had only 70 people.

CAD at Nissan has three main goals and several historical threads:

I asked if noise from engines and transmissions can be analyzed by computer and was told yes.

Other CAE applications include a few attempts to use expert systems to help designers with routine but troublesome problems, or to help junior

2. Design specification control and parts data trees

3. Efficiency of engineering 4. Higher quality

5. Simultaneous engineering