feature libraries are a start in this direction. CAD also permits editing of existing parts, but his goal is to produce a design that is "similar" to one or more existing designs. Since designs involve many parts with interrelated constraints, a similar design might be one with slightly different constraints, or with some parts replaced by new ones while others remain the same. He referred to "experienced product data." By this he probably means data on parts that have stable process plans, assemblies whose tolerances have proven to be satisfactory, and so on.

I asked Okano if he saw analogies between feature-based design and object-oriented databases, or between analogic design and reusable software. To each he answered "yes" immediately. "I am currently reading papers on reusable code," he said. So his thinking is again very similar to our own. (He also said he was keenly aware of Prof. Inoue's new curriculum at Todai.)

It would seem that "top-down design" would be related to "analogic design" if one could come up with a high level description of a product that the designer then filled in with specific geometries and constraints.

Another problem on their agenda is tolerance analysis and synthesis. This is Inoue's project. He has been reading the literature but also has no concrete approach, or else he did not want to reveal it. The group has already rejected tolerance analysis via the worst case method and has opted for a statistical approach. This is the most common choice because the latter is tractable and the results are more economical in manufacturing.

One of the group's first steps was to buy and evaluate a commercial product called VSA, developed by a small U.S. company. This program can do statistical analyses of tolerances by using a Monte Carlo technique. IBM's evaluation is not complete but Okano said in his opinion it was a useful program for designers. However, it has two

drawbacks. First, it deals with dimensional tolerances, such as ± specifications for distances, but does not deal with modern methods called geometric dimensioning and tolerancing (GD&T). The latter is supported by an international standard and attempts to provide ways to describe entire shapes or relations between surfaces, such as flatness, parallelism, concentricity, roundness, and so on, not merely individual distances.

A problem with GD&T, shared by all other methods, is that there does not exist at present a rigorous mathematical description of the various tolerance specifications. Several international standards committees are addressing this problem now but little progress has been made. The TRL people were unaware of this effort.

A shortcoming of VSA, according to Okano, is that it cannot really evaluate tolerances in assemblies. His reason is that VSA assumes surfaces are perfect and that the mating surfaces on two parts are in fact the same surface. He cites tolerance specifications on surface roughness as proof that mating surfaces will not be the same. For precision assemblies such as disk drives, such distinctions can be quite important. He briefly described a situation in which a plant engineer complained that one of his problems was due to insufficient specification of roughness tolerances leading to assemblies that were out of specification.

Regarding the larger issue of design tradeoffs, Mr. Numao suggested that sensitivity analyses might be useful. "Similar to simulations in operations research," he said. I brought up the Taguchi method and asked what they thought about it as a design aid. They did not have strong opinions, indicating that they did not think it was widely used or in demand at IBM.

We had some informal discussion of concurrent engineering (CE). Okano asked about CAD vendors' claims that their product supported CE because

one could design a part and automatically have the computer generate numerical control machining instructions. This is what used to be called CAD/CAM (computer-aided design/ computer-aided machining). He agreed with this assessment. They feel that CE is like groupware in software engineering, that is, many programmers working on the same program. However, they agreed with me that mechanical product design is much more complex. Okano suggested that levels of abstraction and modularity were common themes. This remark, like others, indicates that he is the thinker in the group, seeking to find analogies in far-flung fields.

Micromachines

Repeating last year's visit, they showed me videos of micromachines that the robotics group has been making. These are linear actuators driven by comb devices, which are electrostatic actuators with the stator and motor interleaved like two combs with their teeth pressed together. The comb teeth are typically 4 microns thick. Last October's version is like the original UC Berkeley devices, exploiting mechanical resonance to create motion in the form of vibration at the structural natural frequency (typically 5-7 kHz). No other motion frequency is possible with this design since the force generated by the actuator is too small and needs the amplification provided by resonance. The January 1991 version is able to move deliberately rather than under vibration. The reason is that the gap in the actuator is extremely small, perhaps less than 0.1 micron, permitting larger forces to be generated.

The new design has two interesting features. First, it assembles itself under the action of another actuator built in series with the operating one. The details were hard to see in the video. The assembly motion consists of delicately sliding the moving part of the comb into the

stationary part. Once this is done, the moving part can proceed to move under external control. The other feature is an elastic suspension that greatly improves rotational stiffness over earlier designs. This improvement was necessary since the small operating gap in this design would easily jam if a very slight rotation occurred during actuation.

Last year Hazeki expressed the hope that these actuators would find application in disk drives, perhaps for adjusting heads. However, I was told this year that no applications have been identified.

Final Observations

This group clearly is technically very competent and has a coherent vision about improved CAD, but it does not have much evident drive from within or spur from without to really apply their work to improving IBM's productivity. This is in sharp contrast to both Hitachi and Toyota, where the drive is right on the surface.

SONY

1 July 1991

Background

This visit was arranged by Lynda Strupp, a U.S.-born member of Sony FA (flexible automation) marketing staff. She is bilingual and helps with U.S. marketing. Sony people present were Mr. Yoshihiro Tsukamura, Deputy General Manager, FA Group; Mr. Yunosuke Hayakawa, General Manager, Planning Division FA; Mr. Tohru Fujimori, General Manager, Robotics Products FA; Mr. Hiroyuki Segawa, Engineer; and Mr. Junichi Kuzusako, Assistant Manager, CAD/CAM, Product Technology Group. Tsukamura is the most senior person among these, quite experienced in design methods, although he joined

FA only a year ago.

Ms. Strupp has communicated with me before and stressed that Sony is very protective of its CAD and design methods, as well as of its manufacturing methods for the most advanced and competitive products, such as Handicams. The purpose of this visit was for Sony to determine if there was enough for them to learn from me that they should open up. Apparently this they should open up. Apparently this was a successful visit because they decided to host me for further meetings with Mr. Fujimori and Mr. Kuzusako (see the followup report).

Business of the FA Group

The FA Group makes and sells robots and assembly automation equipment, including circuit board assembly machines, both inside and outside Sony. In both markets it competes with other vendors. There are now several hundred Sony robots inside Sony (the number is not clear, or they did not want to quote a precise number, so perhaps it is as few as 150 or as many as 450), mostly at the Koda plant where VCRS and Handicams are made. In the United States they have had a hard time selling robots or complete systems. This has been true for as long as the FA division has existed, which is about 6 years. Their best U.S. customer bought 24 robots in a complete "turn-key" system about 2 years ago and recently bought about 40 more. This customer will make all the tooling and do all the programming and system integration for the new 40 itself, a major achievement.

Ms. Strupp says that Sony's biggest competitors are Seiko and ADEPT. Both have better programming systems, which attracts customers. This annoys Sony since the best controller is often not needed. ("Only Motorola wants the best of everything.") Sony often has better accuracy or payload and has much more experience with product redesign for robot assembly and system integration. They want more of this kind of business but can't get it, often because they are short of people. This is in contrast to most robot makers who are also system integrators, since it is usually more profitable to sell naked robots. than to do the engineering necessary to tailor a system to a customer's product.

Sony will even sell robots and systems to direct competitors such as Hitachi. I asked if this might result in the transfer of Sony's product design for assembly know-how and was told that Sony sells only naked robots to such customers since they already know how to design the tools and do the programming and system integration. In this sense, Hitachi or other Japanese customers are typical, and European companies also commonly have such capabilities. U.S. companies normally do not, and this fact makes the abovementioned customer's capability especially unusual.

The Sony people were surprised to learn that Nippondenso has 2,600 robots (as of 1990) of which they made 90% in-house. Nippondenso does not bother to try selling robots outside and does not buy many either, since they get better service from their own engineers and in any case domestic robot manufacturers are saturated with orders and cannot deliver fast enough.

Sony regards itself as especially skilled in mechanical design, less so in electronics and software. This is important because mechanical design practices and the usefulness of new CAD

tools were the focus of the visit. It is Sony's impression that they do not need help in mechanical CAD but do in electronics. Yet when I showed them our feature-based design software video they asked immediately whether it would be available commercially and when.

According to Fujimori, Sony's mechanical designers consider assembly right during design, including assigning tolerances and considering assembly sequence. In fact, sequence is determined first and then tolerances are decided, which is the correct method. "Good designers consider both the sequence and the type of equipment that will be used. They start doing this right from the beginning of design, when there are no parts and you must imagine the final assembly." (See the followup report where this statement was made clearer.)

When I asked if redesign was needed for robot assembly, I was told yes. This seemed paradoxical in view of the earlier statement. When pressed for examples, they cited the need to add chamfers around holes if the tolerances could not be tightened. This is an especially trivial change to make and indeed to include in the original design. Thus the discussion did not seem productive and I could not get a better definition of the situation, except for the remark that not every designer has the skills of the best one. I also could not get them to agree that software to help calculate tolerance stackups would be useful.

The latter comment recurs in many of my visits: the hosts are proud of the CAD and CAE software they have and know how useful it is. But they often see no need to have anything better even when they agree that doing some design task manually is either tedious or is even skipped due to its difficulty.

Hayakawa noted that in 1986 top management launched a campaign called "innovate 86" whose aim was (and still is) to improve product design, industrial engineering, and automation. When

Products are designed by teams of engineers, typically 20 to a team. The product design cycle comprises four prototypes called Research, Function, Manufacturing, and Preproduction. These appear on about 4- to 6-month centers. Cost and manufacturability are considered after the functional prototype achieves the required performance. Manufacturing engineers and FAengineers join the design project at that point. A new product will take 2 years to pass through this cycle, whereas a modification of an existing product can be accomplished in 1 year by a team of five engineers. Projects are run by an engineer with 10 years of experience. He may run more than one project at a He may run more than one project at a time but not usually. Previous designs are scrutinized carefully to determine applicable tolerances and assembly methods.

An example given was the optical pickup for a compact disc (CD) player, where tolerances are in the microns. Several people from different disciplines join a discussion about how to best achieve the specifications. Sometimes an adjustment method is used rather than aim for a perfect result in the first place. At various points in the design process there are design reviews, a typical practice in many industries. At Sony it is common for engineers from other projects to attend and make suggestions for improvement or warn of possible problems with the current design.

Sony buys most of its CAD software. The commonly used packages are CATIA and CADAM, both supported by IBM. Sony also has an inhouse freeform surface program called FRESDAM that creates very realistic views of the exterior of a camera, say, using a Silicon Graphics terminal. Recently they acquired a robot motion simulation program from a U.S. company, Silma. However, the database of part designs and tolerances, plus information about robot tolerances, does not exist so full advantage of this software cannot be had. “Japan is weak in robot simulation software."

They do not have any solid modeling software. Instead they model exterior surfaces in FRESDAM with Bezier's formulation. No analysis can be performed in this software. In a few cases, the surfaces are transferred to SDRC's solid modeler, from which a plastic mold flow analysis is often done. FRESDAM has recently been linked with a stereolithography system.

What do they want in the future? Kuzusako says that feature-based design (FBD) sounds like a good idea but solid modelers are too hard for designers to interface with right now. (Many hosts said this.) Kuzusako also said that FBD would be the most help if it were connected to information databases such as for materials properties and costs. Designers should be able to create their own features. This would be useful for robot programming as well as for design.

Currently they have no organized system for keeping track of previous designs in software or of reusing them. Kuzusako felt (unless he misunderstood the question) that it is more important for all the designers of the current product to be able to access a common database about information technology (IT). This is logical if one realizes that many Sony products are

quite different from their previous versions, due to extreme size reductions or performance changes.

What about the need for or usefulness of software to calculate tolerances or make fabrication and assembly cost analyses? Kuzusako allows that this might be useful but he has no plans to start in this direction, only a hope. He has no experience in this area.

Tsukamura, the senior man and most experienced, said that his top priority, based on inputs from the designers, would be for software that accomplishes or aids end-to-end design of circuit boards and their manufacturing processes. Such software exists commercially, I said. Sony has a mix of homegrown and bought, and they are not satisfied with it. His judgment was based in part on the valid observation that the opportunity for using computers is greatest where design is the most routine or uses (or reuses) the most standardized shapes. This characterizes electronics design, not mechanical design.

HITACHI SAWA WORKS
AND TAGA WORKS
3 July 1991

Mr. Takahashi accompanied me to Hitachi's Sawa and Taga Works. Sawa makes automotive components, and Taga makes a variety of consumer products.

Background of Sawa Works

At Sawa, our host was Mr. Sato, responsible for improving Sawa's CAD and CAE capability. He came there 5 years ago from Hitachi's Mechanical Engineering Laboratory, a more research-oriented facility. At that time, CAD at Sawa was little more than computerized drafting of ordinary machine drawings. He has had something of a hard sell and has written at

least one of the new engineering applications himself, a program that does vibration analyses on rigid assemblies. It works by linking models of individual shapes such as bars and plates, for which individual vibration models have been worked out. The assemblies are joined observing the boundary conditions, so a valid model results. Uses include studying the effect of vibration on solder joints between circuit elements and circuit boards. This is a fairly sophisticated approach and indicates that Sato is a good analyst who has the horsepower, if not the manpower, to change how Sawa operates.

Sawa's business comprises electrical and fuel components such as alternators, generators, engine controllers, pressure and flow sensors, microcomputer controlled carburetors, turbochargers, distributors, brake system controls, fuel delivery systems (injectors and manifolds), air conditioning systems, and so on. The plant has annual sales of ¥200B as of 1986 ($1.5B), 2,700 employees, and 100,000 m2 of space. Major customers are Nissan, Fuji, Mazda, Suzuki, Honda, Ford, Chrysler, and Audi. Competitors include DELCO and the powerful Nippondenso.

Their introductory video heavily emphasized reliability, testing, reliability, design, reliability, and so on. Apparently this video is shown to prospective customers. It showed many robots and automated machines, including a line in a clean room for making injectors, a robot assembly line for alternators, and typical automatic production of circuit boards.

Another video of CAE showed use of supercomputers to evaluate flow out of injectors into fuel manifolds, including studies of the effect of fuel particle size (100 microns is too big) as well as supercomputer simulation of air flow from air conditioner vents into the car. Also shown was a simulation of use of active suppression of air conditioner noise using extra loudspeakers. A last supercomputer application was a

dynamic analysis of stresses on a Nissan engine ring gear caused by meshing of Hitachi's starter pinion. Sato stressed the importance of being able to access Nissan's CAD data in order to do this analysis.

So this factory makes heavily engineered products to exacting specifications under strong price and quality competition from other vendors in a world-wide market. They are still in the process of building a strong CAD/CAM/ CAE capability under Sato's direction. An important subissue is computer data communication between Hitachi and its suppliers. (See a later report on Nissan's attitude and methods for communicating with its suppliers.)

Product Development Methodology

Sawa has some luxury in being able to work to the development pace of car companies. This gives Sawa much more than that available to the Image and Media Systems Laboratory people developing video cameras. Also, it is typical that 20 engineers might work on one item such as an alternator. Since alternators are far less complex than video cameras and change much less from model to model, this means that Sawa has effectively lots more engineers per project.

When I remarked to Takahashi later that Sawa should send some engineers to I&MSL, he said that although no exchanges have occurred between Sawa and Yokohama, people from the slow life often cannot adjust to the pell-mell environment of designing fast-paced consumer products. Slow cycle automotive people are used to extreme reliability requirements, to which they respond with lots of analysis, many experiments, many prototypes, and heavy reuse of prior designs.

More importantly, the 4-year pace has given Sato the chance to think ahead about his future needs for CAD/CAM/ CAE.

The product development process consists of a series of prototypes and accompanying analyses developed in response to a specification from a customer. The process was illustrated with their current effort on a new small alternator for Nissan. The original request for proposal (RFQ) came in 1989 for an anticipated car launch in the 1995 model year (fall 1994). The specification included target power, size, weight, and cost. The specification was negotiated and probably adjusted in monthly meetings for a year while Hitachi did lots of computer studies based on varying existing factors and technologies. Later this year Sawa must deliver primary samples which Nissan will road and lab test for performance, reliability, and noise. Responses from other vendors will be tested during this period, and a decision will be made a year from now. If it wins a contract, Hitachi will then have over 2 years to develop final designs and process plans, design and facilitize the factory, and start up production. During this time, Nissan will finalize its engine design, perhaps altering the mounting conditions for the alternator and requiring additional noise and vibration studies.

Use of Computers in the Design Process

The engineering analyses supported by Hitachi software presently include:

FEM to study vibration of cast housings and circuit boards

magnetic field analyses of rotor, stator, gap, materials, windings, etc.

⚫ a spreadsheet for quickly doing the basic electrical calculations such as power-speed-voltage tradeoff curves (an example of what Sato calls a

"handy program") a nice color graphics interface and graphic output, including documentation of all the engineer's calculations, like an engineer's notebook

⚫ sensitivity analysis software to determine the effect of varying certain parameters in the hope of obtaining a less sensitive design or of finding a way to improve performance by changing parameters

In the last program, the designer chooses the parameter values, and the computer creates a consistent set of drawings, except where the combination chosen could not be resolved by the computer. Such areas are left blank and the designer draws them in using typical 2D CAD methods. In this way, routine products can be put out very quickly, as long as they do not challenge the state of the art and require real design engineering. Such programs have been written for alternators,

rotor dynamics to predict bending starters, and distributors. They are not and vibration

⚫ various commercial FEM programs, two mold flow programs, Hitachi's in-house CADAS CAD software, and occasional use of the assembleability evaluation software from Hitachi PERL

On a tour of the CAD laboratory I saw several displays of simulation outputs plus demos of several of the above capabilities. Others were

⚫ fuel particle flow in several vendors' designs of manifolds

⚫ analysis of impact of injector valve on valve seat

starter motor magnetic flux

linked to any of the CAE software so they are good only for deploying existing proven designs. Thus it is really a documentation management program rather than a design program, but it hints at what might be done in the future. In particular, it contains no artificial intelligence (AI), no featurebased representations, and no constraint-based descriptions. Application of these methods would permit substantial design variations to be accommodated without leaving blanks.

With the exception of a few Silicon Graphics and HP9000 workstations, these demos ran on Hitachi workstations or PCs and often utilized output from three Hitachi mainframes elsewhere in the building. UNIX appears to be the operating system (OS) of

choice.

Much of the effort to computerize

airflow over a hot wire anemometer design is driven by the need to make all

⚫ sensitivity and modal analysis of alternator end casting vibration (display on workstation of mainframe calculations done earlier)

parametric creation of new “designs" for alternators by combining menu choices of 18 parameter values (3 shaft diameters, 5 lengths, 3 outer shell diameters, and so on)

components smaller and lighter, with the ultimate aim being the car-maker's need to meet the 1995 CAFE standards. To make things lighter, one chooses lighter (often relatively weaker) materials and thinner sections. These, in turn, are subject to vibration which, in turn, causes noise or, worse, structural failure. Low noise is a major Japanese automotive design goal and competitive feature, so the goals of low