weight and low noise naturally conflict. Identifying and resolving design conflicts is a major challenge for both designers and researchers. Here the response is to build up over the years an extensive analytical capability and drive it night and day to gradually improve and refine the designs. Trial and error on the computer seems to be the main modus operandi.

An interesting point is that a bench test is different from a test in a car, since benches are rigid compared to cars, which have resonant vibration frequencies of their own. Thus the software has been written to simulate either environment so that test data can be compared with computer predictions. One of their test setups is a GENRAD vibration stand, analyzer, and display linked to the computer.

Sato categorized his emerging software tool set as follows:

Design

Stage

Conceptual

Product

Production

Tools Desired

Easier communication of files and data between Hitachi and both its customers & suppliers: includes geometry models & CAD data, plus test data, analysis results, etc.

Sensitivity analyses, parameter manipulation, expert system for effect of tolerances on performance & reliability. Structural optimization for both shape and dimensions.

Better assembly analysis tools? No clear wish here.

No clear wish here.

Visit to Taga Works

In the afternoon we travelled to the Taga Works. There the host was Dr. Murakami, who came to Taga from PERL several years ago with the same mission as Sato has at Sawa.

Background of Taga Works

The Taga Works makes a wide variety of consumer products (50% by yen sales) such as washers, vacuums, and fans; office equipment (30%) like word processors, laptops, and laser printers; and industrial labor saving devices (20%) such as electric motors, hoists, and factory ink-jet printers. Their introfactory ink-jet printers. Their introductory video tied this assemblage together by the fact that all (laptops,

too?) contain electric motors. The plant covers 400,000 m2, has [only!] 2,500 employees, and turns over $960M per year.

Product Development Illustrated with Fuzzy Control Washer

A recent washing machine was used as the example product for illustrating use of CAD/CAM in design. This is a "fuzzy controlled" machine with one button: START. Using sensors, it determines the amount of clothes, the average weight of material in the clothes, and the degree of completion of washing. It adjusts the wash, rinse, or spin time to suit. It has been on the market for about a year and costs ¥119,000, or $881 at ¥136/S. (A Big Mac is Y380, or $2.80, and a good electric iron--without any fuzzy control aspirations--can cost anywhere from ¥9,000 to ¥28,000, or $66 to $207.)

This washer took 3 years to develop, involving 10 to 15 engineers at a time, including all the test and inspection teams. Three prototypes were made during this period. The most difficult design challenges were the motor controller and the shape of the agitator. At least 15 agitator shapes were thought up and tried, being machined out of solid blocks of acrylonitrile butadiene styrene (ABS).

The marketing objective was to produce a user-friendly smart appliance having the ability to run and diagnose itself. It has a timer like microwave ovens and coffee makers so it can be set to run at night when everyone is asleep and be ready in the morning for drying to be done. Since Japanese apartments and houses are very small, it is likely that someone sleeps near the washer, so it must be very quiet. It also must be small and light, so all the parts must be light.

All the noisy solenoid valves have been replaced by quiet motor-driven ones, a quiet balancing system was designed, and gear noise was removed by reversing the motor to create the agitation. They do not know if this will cause motor failures in future years. This must be the reason why motor controller design was difficult. No special materials or sound deadening paints were used. No software for predicting noise was used either. However, like Japanese cars, this washer aims at the trend toward quieter and friendlier products, and much of the CAE used was directed at achieving these goals.

Clothes quantity and weight are measured by filling the tub with a known amount of water and spinning up, then letting the motor coast and measuring the decay time. This is done twice with different water levels to determine quantity and weight separately. A conductivity sensor measures soap concentration and dirt quantity in the water. Wash and rinse times and their water amounts are adjusted accordingly.

While they call this fuzzy control, I think it is in fact a set of lookup tables. Most fuzzy logic applications I am aware of use linear interpolation between preset responses for full set membership to generate a graded response for partial set membership. [For example, "if clothes are heavy weight, use full water level" and "if clothes are medium weight, use half water level" would generate three-quarter water level for clothes that measure halfway between heavy and medium.] So while fuzzy control is a great marketing idea and the machine is truly useful, the technology development is in the sensing and in converting sensory readings into good estimates of clothes weight, not in design of the controller itself.

Final assembly of this washer is entirely manual, but motors and other drive train parts like clutches are built up by robots. The cabinet and tub are

made from sheet metal cut and bent automatically. Controls, lid, and agitator are injection molded plastic. Since both outer cabinets and drive trains are redesigned infrequently, automation can be applied to their fabrication and assembly. But control surfaces and lids change as fast as twice per year. This causes a strain on Taga's design talent. Marketing creates the pressure. Takahashi agrees that these redesigns are superficial but he adds that a shortage of engineers keeps Taga from undertaking the deeper redesigns that they would like to pursue.

Use of CAD/CAM/CAE

Only recently has CAD/CAM/CAE come to Taga. In this sense, the Image and Media Systems Laboratory is well ahead of Sawa which is, in turn, well ahead of Taga. They have had Hitachi's CADAS for several years but only recently got Moldflow for plastic molding analysis and ADAMS for kinematic analysis, both linked to SDRC's I-DEAS solid modeler. Their first attempt recently to transfer CAD data to the machine shop did not work very well. The factory's software development center set all this up for them.

[Prof. Kimura told me later that Hitachi has a System or Software Development Laboratory that is responsible for finding and trying CAD products and getting the factories to use them.]

So the fast response to the market has been accomplished "by our sweat and tears." Right now it is faster to redesign by hand than to create a solid model and run Moldflow. Usually only small changes occur so their experienced designers know what to do to design a new mold. Presumably a featurebased approach would speed things up, since many of the plastic parts consist of regular flat surfaces with rims around the edges and bosses underneath to take screws. Expressing this in terms of features would not be difficult.

They do not have any software to analyze water flow patterns created by different agitator designs. Energy transfer to the clothes is more important than water flow in this kind of machine anyway, but still there is no software.

The CAE Laboratory contains numerous Hitachi workstations operating under UNIX. There are also several Silicon Graphics terminals. In the laboratory they showed me use of Moldflow plus a verification part that agreed fairly well with the simulation. Another demo concerned using ADAMS to predict how the tub would sway for particular unbalance loads. This information can be used to size the clearance between the tub and the outer housing. The FEM program ADINA is used to test the strength of a crimped joint in the tank, based on loads caused by unbalance. Another use for ADAMS is to predict how a vacuum cleaner will track when pulled by the hose, based on various caster designs and locations.

The CAD office also contains Hitachi workstations and UNIX. Drafting is the main activity. Some computers are linked directly to the factory floor or to the purchasing department, to which drawings are sent defining the specifications for in-house and purchased parts. "Engineers have a lot of power over purchasing decisions." Altogether they have many workstations and have invested a lot in computers and LANs.

Another interesting program is a parametric design system for hoists. When an order is received, the engineer types in the specifications, such as capacity, lifting speed, and so on. The computer searches the database of past designs and finds the one(s) whose individual specifications match exactly or come the closest. Some items in the specification may be missing completely so they are left blank. The designer then retrieves the useful drawings and makes the missing ones. Mr. Sagawa, representative of the software service

department, said it took them a year to write this program. It has been in operation for 6 months.

This is a version of what GE calls "purchase order engineering." It is called variant design in the academic world. GE told me that 85% of their engineers do this kind of work and it has been largely automated on workstations. I saw a more sophisticated version at Cooper Industries where high capacity industrial air compressors are made. The compressor vanes are the most complex to design, so existing designs are used to the extent possible. A similar database exists with 10,000 prior designs in it, and the three nearest are found by least squares optimization of the parameter matches.

Discussion

At this and several other plants I have visited, there are a few female engineers and "draftsmen." In view of the rising shortage of engineers that every company is feeling, this trend toward women engineers is expected to increase.

especially disavows the importance of computer aids in relation to management and work methods, ascribing the "Japanese advantage" to precisely the points made by Prof. Fujimoto.

While it is true that engineers at most companies I visited work hard and stay late, I think it is disingenuous to say that computer aids have had little effect. As far as I can tell, every company I have visited has bought the best and latest U.S. hardware and software and applied it intensively. Rates of penetration differ, and some companies use their own hardware (Hitachi) or software (Nissan), but the trend is the same. Still, CAD/CAM/CAE cannot do enough of the job to relieve the pressure on the engineers. They are now so tired out that some companies are openly discussing slowing down the pace of new designs (see Hitachi Construction Machine Co. report).

DISCUSSIONS WITH
PROF. KIMURA AT HIS CAD
RESEARCH LABORATORY
ABOUT PRODUCT
REALIZATION, IMS, AND
PRODUCT DEVELOPMENT
CYCLES

5 and 19 July 1991

It appears that the shortage is causing many problems. A major one is that companies can no longer make the design changes they want to, or cannot put them into effect as fast as they would like. Yet rapidity of model changes Background is the main competitive weapon between Japanese companies and is the biggest advantage they feel they have over international competitors. This must mean that every Japanese company has the same problems, but Hitachi feels that some of its competitors, certainly Sony, can do things faster. They feel that Sony may have some "secret weapons" such as better management methods or better computer tools. They are frankly puzzled and worried.

It is interesting to compare this reaction to Sony's, which is that the engineers just work hard and marshall their experience effectively. At Nissan I was told the same thing. Nissan

The general topic of research in the laboratory is product realization via laboratory is product realization via computer. The work is currently sponsored by the Japan Society of Precision Engineers, of which Prof. Kimura is a prominent member. Similar research has been going on for at least 5 years under the original leadership of Prof. Sata (now retired) and Prof. Yoshikawa, now Vice President of Todai. Kimura said that he expects a new large government grant starting this fall.

Sata and Yoshikawa apparently are responsible for the highest level view of this topic. It has matured from longstanding research by them and Kimura

on numerical control (NC) machining of complex surfaces into the notion of CAD systems that contain engineering knowledge, capture the designer's intent, and connect design and manufacturing. Its most recent manifestation is in Yoshikawa's proposal for internationally supported research on intelligent manufacturing systems (IMS).

Note: On the second visit I was accompanied by two researchers from Hitachi Production Engineering Research Laboratory.

New Approach to Manufacturing

"Manufacturing" often connotes fabrication of single parts, typically by metal cutting. Interpreting IMS in this way seriously misses the point. Taking the more general view that “manufacturing" means fabrication, assembly, test, and logistics activities in factories also seriously misses the point. It is important to realize that the idea of IMS extends well beyond "manufacturing" all the way back to conceptual design.

Therefore, IMS is the same as "product realization," which encompasses all of the processes, intellectual and physical, that are undertaken as an idea for a product is hatched, refined, turned into design data, analyzed and simulated, cost-analyzed, turned into process plans and instructions, the parts made, assembled, tested, and shipped, the factory designed and operated, customer responses factored into the next design, and so on.

Kimura's career illustrates the process by which these ideas have grown and matured. His background is in 3D modeling of complex shapes such as car bodies and creation of computer software for representing these shapes. He was involved in several MITI projects that created GEOMAP I in the late 1970s and GEOMAP II in the early 1980s. This software was taken up by Toyota and used in their car design

Such software ultimately must represent two distinct but mutually indispensable components: (1) the mathematics of the surfaces themselves, so that they can be displayed realistically and manipulated into the desired shape; and (2) the physical processes by which metal will be pressed and stamped into those shapes, so that the designer can find out if the shape is possible to make and how much it will cost. A complete design system must therefore contain a great deal of manufacturing process knowledge. Developing this knowledge has been a 20-year effort for most automobile companies but is only being integrated into design software in the last 5 years or so at the most advanced companies. On the other hand, software to represent the shapes mathematically has existed in various forms and levels of sophistication since the 1950s.

The above pattern (namely, that computer-aided design software is incomplete without manufacturing/ assembly/test process knowledge) is at the heart of a new wave of research and commercial progress in CAD. It is redefining the meaning of CAD, once embodying only making mechanical drawings of single parts, into someday representing all the knowledge needed to design, analyze, make, and even sell entire products.

State of the Art in Different Design Domains

The use of computers in design has grown rapidly in the last 30 years, but the most complete representation of product and process design methodologies is in the domain of microelectronics. Here, the products are so complex that they are impossible to design without pervasive use of computers. A typical microprocessor design looks like the map of a city's streets.

mation as the entire New York metropolitan area including much of western Connecticut and northern New Jersey. Because microelectronic designs consist of layers of 2D information, mostly in the form of circles and straight lines, they are easy to represent mathematically. Similarly, all the manufacturing tooling consists of photographic replicas of exactly this information. Thus the conversion of the design into manufacturing equipment is relatively straightforward in the sense that no further complex mathematical processing is required to convert the design data.

By contrast, the shapes of car bodies cannot be made by making the press die the same shape as the mathematical model because of the mechanical properties of the sheet metal (springback) and the behavior of the metal-die interface during stamping. Accurate models for these factors are difficult to create. Bridging this design-process gap is the essential obstacle in car body design-manufacturing. The trial and error involved here is responsible for much of the time needed to design cars, and thus better prediction and less error are of extreme competitive value as well as being an extreme intellectual well as being an extreme intellectual challenge.

The conversion of microelectronic designs into working manufacturing processes is not straightforward either, but the obstacles are in keeping the processes free of contamination and in learning how to make the patterns of lines and circles smaller and smaller. The limits of the manufacturing processes create "design rules" that must be followed, but these are relatively simple to state and to check for during design. This is not to say that making microelectronic products is easy but rather that representing the designs in the computer in ways that translate directly into manufacturing instructions is relatively straightforward.

In addition, the geometric layout of a microcircuit determines its electrical behavior, so it is relatively straightforward to obtain a computer simulation of the behavior during design.

On the other hand, drawings and mathematical shape representations of mechanical items contain no clue at all as to how they will operate and little about how they should be made, except when their shapes are simple circles, cylinders, and so on. On the contrary, most mechanical drawings and CAD versions of them are merely symbolic notation that refers to both physical things (circles that refer to circular holes) and nonphysical things (arrows pointing to extensions of surfaces that represent measurements between those surfaces). Drawings thus represent a well-developed language that people know how to interpret. Many statements in this language are inferred and do not appear directly on the drawing.

Another difference is that in electronics most of the parts are standardized. In larger parts, the externals are designed almost exclusively for handling. In smaller items like microcircuit elements, the shape completely determines the performance, but these shapes are simple and are also standardized. The behavior of each item has been well modeled and the behavior of both single elements and large systems of them can be predicted with excellent accuracy.

By contrast, the shape of mechanical items must be tailored specifically for each item to the tasks it must perform. Rarely does a part do one thing. Most items do several, such as provide strength, electrical or heat conduction, and geometric arrangement. All of these functions can be difficult to model and predict accurately even when performed in isolation. Little modeling capability exists for sets of mechanical items operating together as a system except if many simplifying assumptions are made.

Thus the essential elements of a complete design-manufacturing system

are in many ways much easier to estab

lish for microelectronic items than for mechanical items.

Research Activities at Todai

Kimura has two bright assistants, Prof. Suzuki and Prof. Inue. In other laboratories there are additional faculty with similar aims. Kimura and his assistants have 12 graduate students, 8 undergraduates, and 13 visiting researchers from industry. Their research breaks down into three main categories: geometric modeling, product modeling, and design/manufacturing. Their participation in the IMS is in such areas as concurrent engineering, virtual manufacturing (meaning wide use of simulation of product and processes), design by customers, information transparency (meaning good user interfaces) and self-organizing and distributed systems.

Kimura also divides up the topics into objects, processes, and environments. All three are essential, he says. In this sense, he reflects his machining and NC background. "There is nothing without understanding of the basic processes," he says, and I agree. Yet right now their research tools consist entirely of workstations, due to lack of space, funds, and a process-oriented colleague to replace Sata.

Geometric Modeling. The laboratory has a long history of studying geometric modeling. Recently they have been overtaken by developments in commercial software and by the fact that students do not like software thesis topics. So he is using commercial solid modelers as the basis for additional work. This work is heavily related to constraint-based modeling and design. A typical project is on use of constraints to determine the size or shape of an item based on a statement of its performance requirements. Feature-based design is another related topic.

A typical project involves allowing the designer to call forth standardized elements such as shaft ends and their mating bearings, which together form a functional feature made of several geometric features. These functional features can be given specific dimensions, or some larger requirement such as a torque or side load can be given by the designer and the computer will resize the parts. The example shown was hardwired (i.e., preprogrammed in LISP) by the student. There is no graphic design interface.

Another project is design of sheet metal parts based on partial input of geometric constraints. These include locations of holes, portions of the part's outline, and areas the part should avoid. The computer draws the rest of the outline. This project is similar to others where the spirit is to permit partial information to be used effectively.

Another series of projects deals with tolerances. These include use of solid modeling methods to predict all the possible mating arrangements of parts with deformed mating features, such as machine tool ways or peg-hole mates. These methods are enumerative rather than statistical so they are threatened with combinatoric explosion.

Another series of projects involves specification of geometric constraints. The portion that accepts 2D constraints has probably been overtaken by commercial modelers such as SDRC's I-DEAS Level 6. Three-dimensional constraints are much harder to deal with and include lining up features on different parts in order to describe constraints in an assembly.

Product Modeling. Several of the above projects could fall into product modeling, since the line is rightfully indistinct. The goal in product modeling, however, is to work top-down from specifications toward assemblies and finally to parts, with the computer progressively doing more and more of

the routine work. Kimura wants to capture what he calls "product structure," something he distinguishes from assembly structure or machining structure, which have their own geometric features.

Product structure is about function and its relation to technological knowledge: kinematics, dynamics, dimensions, and tolerances. In one form of this idea, a product model is realized as a set of objects (in the database sense) together with object models of the tools that make them. To these models must be linked models of the processes (machining, assembly, test) that will make the parts and the product. He agrees that "this is a quite difficult topic."

Clearly, AI has deeply penetrated this laboratory. One can see it in the magazines the group subscribes to. They cover robotics, graphics, CAD, AI, graph theory, UNIX, and a wide variety of nonmechanical engineering topics.

He is also interested in modeling the design process. He does not approach this directly by observing the behavior of designers but rather by observing the work they do. He was inspired by how Toyota did this. It is very pragmatic and seeks to identify what takes the designer's time that could be computed readily, or what tools the designer uses (colors, multiple views of solid examples) that the computer could reproduce. Right now he has identified "routine design" as an approachable topic. An example is the featureconstraint-driven model of shafts and bearings mentioned above.

He notes that such models are easy to construct so as to obey physical laws. The same is not true of more innovative design. This is not a linear process dominated by well-trod paths and procedures. Here he suggests that the computer models not be bothered by physical reality, as indeed perhaps the designer's thoughts are not either. When something useful emerges, the designer can impose reality. The likelihood of