of several laboratories and works to form his team. Hitachi is quick to claim that this method is superior to the Boothroyd and Dewhurst method for Design for Producibility Evaluation Method Assembly (DFA) that is widely used in The Producibility Evaluation Method (PEM) is an outgrowth of the Assembleability Evaluation Method (AEM) dating from 1976 combined with the Machining Productivity Evaluation Method (MEM) developed in 1985. Since AEM and MEM can have conflicting recommendations about the design of parts, the goal of PEM is to blend the two and give a composite evaluation. These evaluations are done on a part-by-part basis and are intended to give a designer a way to evaluate his own design. The method is designed to be simple and easy to use with minimal training. Thus it demands very little of the engineer. The evaluation proceeds by means of a checklist in which the designer deducts points from each part's score based on various undesirable features. For example, one might deduct 5 points if the part must be twisted while being inserted; another 10 points might be deducted if there is almost no space for fingers or a tool around the part during insertion. A perfect part gets 100 points and a part with a score of less than about 70 is a definite candidate for improvement. The procedure for a group of parts in an assembly is to attack the lowest scoring parts first, then the higher ones, until no further improvement can be made or the average reaches about 80. Hitachi's main claim for the AEM and MEM is that they have validated some cost reduction ratio predictions that accompany the checklists. These predictions are said to be valid within +15% on individual part assembly cost reduction and ±10% on product assembly cost reduction. Details of the method are hard to come by because Hitachi sells it and reveals little to those who do not buy. the United States. The difference seems to be that Hitachi's cost reduction predictions have been validated in an industrial setting whereas Boothroyd and Dewhurst's (according to Hitachi) have not. Otherwise the methods are quite similar. How Will Hitachi Drastically Improve Its Product Design Cycle? Presently, Hitachi's I&MSL depends on commercial CAD and PERL's PEM for its main design aids, plus a lot of very hard work by skilled designers. From these two groups I did not get a feeling that long term design tool development is underway. They are quite concerned that just buying CAD software and hardware from outside will not give them a competitive advantage. The one step they are planning to take, or hope to take, is to combine CAD and the PEM, but as stated above they do not have a clear methodology for doing this. Furthermore, they do not see Japanese research institutions as having anything to contribute to such problems. For security reasons, they were reluctant to discuss their progress in this area. It was interesting to compare this attitude to that of a researcher familiar with Hitachi's VLSI computer support. This software does what most such systems do, namely, supports all aspects of VLSI design from circuit simulation to manufacturing. Naturally, I asked what competitive advantage it gives Hitachi and the answer was "Hitachi's extensive database." Such data include materials behavior, design rules for line width, methods for calculating capacitance, and so on. Thus a lot of engineering expertise and company experience and standards have been captured and is available to other designers. While no definitive competitive comparisons can be made, it is clear that Hitachi has been able to put a great deal of its own work into this system, differentiating it from anything commercially available. What is the analog of such a system in design of mechatronic items? Apparently Hitachi has no vision of such a system. However, several research groups in the United States, Europe, and Japan are investigating such problems under the covering name of integrated product models. A common theme is "Feature-Based Design." This approach seeks to extend the idea of using a computer to capture the shape of parts so that nonshape information is captured and stored along with the shape. Such information includes materials properties, tolerances, assembly approach directions, manufacturing process plans, process costs, and so on. Typical features might be threaded holes, slots, ribs, round passageways, and so on. Since many of these have obvious functional attributes as well as manufacturing and assembly aspects, one can imagine building up enough information to support a physical simulation of the design. At this time, there is no agreedupon approach to constructing such design systems. A number of barriers exist. One is a lack of understanding about how designers like to work and express their ideas. Another is a lack of mathematical models of physical behavior of sufficient accuracy to capture the behavior of a whole product. A third is how to achieve unity in a model of a part or product by building up from many individual features. It is not clear how a concept of the whole item can be achieved based on many little pieces or if this is necessary in every case. Other important engineering and computational barriers exist. Foremost among these is the need to represent the stochastic nature of manufactured items. While the design describes the perfect item, each real part differs in many ways from the ideal. This fact is impossible to overcome. Instead it is acknowledged that reality can be approached more nearly with increasing cost, and that a point is reached where cost overwhelms the effort or adequate performance can be had without additional improvement. Knowing where this point is constitutes a major challenge in every design activity, and virtually no computer support exists for it. It is called the tolerancing problem and is "solved" in every company by using experience, company guidelines, and prototypes. In many cases, tolerances are set based on the best the factory can be expected to do. of the problem or an alternative list of manufacturing). His view is entirely that of the design engineer focussing on performance of the product. He is less aware of design for producibility issues and does not know Prof. Fujimoto. We had a wide-ranging discussion of many issues: what the elements of nextgeneration CAD should be, how to effect technology transfer from R&D to practicing engineers, and what characterizes design problems in general. It is important to note (see the report on my visit to Prof. Kimura at Todai) that in spite of any shortcomings in Kimura's Current Research either Hitachi's CAD or view of future design tools, they are able to design remarkable products just the same. Would "better" design tools make them supermen or just get in their way? Also, Hitachi is not alone in having goals but few methods for achieving them. Many other companies seem to be in the same fix. Only a few have detailed plans and a clear statement of the blockages. DISCUSSION ABOUT DESIGN 13 June 1991 The only systematic approach to this problem that I know of is called the Taguchi method in the United States, or statistical quality control and statistical process control in Japan. Usually this involves making a series of controlled experiments to determine the most important variables, typically utilizing ANOVA, and then focussing design efforts on these variables. The approach is to choose values for the most sensitive parameters in order to minimize the effect that these variables Background will have on performance. The Taguchi method is most easily applied to tuning a manufacturing process, since the experimental series can be accomplished by varying process parameters. To apply it to design requires making many prototypes or manipulating a mathematical model of the design. Both are difficult to do: prototypes are expensive and take a long time to create; accurate comprehensive math models usually do not exist. Among the people I talked to at Hitachi I could find no one who had formulated these long term problems in terms similar to those used above. I did not even hear such blockages listed and identified as ones whose solution was fundamental to drastically improving the product design cycle. Finally, I did not hear an alternative description I visited Prof. Fumihiko Kimura to plan my visits to companies and to discuss research problems in design. Kimura is an expert on the mathematical models of complex 3D surfaces. He did his Ph.D. with Prof. Hosaka who, along with Coons in the United States and Bezier in France, is responsible for all math models of surfaces used in CAD (computer-aided design) of automobiles and airplanes world-wide. Kimura led two national big projects through the Ministry of International Trade and Industry (MITI) on development of surface modeling software for CAD. Toyota, among others, is a major user of this software. Kimura is a respected thinker about design research, product modeling, CAD, and CIM (computer integrated He has just launched an industryfunded project on product modeling to support next generation CAD. He admits that he has no experience as a designer himself and it is interesting to hear him bemoan the complexity of design as it is described by engineers. As an academic he is looking for clean categories and well-defined steps in the design process. What he finds instead are intuitive leaps, a mixture of primary and secondary issues being considered at once, and unlikely linkages between causes and effects that “good designers" seem to know but are not visible to the uninitiated. The problem of technology transfer is especially perplexing. He wants to define next generation CAD but can't describe it to engineers who know only existing CAD. They, in turn, cannot imagine what does not exist and cannot formulate or describe design tools they do not have and for which they substitute experience, intuition, and naive or sophisticated mental models. He notes that the IMS (intelligent manufacturing system) proposal is blocked by doubts about the efficacy of technology transfer. Apparently he feels that this is an illusion created by typical communication barriers between researchers and engineers, as described above. Foreigners are especially prone to think that Japanese engineers are holding back whereas the Japanese feel they are making rather full disclosure. Problems like this lead to bad feelings and slow progress. A Design Technology To encourage this discussion, I brought up the case of IHI's shipbuilding methods. These were developed over about 15 years beginning in 1955 by Mr. Shinto, who became famous for them. They are a sophisticated extension of U.S. modular shipbuilding methods developed by H.J. Kaiser during the war. The extensions include ⚫ adding statistical process control so that very large modules can be built accurately choosing module shapes skillfully so that most of a ship is made of simple modules ⚫ training groups of workers to deal with the same type of module day after day flat panel in large quantity, panel curved Chirillo told me that he had to evoke Ultimately, technology transfer to This story is interesting because it The method is an application of Group Technology, a technique that combines similar but not identical things into groups where they are treated as though they were identical. Except for this application to shipbuilding, Group Kinds of Design Technology usually is applied to machining. The entire approach used at IHI apparently grew almost organically in the minds of its originators. In the early 1970s the U.S. Navy hired a consultant, Mr. Louis Chirillo, to visit IHI, write down their method, and transfer it to U.S. shipyards to reduce the cost and construction time of Navy ships. This effort took him over 5 years and resulted in about 2 feet of shelf space of reports. These reports describe the method in full pedagogical form, complete with terminology, hierarchies of entities. (Grand Block Modules, Modules, Submodules, Panels, Subpanels, etc.), kinds of production entities (simple It is customary to distinguish three kinds of design: original, variant, and routine. These roughly distinguish design of totally new things, redesign of existing things, and routine modifications of catalog items. Kimura also distinguishes products whose design distinguishes products whose design method is understood and those whose method is not. He feels that basic automobiles and copiers, for example, are understood, whereas space stations are not. There is little to be learned from studying cars and copiers, he feels. One can observe a set process in operation but cannot witness the struggles that occur when something really new is being created. If we acknowledge that designing a car or copier can be quite challenging, what is it that makes it so? Again roughly speaking, he is distinguishing innovation from complexity. Cars are merely complex, made so because they comprise a combination of very many but probably simple things. These things are well modeled in an engineering sense, he feels, so the real challenge is past. This is probably the reason why he does not feel that Fujimoto's work bears directly on the main challenges of design research. "It's just a social problem," says Kimura. Yet if cars were a done deal, so to speak, then why are so many mistakes made designing them, why do they take a long time, what is it that some companies do that permits them to design cars faster and better, and is this question worth pursuing? And, is Kimura seeking an unachievable goal in trying to capture innovation? Can one imitate the experience of a designer who “knows" that a particular structural material, when used in a certain way, plays, say, a vital electromagnetic role in the function of the product, desirable or undesirable, and that this effect or a chain of such effects must be considered when designing with that material? Can such knowledge ever be captured? Put another way, could Chirillo have written his reports in 1955, 1960, or 1965, while Shinto's work was still evolving? Another way to pose the question is to ask if computers can ever be more than mere tools that help designers in routine ways. To go beyond the routine requires "knowing" a great deal about nature or capturing it in engineering and mathematical models. Engineers must be able to construct such models with the expectation that behavior they did not anticipate will be represented, and represented accurately. This might be called Data in, Genius out. (See the report on Prof. Tomiyama, who is attempting something like this.) Finally, can such tools be generic, or will we have copier design systems, car design systems, and so on? Kimura worries that the latter will be the only result. IBM TOKYO RESEARCH LABORATORY (TRL) 18 June 1991 Background My hosts for this visit were Mr. Chihiro Sawada (Robotics Group) and Mr. Akira Okano (CIM Technology Group). Also attending were Mr. Masayuki Numao and Mr. Keisuke Inoue of the CIM group. These are young researchers who work for Mr. Hazeki and Dr. Koda, whom I did not see on this visit. This group has close ties to Todai, both because the campus is nearby and because all are graduates. Okano was a student of Prof. Inoue, Inoue was a student of Prof. Kimura, and Sawada was a student of Prof. Miura. IBM TRL is a member of Prof. Kimura's new Product Realization Project and one or more of these people attend monthly meetings of the industrial participants. CIM to these people means more than computer integrated manufacturing, since the latter implies communication, networking, databases, and so on. This research group focusses on advanced CAD, robotics, automated assembly, feature-based design, and concurrent design. Their views are very advanced and their knowledge of the research literature and the status of commercial CAD is very good. Their opinions on fruitful research directions and approaches are very similar to my own, as discussed below. The topics we discussed were ⚫ the status of their recent research recent work in micromachines Informally we discussed the status of young engineers, salaries, cost of homes, commuting distances, style of communication in Japan and at IBM Japan in particular (slightly Americanized, they said). [See IBM Fujisawa plant visit for more on this last point.] Recent Research For the past 2 or 3 years the CIM group has been developing a solid modeler for mechanical parts based on features and constraints, plus some assembly modeling and assembleability evaluation (Ref 8). I was shown this work when I visited a year ago. In the intervening year, there has been no intervening year, there has been no additional progress because Okano, the additional progress because Okano, the leader, has been busy converting it for use by designers in the factory. He is aware that SDRC's recent I-DEAS solid modeler Release VI has both featuredefinition capabilities and constraint modeling. TRL's feature modeler is very similar to our own (Ref 9), capturing shapes of parts and form features and expressof parts and form features and expressing disassembly directions of assembly features in a database that accompanies each feature. (No commercial modeler can do this.) Assembly sequences are determined by a method again very similar to our own, in which the computer determines all the easy assembly constraints (such as parts completely trapped by neighboring parts), and the designer supplies the rest of the constraints. In the TRL system, assemblies are defined by means of mathematical solution of constraints: the designer designs each part and tells the computer which surfaces mate to which; the computer solves for the relative positions of the parts. This is more complicated than our method, which permits the user to name or click on the ⚫ their plans for the next phase of mating features directly. Okano notes research that he has incorporated into his software our method for visualizing networks of assembly sequences and the accompanying techniques for editing sets of possible sequences and eliminating undesirable ones (Ref 10). He has also added some assembleability evaluation capability. The assembleability evaluation is fairly simple at this point. It seeks to determine when parts need extra fixtures during their assembly as well as how many screw or other nonsimple moves are required. An algorithm selects the "best" assembly sequence. There is no assembly cost analysis. The scoring system for evaluation bears some similarity to the Hitachi method, which they purchased. However, their papers and discussions omit any detail about the Hitachi method due to proprietary restrictions, so I cannot tell if their software actually implements any of it. I believe, however, that they are dissatisfied with the Hitachi method and this is why they have tried developing their own. Example scoring in the TRL system is relative to full credit for downward simple insertion, with some points off for upward insertion, more off for horizontal, rotating, or diagonal, and still more if a separate fixture is needed to establish or maintain special relationships between parts during assembly. Their software also has, or will soon have, an expert system shell and some rules for simple ease of assembly judg. ments, such as determining if holes and pegs have chamfers on them. Such facts about individual features are fairly easy to determine and can be deduced from simple table lookups if the form features have been described in the database properly. However, I do not know how they store their features. CAD in general at IBM seems to be based on CADAM, a two-dimensional drafting system. About 70% of the designers use it; perhaps 30% use CATIA, a 3D solid modeler. Both of these are commercial products. CATIA is too hard to use, say the designers, who were trained in conventional drafting. The problem of difficult user interface to 3D modelers is widely discussed in Japan, not only at IBM, but no one has a really good solution. According to Okano, shortcomings in CAD and in design evaluation methods have a common factor, namely that CAD, especially solid modeling, enforces the wrong kind of design methodology. One wants to begin with a complete rough layout that shows all the parts in approximate relative locations; then one wants to design each part in detail. CAD supports the second step but not the first. He feels this is one reason why products turn out to have too many parts: the designer begins designing single parts right away and does not see the whole product until too late. Reducing the number of parts is a key assembleability evaluation technique. Several methods exist for judging if a design might have too many parts but no method exists for advising the designer on which parts should be eliminated. The same is true of any other evaluation, such as tolerances (see below). One can tell if a given assignment of tolerances permits assembly but no method exists for advising the designer on how to do the original assignment. Research Priorities and Okano and his colleagues apparently have come to these conclusions more or less independently. I could not detect strong interactions with the factory or any attempt by upper management to encourage such contact. Management does not provide a research agenda and neither does the factory. The researchers do not actively generate research ideas by regular consultation with the factory, but instead use their own judgment and pursue their own goals. Transfer of the assembly modeling technology, like most of this laboratory's work, seems rather rather haphazard. Okano said that he merely performed by the plant engineers. This Yet no one else in IBM world-wide Sawada knew of Toyota's work in rumored to have regarding releasing So, while the group understands that The group is currently in the process of developing its plans for future research. I had some difficulty determining any details. The emphasis will be on assembly. (They have no interest in machining, molding, or other fabrication processes.) A major theme is providing active advice to designers. This extends any existing work on evaluation or scoring of a design and hopes to provide particular suggestions for how redesign should proceed so as to improve assembleability, reduce part count, or improve tolerances. Another theme is improving CAD interfaces, most likely by extending feature-based design to include "analogical design." This term refers to reusable designs. Also, they want to explore the idea of "top-down design," meaning designing all of the parts of an assembly first in a rough way and then doing detailed design on each part. A fourth topic is sculptured surfaces. Finally, they are interested in helping designers evaluate tradeoffs that occur when product performance must be balanced against other factors, such as producibility. Peals of laughter greeted my question: "What do the designers do now?" Reusable designs seem to have been given the most thought so far. Okano was quite clear in several of his points. First, current commercial CAD focuses on design of completely new parts. There is no library of previously designed parts, although he and others realize that |