The manufacturing engineering department responsible for making forging and casting dies for engine parts has responded to shortcomings in existing 3D CAD by developing its own 3D modeling system. Other departments have done the same and now several modelers exist. The manufacturing department's program is called F-CAD and was introduced in 1988 as the result of a Ph.D. by one of its engineers, Mr. Suzuki. F-CAD's main strength is its ability to handle arbitrary and complex fillets. But it can't support revision, so a problem occurs because many dies are changed during tryout. As a result, the drawings are not updated to a data file. Future Needs The opinions expressed here are those of my hosts, who are trying to convince their management. Whether Nissan will adopt their ideas is unknown at this time. New engine designs promise to be quite different from existing ones. An example is modularity, in which two cylinders can be added to a four cylinder engine to make a six, for example. Interior and exterior engine parts will be different from current ones. More commonality will be needed, and engine performance must not be allowed to suffer because of it, due to the need to meet the CAFE requirements. Kobayashi wants 3D modeling to be adopted throughout engine design. He also wants better cost prediction methods early in design that can analyze complex parts, not just screws and brackets. He does not know the proportion of engine cost that goes into assembly. One reason is that so many components are purchased. He feels that part count should be reduced. When I asked if that might increase the cost of each part more than the amount saved in assembly cost, he replied that representatives of Boothroyd & Dewhurst claimed not. But apparently the cost-complexity issue is still open. my video about feature-based design for assembly a second time he did not understand it. "I thought features were only for machining," he said. He feels that in general relations between Nissan and university researchers are weak. The university research generally does not address industry's needs. On the other hand, advanced methods like Taguchi's are just now starting to be investigated. In addition to encompassing modu- 5 September 1991 Any time cost, performance, and Estimating assembly time/cost is This visit was a followup to one on 1 July. Our hosts were Mr. Tohru Fujimori, General Manager of the Robotics Division of the FA Department; Mr. Yasuyuki Yamagiwa, creator of the DAC assembleability evaluation system; Mr. Junichi Kuzusako, Assistant Manager of the CAD/CAM Group; and Mr. Hiroshi Harimaya, Manager of System Development in the Production Technology Development Group. We were joined for lunch by Mr. Juzo Akiyama, who developed the first Sony robot about 10 years ago. A description of his strategy for developing Sony's family of assembly robots may be found in Reference 1. The main subjects of the meeting were DAC, Sony's preferences for CAD capabilities, early design methodology for videocameras, and robot assembly of precision mechanisms. To support this kind of design does Getting him to focus on the differ- A new camera requires about 30 to 50 mechanical designers, whereas a modification of an existing design requires about 10. All these people are university graduates and there are no assistants like draftsmen. Sony is the only company visited that has more applicants than it can hire. The exception is experienced manufacturing engineers, who are not produced by universities. CADAM, a 2D drafting program, is used for 80% of the parts, while CATIA is used for the rest, to produce cosmetic designs of camera exteriors. Sony's own surface design software called FRESDAM is also used, along with a small number of seats of Pro Engineer and Design-Base. According to Kuzusako, 3D systems are simply too hard to use, having about 100 commands compared to 10 for CADAM. They are also much too expensive, costing as much as ¥60M per seat (clearly he has made a factor of 10 error here, since ¥60M was $444,444 at that time) with all the features. So 3D is used only when the shape is too hard to make in 2D, or if stereolithography will be used to make a rapid prototype, to check interferences, do CAE, or teach robots offline. If 3D systems were simpler to use, say because feature-based design (“a very important idea") were incorporated, not only would many designers use them but their output would be understandable by the manufacturing people. At present, design and manufacturing people are separated and communicate by passing drawings back and forth. DAC Assembleability Evaluation and Its Use in Fine Mechanism Design Videocamera and tape recorder mechanisms can be very complex (see Hitachi I&MSL reports). Much simpler cassette tape recorder/players were used as the example in this meeting. A design consists essentially of a base with parts stuck onto it from both sides. The direction of approach for the parts thus dominates the assembly design evaluation. Fastening method has next priority, while part shape is lowest. A new design often starts with a goal such as to cut the number of parts by a factor of 2 from the previous design, or to cut the cost or weight. "By half" is the division's motto. The previous tape recorder design had a metal chassis and a circuit board. Most of the parts were attached to the chassis, including the motor. Wires were therefore needed between the chassis and the circuit board. Sony's next design used the circuit board itself as the chassis, allowing the motor to be attached to it directly and eliminating almost all the wires. This made assembly so simple that the unit is made in Malaysia by simple pick-place robots. Detail design consists of making pencil sketches of multi-part mechanisms in exploded view form. At the early stage, many such sketches are made, and the assembly evaluation is one of the main criteria used to choose one concept over another. This makes Sony's DFA approach different from any of the others observed during this study. The part sketches are remarkably detailed. Thus fairly good DAC judgments can be made. Since one person accounts for 50 parts, he can make a significant portion of the design and evaluate it himself. Sony's design methodology may be different from other companies', too. The common practice is to start with an assembly drawing or assembly sketch showing the final locations of all the parts in a unit. Sony's practice as illustrated to me is that the designer starts by making an exploded view of the parts. This is consistent with a remark Fujimori made at the 1 July meeting, that designers consider assembly method and sequence during concept design, a remark I found hard to believe at that time. Yamagiwa made the point (commonly quoted in the United States) that 75% of the cost of a product is determined by early design decisions, and DAC is therefore aimed at this stage. Fujimori said that Sony has no data to back up this estimate, merely their feeling and an informal survey of their engineers. This is interesting because no one else has any data either; but everyone quotes this number! One designer may be responsible for as many as 50 parts. At various times in the meeting, some people called this a reasonable amount for one person while others said it was "too many to remember." The conflict is between the older people who do not want anything to block the designer's creativity and the younger people who see the need for computer tools to replace the sketches. Fujimori points out that Sony has a database for common parts like screws and springs but not for any of the important parts. These are radically different for each new design. Past designs are of little or no use in providing design data such as parts of nearly the desired shape. Since I had given a talk on feature-based design, he remarked that Sony's lack of part data "may be a shock to you." Design evaluation has several facets: ease of assembly, precision of the resulting part mate, and cost. Often these conflict, and precision may dominate, especially for shafts and capstans that guide tape. The parts of the tape recorder are all rather simple, and Fujimori claimed that combining them to reduce part count and create "multifunction parts" does not create any cost penalties due to increased part complexity. Part count reduction, therefore, always yields a net savings. I gathered that the assembly cost estimation method in use is not very sophisticated. Details and Demonstration of DAC DAC means "design for assembly cost effectiveness." It is at least 7 years old and was originally a pencil-paper procedure because at that time few computers were available to designers, except for drafting. Even now, it can be used by referring to a single sheet of paper that lists a series of keywords describing the points the designer should keep in mind. These 35 keywords are in three classes covering part shape, method of attachment, and direction of assembly. Direction includes considerations like space for fingers or tools. Method of attachment includes what kind of fastener (if any), how far the part must be pushed or twisted (if at all), what kind of screw head, and so on. Part shape includes general descriptions like "flat" or "cylinder," plus the largest and smallest dimensions, the weight, and whether the part is rigid or flexible. Other information in a general category includes whether lubrication or cleanliness is needed, whether the part is fragile, or whether a wire is attached to the part. Fujimori points out that “precision" of assembly is a critical factor. This really means how perpendicular a shaft will be to the base in which it is inserted. This is often more important than the exact location on the base. Press fits are deemed the most precise for their cost, although a shaft with a wide base fastened with four screws is more precise but more costly. Press fits are notoriously hard for people to accomplish and ordinary robots do not have the strength, so special high cost robots are needed. The ones in use are descendants of the first ones Akiyama developed. A more recent chassis design had a very thin metal base to save weight. Here press fits could not be used, so the shafts were riveted instead. This indicates that ease of assembly is unlikely to be the main criterion in many of these assembly steps. Yamagiwa demonstrated the DAC evaluation using a program on a NEC personal computer. There are several windows onto which the designer enters data about each part in the 35 categories. The computer calculates the score and keeps various running tallies like total part count, average score, and estimated assembly time for each part. One can input various goals such as "no need to turn the product over" or "less than four screws." The computer will then prompt the designer if a goal is not achieved. Wide differences between assembly times are noted as "poor line balance." The fact that two short tasks in sequence could be given to the same robot or person is not taken into account. Robot Assembly of Complex Mechanisms Another is a different assembly sequence. (Up to now Fujimori had claimed that the product designers think up the best sequence, so software like I demonstrated for generating alternate sequences is of no particular use...) As a first step toward both goals, Sony is developing a “computer-based manufacturing system." It will contain vision and robot tool changing, plus a top level control system run by Sony NEWS workstations. Vision will be used to aid the teaching process, while tool change will be used to permit a robot to do the work of another one that breaks down (requiring a different assembly sequence?). In the future, force sensing will be added to permit assembly the way people do it. Sony began robot assembly about 10 years ago with its own robots and now has several hundred in its plants. It also has a business selling robots and complete systems to other companies, including such rivals as Hitachi. Remarkably complex assembly moves are made by these robots in less than a second or two. Examples include threading a small TOYOTA rubber belt around several pulleys or inserting coil springs that need to be 6 September 1991 wound up during assembly to create preload. Fujimori points out several problems in such systems. Teaching the robots these intricate tasks takes time and can be positively dangerous since the parts are small and the programmer puts his head a few inches from the tool. Assembly cycle times are now as short as 2 seconds, so the loss of even 0.3 second at one station on a line can spoil the efficiency of the entire line. So he would like an offline programming system driven by solid model data. His dream: "It works the first time." The problem with this is that the data are probably not detailed enough. The robot's dynamics, not easily modeled, plus the friction and inertia interactions between the parts, will make offline teaching too inaccurate. His other dream is to have some kind of artificial intelligence built into the robot, its sensors, and its controller so that it could gradually learn a faster way of doing the task. An example is a shorter motion path that comes closer to an obstacle than originally taught. Background This visit was a followup to the one on 31 July. Our hosts were Mr. Kuranaga, Manager of Information Systems Div 1, Mr. Takatoshi Negishi, Manager of the CMM Group in Body Production Engineering, and Mr. Yasuhiko Ichihashi, Project Manager in the Chassis Design Department. Other attendees were Mr. Kato (CAD/CAM systems for suspensions), Mr. Sasano (body engineering CAD), Mr. Hatano (engine engineering CAD), Mr. Kobayashi (power train engineering R&D), Mr. Shimizu (formerly engine design, now in CAE), Mr. Yoshida (power train gear parts design), and Mr. Iwase (Info Systems Div 1, planning new CAD/ CAE/CAM systems). The general subjects were the overall vehicle design process and use of computers in engine and transmission design. Afternoon plant tours covered die design and manufacture and Toyota's in-house manufacturing equipment building division. Figure 14. Overlapping design tasks. In this figure, the leftmost vertical line in each balloon indicates when official information is received from the previous balloon; the rightmost line indicates when official information is passed on to the next balloon. Preliminary information is released earlier than the official release so that the "flying start" can occur. Activity in each balloon continues after final release of information to take care of downstream problems ("continuing responsibility"). Courtesy of Toyota Motor Corp. Vehicle Design Process Sasano presented Toyota's car design methodology. He said that, overall, it takes 3 years from start of concept design to start of production. New technologies and power trains take longer. The concept, styling, and preliminary design take the first year; detailed design, prototypes, and evaluation take the second year; and tooling design and production take the last year. Later he amplified this breakdown with Figure 14, which shows that several of the processes extend over more than 1 year. He used Figure 14 to make the point that European and U.S. car makers take longer to design cars because of a different attitude toward "responsibility and competence." In Toyota, designers have responsibility for their own designs, updating them where necessary due to the needs of downstream processes like tooling design. stream processes like tooling design. Sasano thinks that in Europe and the United States the downstream designers take over responsibility and can change the design. It is not clear why this should make the process take longer but clearly Toyota's method will cause the designers to learn downstream problems and avoid them next time. The two important features of Figure 14 are the "flying start" and the "continuing responsibility" or downstream followup. Flying start involves two processes: an essentially free check of the design by the tooling people and a costly/risky early start on design and construction of the tooling itself. The vertical arrows in the figure denote official drawing release but there are clear cases where flying start extends back almost to the beginning of the previous task or even two previous tasks. Flying start information is not informal and is not transferred freely “among friends" (my terminology to indicate unsanctioned transfer). He was very firm on this point and it seems to contradict Prof. Fujimoto's description of a process that depends on long-term personal contacts. According to Sasano, there are various levels of approval, but all information transfer is approved. Final release has top management approval, but intermediate release of incomplete information can be done with departmental approval. "Incomplete" can mean preliminary and subject to change, or it can mean that main structure is shown but details are omitted. How is the timing and content of these intermediate transfers worked out? What was the history of its evolution? "Difficult questions." The answer was typically Japanese: Our engineers are highly educated and have universal experience (i.e., they do not have a few narrow skills). They must harmonize with their job environment. Sometimes information is offered, sometimes it is requested. The correct action is a common subject of discussion among themselves. Die design, power train design, and chassis design all utilize similar principles. The whole process is controlled by the Chief Designer (what Fujimoto calls the Heavyweight system). He is responsible for launching the effort and deciding all the tough technical issues. Among these is allocating space for wheels, engine, and transmission in a front wheel drive car. Styling sets the spacing and the power train people fight it out for the space between. To catch problems early, Toyota starts engine compartment layout as soon as early styling sketches are available because many of the "hardpoints" (fixed dimensions like wheelbase) are determined at that time. A special crossfunctional working group chaired by the Chief Designer is formed to handle this problem area. Wireframe 3D computer data are used to aid the decision process. Often its meetings are short because design consists of choosing an existing engine-transmission pair. When a totally new engine and transmission are used, car design can take longer than the 3 years cited above. Computer Support for Concurrent Engineering Sasano showed a chart (Table 2) delineating the two most important issues, databases and communication, showing the present status and future goals. For current data representation, there is a CAD model of the car in the form of a 3D wireframe or surface model, plus some parts that are modeled in a hybrid of wireframe and surface representation (the latter for complex surface regions). Communication is currently supported by data conversion software and by means for distributing data electronically to downstream processes. No mention was made of communication upstream. Since electronic data now consist mostly of geometry, a lot of attribute data are still issued on paper. Therefore, Sasano seems to think that electronic data representation is merely a change in media. Several of us in the meeting disputed this and he relented. Hatano pointed out that in fact only body data are in 3D while all the rest are in 2D. The result is a lot of conversion to 3D to permit interference checking. This is painful and in fact the checks are done semi-manually. Multi-color wireframe computer drawings are projected onto a big TV screen in a meeting room and everyone talks them over. It is an awkward process and Kuranaga asked them if they plan to switch to all 3D in the future. Their answer is the typical one, namely, that 3D takes too long and the designers like 2D. In the future, Sasano says that the database will be a combination of CAD models, bill of materials, part attributes, and documentation like process instructions. The future communication system will comprise a universal communication network. This will create "integrated engineering." Prof. Kimura says he has heard Toyota people talk about this future. dream system before but he feels they are still debating its details. It appears to be less developed than the ideas outlined by Mazda. Neither company seemed to have made explicit provisions for passing information back upstream. The issue came up at Toyota |