Lapas attēli
PDF
ePub

They had no access to the first round capabilities are represented by a Monte process.

The first round takes 3 months, the second 4 months.

Optical System Design

The optical system division designs both glass and plastic lenses. While plastic poses problems from thermal expansion, it is seeing increasing use in laser printers. These lenses are manufactured in a plant 500 km north of Tokyo. Lenses can have between 3 and 20 elements, the latter applying to complex zoom lenses. Design takes 3 to 6 months and each lens is designed by one person, of whom there are 30 in the department I visited.

Lens design is supported by both commercial ray tracing software (CODE V from Optical Research Associates in Pasadena), which operates on DEC VAX workstations, and by Ricoh's own, which operates on an IBM 3090 mainframe in Yokohama. Ricoh makes most of the equipment it uses to make lenses. This includes polishers for glass lenses and molding equipment for plastic ones. The spot size of laser beams emerging from lenses is measured for quality control purposes using equipment Ricoh builds.

The process of designing a lens starts with a given specification in terms of cost, focal length, product application, and so on. The first step is to choose a lens type and number of elements. The main construction parameters, such as geometry, lens radii and separation, type of glass or refractive index, and so on, are chosen, often using those of a similar previous design as well as patent data. Ray tracing plus trial and error are used to modify these parameters until the specification is met. The expertise of the designer is the main ingredient.

A major issue is deciding what the tolerances should be. These include all the above parameters plus those of the housing, called the cell. The factory's

Carlo analysis, and many studies are run. Particular attention is paid to the "sensitive parameters." These are identified by numerically developing a coefficient table, essentially an empirical array of partial derivatives. The effect of small changes in parameters on lens performance is judged by noting the change in aberrations and the mean time to failure (MTF), a measure of image contrast.

Interestingly, the study varies only the parameters that engineers think about, such as radius of a surface, and does not consider that the ideal spherical surface might not be spherical at all. In fact, lens polishing can produce quite spherical surfaces, but this is not yet true of molding. Another surprising yet true of molding. Another surprising problem revealed at a later meeting is that the periphery of a lens cannot be held concentric with the optical axis to much better than 10 microns. In a stack of many lenses, nonaligned optical axes can seriously degrade overall performance. A more accurately made cell does not help this problem.

If performance degrades too much in the face of the stated tolerances, then they are tightened, although some attempt is made to avoid a “sensitive" design, that is, one whose performance is easily degraded if one variable varies even a little. At this point I asked if they use the Taguchi method, a technique often used to attain low-sensitivity designs. I was told that Dr. Taguchi visits Ricoh regularly and gives seminars so that the engineers by now are familiar with his philosophy and follow it in principle. However, the main design approach is numerical optimization, using a damped least-squares method.

This optimization, however, is not used to seek a low-sensitivity design. used to seek a low-sensitivity design. Instead, it is used to find a set of parameters that best balances a variety of conflicting aberrations, some of which get worse when others get better.

Camera Mechanical Design and Use of CAD

Use of CAD is widespread at Ricoh. In 1990, 60% of 50,000 mechanical parts were designed by CAD by 1,000 engineers using 300 networked workstations or terminals. Electronics is even more advanced: the database has 35,000 parts, and 400 engineers have access to 200 networked terminals.

Like Nikon's, Ricoh's cameras are made almost entirely from precision molded fiber-filled plastic. Typical part

count is 300 to 400. I was shown the new Mirai Zoom 3, an autofocus, auto zoom, auto exposure, auto flash camera. It sells for $250 in the United States. Every part was laid out on a series of foam-core boards around the room where we met. The largest of them is the condenser for the flash. There are four motors which are either somewhat or dramatically smaller than this condenser.

The Mirai took 2.5 years to design, 6 months more than planned. There were 30 full- or part-time designers on the project: 10 interior mechanical, 5 optical and factory lens people, 10 other factory designers, and 5 assembly engineers. Considering that Nikon did not count factory people for me, the size of teams from the two companies is similar, as is the time required. But Nikon mostly designs much more complex single lens reflex (SLR) cameras. For example, Nikon uses focal plane shutters while Ricoh uses simple two-blade wing shutters in its compact cameras.

The design process has four parts, of which exterior design comes first. The others follow and are undertaken somewhat in parallel with the expected precedences: electronics, optics, and mechanical. The last is done with Intergraph systems. Electronics is supported by other CAD that does flex circuit layout but cannot support circuit simulation or analysis. Similarly, the mechanical CAD cannot support

mechanism or structural analyses. Another computer is used occasionally for some FEM work.

how this is done. Sony has a DFA method
that they plan to try soon.

At present, all mechanical assembly
is manual, mostly done overseas. Next
spring they will install a line of 30 Sony
robots to assemble a shutter mecha-
nism. I think the possibilities here are
large.

The Intergraph system supports 3D wireframe modeling as well as external free surface shapes. If parts interfere with the case, they stick out in the picture. Design often consists of the mechanical people fighting the exterior shell people for space. However, Future Needs they admit that CAD cannot adequately predict these interferences. "We can't really tell until we have the parts in our hands."

As at Nikon, camera design is done in blocks. Each block is done in parallel by a separate team that tries later to fit its block to the next team's block. The

defects in this method are apparently known to them and their "restructuring" process is aimed at reducing such problems.

External parts are defined by breaking up the surface given them by the industrial design group. This is just like car design, but the camera people are not used to it. There is no strong tradition for deciding where the boundaries between sections should be, and no FEM is used to see if a choice might be bad for strength or dimensional stability.

Ricoh Optical Company makes the molds and parts from NC data provided by the above process. Typical tolerances in critical areas are about +0.05 mm or 50 microns (0.002 inch). The parts are made from 20% glassfilled polycarbonate.

Flex circuits are also designed on the Intergraph system. Ricoh is trying to locate integrated circuit design software. I mentioned Mentor Graphics and Harris Corp.

In all, Ricoh agrees that its current CAD is an electric pencil. The main advantage over real pencils is that changes can be made quickly.

No systematic DFA methods are in use. However, the factory critiques the designs and weights ease of assembly 60% of the total. I could not find out

second lens using an optical measuring system. When adjustment is finished, an ultraviolet (UV) curing adhesive is injected, which cures in 15 seconds. This is, of course, the classic tradeoff against tightening tolerances. The cost may lie in the fact that 15 seconds is barely short enough to support production.

SEIKO-EPSON

They want better ability to predict
fabrication and assembly costs, plus 30 August 1991
ways to predict the effect of tolerances

on lens performance. Thus they were Background
very interested in our assembly anal-
ysis software. They also want better
solid modelers that can check
interferences for them fast. Such is
essentially available now. They also want
ways to set tolerances automatically.
No such thing exists.

Mr. Nishi showed me their early
work on tolerance analysis. It has two
components: identification of contact
chains between parts and identifica-
tion of possible detailed contact condi-
tions between various surfaces and edges
on parts. The contact chain represents
the nominal situation. The possible
contacts can pose a combinatoric
explosion. It is not clear if his method
explosion. It is not clear if his method
actually investigates them all or if he is
just illustrating some of the possibilities.
Lenses rest against each other and
determine each others' positions, but
his method does not analyze the fric-
tion in these contacts or the symmetries
imposed by resting spherical surfaces
on each other. Thus he may permit odd
contacts in his analysis that are unlikely
to occur in practice. It is OK to be
conservative but there are cost penal-
ties for doing so since one may pre-
scribe tolerances that are unnecessarily
tight.

Later Mr. Ageishi told me that they recently built a machine to install lenses recently built a machine to install lenses and check the optical alignment. The machine installs the first lens, then installs and adjusts the position of the

The plant I visited is in Shiojiri, about 2.5 hours on the train from Tokyo. It makes dot matrix printers, laser printers, ink jet printers, and personal computers. Discussion on this visit was exclusively on dot matrix printer design and manufacture. My hosts were Mr. Akio Mitsuishi, General Manager of printer design and development, Mr. Asada, a print head designer, and Mr. Mitsuyoshi, General Manager of CIM.

Seiko-Epson was formed in 1986 through the merger of Suwa Seikosha Co., Ltd., a famous maker of watches, and Epson Corporation. The combined company has 13,300 employees in Japan and 11,000 in 27 other countries, including the United States. Products include printers, personal computers and laptops, memories, semiconductors, motors and rare earth magnets and, of course, watches. The company makes liquid crystal displays (LCDs), some of which it uses in its laptops. It also makes, uses internally, and sells a line of accurate assembly robots.

Epson makes two main kinds of dot matrix printers, those with 9 pins for low end use and those with 24 or 48 pins for high end use. Heads are made at another plant using automatic assembly. Mitsuishi claims that no one else can assemble print heads automatically, especially insertion of the wires. I confirmed on one other visit that a competitor inserts its wires manually.

Like most Japanese companies, Epson has trouble finding new employees, especially software engineers. Shiojiri is not as exciting as Tokyo but "houses are inexpensive and the environment is good."

Product Development
Process for Printers

The organization of the design process changed about 2 years ago when Mitsuishi introduced team design. His motto "common goal" helps the designers create the best printer, not the best print head, for example. Prior to that time, separate groups designed each subassembly, which led to problems. However, even before team design was introduced, there were design reviews in which each part was critiqued.

Design occurs directly on the computer screen with little or no pencil sketching first. They use their own CAD software for this (see below). However, they do not use any formal DFA methods, although they have considered it for 2 years. He studied the IBM ProPrinter, which was heavily publicized in the United States about 5 years ago as an excellent example of DFA. It contains many complex plastic parts whose use has greatly reduced the number of parts compared to competitive printers of that time, including Epson's. These parts snap together, reducing the need for screws.

Epson's recent printers are similar to the ProPrinter in this respect. In spite of this fact, Mitsuishi was not completely satisfied with the ProPrinter. He admires its ingenuity but worries (with some personal knowledge, apparently) that it is hard to find vendors in the United States who can deliver the complex plastic parts that such a design method requires. (Many Japanese say the same thing: high quality suppliers are hard to find.) Epson has its own precision mold making and plastic part making division for such items.

Product design, excluding technology development and print head design, typically takes about a year and involves a team of 15 to 20 people. A typical printer has about 100 parts plus at least another 50 in the print head. About five production engineers support design.

robots inserts each wire individually by grasping the armature, using no vision or force sensing. I tried this with my own hands and poor vision and found that the wires almost fall in by themselves.

In other respects, too, the print head is a good example of DFA. All assembly is from one direction, and it is held together by a few spring clips.

A tour of an automated printer
assembly line (details below) showed
that their printers have been designed
so that most of the assembly opera-
tions are from above or the sides and
can be accomplished by rather accu-
rate XYZ robots (±0.03 mm repeat-
ability). Manual intervention is needed
for one turnover and to handle wires Use of CAD/CAM/CAE
and a few tests, as well as to undo jams
in the screw feeders for five screws that
hold in the power supply and a circuit
board. Almost all the parts are plastic,
except for a sheet metal foundation for
the head transport mechanism and
several precision steel shafts. The plastic
parts are glass-filled ABS and many
are quite precise gears and levers. The
division where these are made also
makes lenses for eyeglasses and laser
printers.

In general, print head assembly appears to have many aspects in common with watch assembly, and the sophisticated expertise built up in watch assembly has been utilized in printer manufacture.

They are particularly proud of their print head design because it permits automatic assembly. The wires must pass through entry holes that are arrayed in a circle and exit through holes that are arrayed in two straight lines. To guide them from one pattern to the other, four intermediate guides are provided, each with holes in a different oval pattern. Special software was written to decide what these intermediate hole patterns should be so that the six holes for each wire lie in a straight line. Assembly then requires only the correct straight line motion.

Wires are about 0.2 mm in diameter, and the holes are about 0.25 mm. This size is typical in the industry. Each wire is about 2.5 cm long and quite straight. It is welded to an armature at one end, giving the whole thing the shape of a tiny golf putter. One of their

Seiko developed its own CAD software for 2D drafting in 1979. Originally used on a Univac machine, it was ported to Apollo workstations in 1985 and transferred to the printer division at that time. The company now has 400 networked workstations, 250 of them in the printer division. Printer design is done by 350 people, of whom 100 mechanical designers each have their own workstation. Electronic design is done on DEC or HP computers, while software development is done using Sony's NEWS workstation. In the production engineering division, 200 production engineers are networked to the design division's database.

For the moment there are no plans to convert to 3D modeling for general product design. The reason given is the investment in special software to support printer design, such as that mentioned above for print head wire trajectories. However, SDRC's solid modeler or ProEngineer is used to support NASTRAN, MARC, and RASNA (all FEM packages) for various kinds of CAE. Moldflow software is used by the molding division to critique part designs, but the part designers themselves do not use it.

In spite of Epson's reputation for high reliability printers, I was told that no special software is used to predict

lifetime of critical parts, such as the welds that join print wires to armatures, the belts that drive print heads, or the coils in the head (subject to a lot of heat). They simply test or attempt accelerated life tests. The vibration behavior of print wires is very complex, giving rise to base frequencies of 1 kHz or more plus many higher harmonics. They are attempting to simulate the physics of fatigue in such parts but they have not made much progress. Mitsuishi is willing to fund research on this topic.

Future Needs

Mitsuishi cited four areas where he would like improved computer support: tolerances, design for assembly, routine data transfer, and cost modeling.

He feels that CAD generally is just an electric pencil with the advantage that it is easier to erase electric pencil than physical pencil. Other than that, CAD offers him no real engineering support. He would love to have a way to decide if the correct tolerances have been specified.

He is aware of simple rules for DFA such as making assembly moves from one direction (not possible with his current printer designs) but he does not know of many others that are really helpful. [Note that his printers are much less complex than, say, Sony's or Hitachi's video cameras.]

Routine data transfer causes problems all the time, not only between Epson and its vendors but even inhouse. The problem is most acute when transferring NC data from design to mold makers. Molds require draft angles and different vendors require different angles, depending on their skill. The base design usually has no draft angle. Keeping all this straight is apparently a big problem. Another source of data transfer problems is in tolerances, where designers may specify asymmetric tolerances (+7, -5) whereas most CAM software requires symmetric specifications.

[blocks in formation]

This system has 35 basically identical XYZ robot assembly stations. The 2 September 1991 first half was installed 7 years ago and

took 15 people less than a year to Background design and install. The second half was installed 2 years ago. It can build two kinds of printers, but these apparently differ only in the number of wires in the print head. Production capacity is print head. Production capacity is 350,000 per year, two shifts.

Parts are delivered by automatic guided vehicles (AGVs) to places where people dump them into vibratory bowl feeders or place delivered pallets on racks.

A typical station has one or more feeders (two bearings, for example) and special tooling on the end of the robot that can grasp one bearing at a time and then assemble them one at a time. Some robots have more than one gripper on the tool and can grasp and install a second part or even a third one, since the cycle time is a comfortable 30 seconds.

Gear teeth are mated by slowly turning one gear; in one case the gear on the rubber drive belt is turned by a finger that stretches the belt; in another case, a blast of air is used.

Several shafts are installed by the "parallel parking" maneuver since both ends must be inserted in closed holes. These assembly moves are effortless sequences of fixed pivot tool rotations and linear robot motions.

At the end of the line, ladies handle the wires and paper needed for final testing, but packaging is entirely automatic. A curious feature of final

Our hosts were Mr. Yoshimi Okada, General Manager, Information Systems Dept 1, and Mr. Masahiro Matsumoto, Assistant Manager, Production Planning Department. Others in attendance were Mr. Yasuto Tatsuta (Matsumoto's boss), Mr. Hideki Kita (CAE for engines), Mr. Tuginobu Tomita (CAE facilities), and three people from automatic transmission design and production: Mr. Yoshinori Kurihara, Mr. Haruto Kurihara, and Mr. Susumu Ihara. Matsumoto did most of the talking due to his command of English.

Mazda has about 30,000 employees and sold 1.42 million cars in 1990, of which 833,000 were exported. Sales in 1990 were $16.3B. The headquarters and main plant are shoehorned into odd-shaped and separate pieces of land around Hiroshima's waterfront. Use of space and arrangement of material flow must be done carefully. Like Toyota and Nissan, Mazda has an Information Systems Division that creates advanced CAD/CAE systems for body engineering and other applications. However, Mazda cannot yet design cars directly on computer screens. They see additional computer capability as essential to improve product quality, shorten product development and preproduction lead time, reduce the cost of prototypes, and enable more alternatives to be investigated.

Product Development Process drawings. Following tests, the final drawings are made and issued and a prototype is made from these drawings. No final prototype is made. Instead, the last prototype is used as the basis for final revisions after which pilot production models are made at the factory. They acknowledge that this method is risky and involves changes. Die design is affected the most, since engines and transmissions are designed ahead on a much longer cycle. Die change affects outside suppliers who now provide most of the dies.

Mazda has about 4,000 car design engineers and 1,800 production engineers. Typically about 20 car design projects are underway at any one time, of which half are major product specification changes and the rest are face lifts or partial redesigns. Two-thirds of the designers work on body and interior, while the rest work on power trains. As at Toyota and Nissan, many components are purchased from suppliers such as Nippondenso, and much of the production equipment is made in-house. However, Mazda has spun off its machine tool division to form the subsidiary Toyo Advanced Technologies, Inc., and is using the resources to "in-source" many high value-added items, thus evolving toward the structure of U.S. car makers. This is "the road to survival" in their opinion. Mazda is having problems finding people to do the engineering, especially software people. As at many companies, women are being increasingly recruited into engineering, drafting, and design tasks.

Kita passed out four confidential charts from which he explained the overall product development process. While confidential, these charts did not show anything really different from what other Japanese car companies do, including the overall time required from beginning of exploratory styling to start of production.

Product planning and development are two separate stages, and design is governed in both stages by a series of design reviews. Production people take part in the reviews in both stages. A mechanical prototype to test engines and transmissions is built during the planning stage and uses an existing car as the platform. Preliminary drawings are issued at the beginning of product development, and tooling design begins immediately. (However, some preliminary tooling design begins even before preliminary drawing release.) A prototype is built from the preliminary

Later Tomita, in explaining this process again, remarked that car design takes "too long" in the United States and Europe. When I asked him why, he said, "That's a good question." (Before moving to the CAE department, he spent 3 years in car design.) His reply was that Japanese car design is so competitive that all the companies struggle to design faster just to keep up with each other, essentially ignoring overseas competitors.

He says that small cars are easier to design and build, and it is easier to change their design. When I replied that small cars present severe space allocation problems, he replied that years of making small cars have given Japanese car makers the required knowhow to attack the space allocation problem inside the car, especially in the engine compartment. When I asked him for examples of this know-how he could not provide any. The older engineers merely teach the younger ones what to do.

However, Mazda has developed a number of "design manuals" or "design standards" that give criteria and instructions for design of parts and components, and young engineers can learn from these. Each item, such as a radiator, cylinder block, or piston, is evaluated according to a set of criteria such as stress, strain, producibility, clearance around it, serviceability, and so on. Each criterion has a stated reason, an important point since the reasons

probably contribute to resolving conflicts and educating young engineers. Among the producibility evaluation criteria are: justification of tolerances for function and surface finish and access for assembly. These are presented in the form of a manual checklist. No formal producibility or assembleability evaluation software is used. Instead, lessons learned from previous designs are written up according to formal Production Requirements Procedures and sent back to the designers for use on the next design. A database of these writeups is starting to be made.

The design manuals are going to be the basis for an effort to create “automatic design." This apparently means taking the steps in the manuals and programming them into a questionand-answer (Q&A) interaction between a designer and the computer. Mazda hopes to integrate such programs into one large program that coordinates the activities of many engineers, launching their work, receiving their replies, passing those replies to other engineers, and so on. Naturally, a lot of analysis of the design procedures and information flows will be needed before this integration can be achieved.

Longer term, their objective is to further reduce the need for mechanical prototypes by the process of "designstage verification." The tools to be applied to achieve this goal are more design reviews, computerization of design standards and creation of standards for more items, and use of reliability analyses (failure modes and effects (FMEA) and fault trees (FTA)) in cases where conventional design analysis techniques are not sufficient or a conventional product technology cannot be adapted.

In this regard, I asked them about a joint program they had with a U.S. company several years ago. Due to my own experience, I knew that both companies shared production responsibility for a certain automatic transmission, so I asked if they made any

« iepriekšējāTurpināt »