by Daniel E. Whitney, Senior Research Scientist, MIT

Hosts:

Mr Mineo Hanai, General Manager,
Mr Hiroshi Harada, Manager, Production Engineering Department,
Mr Fumio Kojima, Project Manager, Production Engineering Department,
Mr Tom Kita, Production Engineering Department,
Dr. Kazuo Matsumoto, Director of Production Engineering Department and member of the Board of Directors

Nippondenso remains one of the premier design and manufacturing companies in the world. It is a prime example of a vertically integrated manufacturing operation, with strong capability to develop its own manufacturing equipment and systems. When I last visited in 1991 I was told that there were 2500 robots in the company, 90% made in house, with plans for 1000 more the next year. At the time I wondered if this pace could be kept up; on this visit I learned the answer: ND now has 7000 robots, nearly all made in-house. It has also won at least three of the prestigious national automation excellence awards. The Instrument Cluster factory contains two such winners.

ND's focus on production technology is supported in several ways. Many departments required for production, such as Flexible Automation (maker of the robots) and Production Engineering (our hosts' department) are part of corporate headquarters and are support in whole or part by corporate funds. Furthermore, ND devotes a much higher proportion of its engineering staff (35%) to production engineering than most Japanese companies (10%). The Production Engineering Department makes complex automation systems and designs many that are made by contractors. It also provides expertise such as design for assembly and simulation. In spite of the availability of software tools, it is Mr Hanai who decides finally what will be automated and what will be done manually. Bosch is still a stockholder of ND, based on licensing of Bosch fuel injector technology to ND in the 1950s. Now ND has some good product technology which it licenses to Bosch, but product technology flows mostly from Bosch to ND while production technology flows mostly from ND to Bosch.

In several ways, ND is attemping to decrease its dependence on Toyota. It sells to many other car makers (not to Nissan, apparently) and it has recently branched out into non-automotive areas that exploit its technological strengths. These include cellular phones and other mobile communications systems, and sales of its line of robots, in total comprising about 6 - 7% of sales. When I asked about Toyota's attempts to gain more control over electronics, I was told that this was "a delicate question ... a matter for top management." ND buys about half of the ICs it uses such as microprocessors and memory, and makes the specialized ones. Everyone seems to understand the fact that cars are now "intelligent" so the importance of this issue is clear.

Prof Taka Fujimoto told me on January 22 that ND has attained such power in the Japanese car industry that it can sometimes dictate system integration factors or conditions to its customers. A prime example concerns space allocation under the hood, surely a delicate topic due to the fact that such space is highly limited. If ND proposes a component or system (alternator, air conditioner), it may also propose the space that the component will occupy. This is quite the opposite of the usual situation, where the car company proposes the available space as well as the component's operating requirements. That ND should have attained such power is certainly an event of note. Presumably ND offers something in return in such situations, such as a proven technology, a finished and reliable design, or something else that will save the customer money or time. I have no examples at the moment and I did not know of this issue during my ND visit.

Harada and Kojima are especially interested in assembly. While in the US for three years, Harada worked on developing a Design for Assembly (DFA) method whose implementation is now Mr Kojima's responsibility. Some questions are directed at ND's main problem, namely control of variety of parts. A relevant question scores a part positively if it can be used in several versions of the product and negatively if it cannot. Toyota has reduced the variety of the things it orders by about half, but this reduction is like from 20 to 10, say. Even if 10 varieties are "in the catalog," complex equipment and scheduling are needed. Only if the variety were cut to 3 or 4 would a significant reduction in manufacturing complexity result. At the present levels, one saving grace is that intra-day changes are fewer than in the past, but inter-day changes are as big as ever.

The current DFA method requires designers to answer 65 questions about each part. Even though the new method requires fewer, the designers seem to need to be forced to comply. For this reason, Mr Harada is interested in any new methods that could be used to provide automatic answers based on CAD representations of the parts. Such on-line help would improve the designs, inasmuch as designers' knowledge of assembly and manufacturing is limited. In any event DFA is credited with saving about 30% of the cost on a new product, based on doing typical cost estimates before and after applying DFA. Part count reduction is the main source of these savings. But he admits that their DFA is still part-centered, that is, it deals with one part at a time and does not force multi-part issues to the forefront. A massive cost saving or part count reduction would require major redesign and a new product concept.

ND still uses its 20 year old internally-written CAD system called NADAMS, which is a 2D system with some 3D capability. There are rumors that ND will purchase a modern 3D solid modeling system and apparently no thought is being given to writing one internally. Most of their products are basically 2D or 2.5D so they do not feel a strong need to upgrade. Better cost estimating methods are also desired, to supplement the current method of looking up the cost of similar products in books and relying on human experience and memory. It is not clear if NADAMS could be the basis for an automated DFA/cost system in view of its inabilty to capture real geometry.

To overcome variety in small batch production is especially challenging. Sometimes batches are as small as 20. For air conditioning fans, ND built a robot assembly line complete with force-sensing robots to insert rotors into stators. This line won another of the above-mentioned automation prizes. Mr Hanai refered to this machine as having some human-like features, by which he meant learning and improving. These he grouped, in typical ND fashion, into three classes: automatic adjustment that is programmed in, ordinary Kaizen carried out by shop floor people, and longer term upgrades by engineers in response to problems or opportunities.

ND is facing the need to outsource production to other countries, as are many export-oriented companies, and for some similar reasons. These include cost, local government regulations, and the need to have a plant near where the assemblies will be used. Often it takes several years for plants in SE Asian countries to become profitable, but immediate profit is not the goal. Some simple items are made outside Japan for pure cost reasons, especially if manual labor can be used. In such locales, robots are used only when quality demands it. In Japan, robots are used to reduce the shortage of workers or to cut labor cost. Interestingly, ND now has 38 overseas affiliates or plants compared to 11 in Japan. But most of the employees are still in Japan.

Visit to Takatana Plant
This plant occupies 78000 m2, employs 1509 people, and manufactures dozens of instrument products, such as sensors, displays and gauges. Most items are made from raw materials. For example, the magnetic coil rotor of the speedometer is a plastic molded piece with four terminals, a steel shaft, and four fine wire coils. ND buys the plastic in pellet form, the steel in wire form, and the magnet wire. It does everything else in house to make this relatively simple part, along with thousands of others. ND has the market share lead in both Japan and the world in these products, followed by Yazaki in Japan and DELCO world-wide.

The product we saw being made is an instrument cluster. These are required in hundreds of varieties made for about a dozen companies. Production rates are a quarter million to a million per month, depending on the assembly. Each cluster consists of a plastic rear base, a flex circuit board, many lamps, several instruments such as speedo, tach, fuel, and temperature gages, and a clear bezel. The plant is divided into three departments: molding, movement assembly (such as tach or speedo), and final assembly (base, flex circuit, lamps, movements, bezel, and test).

The machinery in each department operates essentially without human intervention. People are required for most material handling between major departments, and quite a lot of people are running around fixing the equipment as minor stops occur. The movement and final assembly lines are quite complex and stops occur every minute or so.

The molding department contains about 80 machines which are unloaded autmatically by robots which stack the parts neatly in bins. These bins are moved manually to the assembly departments.

The dial face printing department won an automation prize. It makes dials directly from bulk polycarbonate stock. The face designs vary a great deal and are made by multi-step silk screen printing in a set of automated printing machines. People change the screens frequently, on the fly while the machines are working. Dial shapes are complex and vary a lot from style to style. They are cut out of the stock with a laser cutter. Operating, locating, and fastening holes are cut by die cutters. This selection of processes was motivated by the relatively high speed of die cutting of simple shapes and the corresponding relatively high speed of laser cutting of complex shapes.

The movement assembly line operates with a cycle time of 3.35 seconds, making tachs on one branch and speedos on the other. A common portion of these lines covering the first dozen or so steps operates at a 1.6 s cycle time. The stations are mostly small fixed program mechanisms or relatively fixed cycle pick and place robots. This line makes galvanometer movements comprising a rotor with a shaft and coil, a base with bearings, a hair spring, and a small circuit board on the back. When it is complete, the dial face is put on and then the needle is attached. A vision system aids needle attachment, setting the needle so it reads 40 km/h in the no-power state. There is no calibration or zeroing done on these meters. They are simply tested using standard voltages and RPMs at a later stage and adjusted manually if they do not pass. This is quite remarkable to me since I am familiar with such lines in other companies and businesses, where zeroing and calibration are both required. ND used to do it as well.[1] Obviously a lot of process consistency has been achieved here in controlling such things as hair spring stiffness and coil magnetic field strength.

Final assembly occurs on a line containing about 46 multi-axis robots of several configurations and sizes, all made in house. The cycle time is 13 seconds. Both robot programs and multiple part feeders are used to achieve flexibility of product type. Some stations do complex things, such as folding a flexible circuit board over a 90¡ step on the back of the instrument housing or inserting lamps by a push-twist action. Final test requires that different plugs be connected to different styles of cluster. A robot loads each test line pallet with the required plugs, then places the cluster in the pallet, then plugs in the plugs. All testing is automatic, with vision systems looking at the lamps and gages while they are exercised by the machine.

This line is quite complex and contains several vision-guided robot actions. The most complex of these involves folding the flex circuit around the back of the case and fastening it down. Problems with this and other stations raised the cost and lengthened the development time of this line. The line stopped quite often while we were there. These facts indicate that ND may have reached at least a temporary plateau in automatic assembly.

The Production Engineering Department designed and implemented this line, using robots made by the Flexible Automation Department. PE also wrote the robot and line control software and the specifications for the scheduling software. Programming of the scheduling system was 50% outsourced, and the relational database to support it was purchased. "You can't buy suitable software for controlling a JIT system."

Conclusions
This visit reconfirmed my earlier impressions of ND and indicates that it is keeping the major strategies, design methods, internal technological strengths, corporate organization, and in/out sourcing policies mostly intact from the 1980s. Internally-grown technical capability in product development, production automation, and production flexibility are its core strengths. Its only major concession to current conditions appears to be a slowdown in replacement of retirees in an attempt to reduce domestic headcount as painlessly as possible. Headcount appears to be down about 10% from 1991.

Footnote:
ND's former calibration and adjustment method, which involved computer vision, motor-controlled screwdrivers, and long term learning, is described inNevins & Whitney: Concurrent Design of Products and Processes, page 286. The gage described in the book contained a bimetal strip that had to be bent in order to achieve calibration, so the design was different from these speedos. Nevertheless, other galvanometric gages I have seen made required two adjusting steps: full scale calibration and zero-setting. ND does neither on the new speedo line.