approval of issuing partial information in an overlapping job environment and how to organize feedback and design critiques that are based on partial information.

An essential element in Nakajima's approach is to restructure the design process to optimize the flow of information. Last year at the Japan-U.S.A. Manufacturing Research Exchange, Nakajima presented the following fourstep methodology for finding a good sequence of design tasks:

1. Analyze the steps in the designdevelopment process, find out how long each step takes and what their precedences are, and determine the critical path.

2. Subdivide the steps on the critical path and carefully determine information precedences for the newly subdivided task steps.

3. (Somehow) rearrange the sequence of these steps in view of the information available upstream that is needed downstream.

4. Determine the new length of the critical path and repeat steps 1-3 until the best arrangement is found.

Nakajima had no plans to use any computer aids in this activity except for conventional critical path analyses.

As of this year, the methodology presented by Mr. Chikata for use on the HYPER 90 project consists of

1. Find critical path as above.

2. Analyze information flows as above.

3. Begin ordering long lead time items. and designing tooling and fixturing before design is finalized, using smarter awareness of when critical information is available and guessing the rest.

A computer-aided process planning system that uses such features and a group technology segmentation of their parts plus knowledge of expert machinists. This is written in OPS 83 and features easy entry of tabular process data by the designers themselves, since IHI has no knowledge engineers. A major objective is to extend their current ability to sequence single machining cuts into the ability to plan the sequence of major operating steps such as cut, measure, heat treat, and so on. Choice of the last operation to guarantee part quality is an important element of such plans.

Automation of optimized blade design, including direct data transfer between aerodynamics, stress/ thermal analyses, vibration, and preparation of process instructions and NC programs, with the goal of reducing blade design time from 6 months to 2. Right now 80% of the time goes to obtaining, translating, and verifying blade shape data from aero design to structural analysis!

Stated only obliquely but clearly on their minds is the shortage of engineers and the need to leverage the experience of their senior people. Much as

My host for both visits was Mr. Awane, formerly of Hitachi PERL, now director of Mazak's new Tsukuba R&D Center. Mazak is a privately held company and no sales figures are available. It has only 4,200 employees world-wide, 2,600 in Japan. Of these, about 400 are designers, split evenly between electronic and mechanical. The main products are NC lathes and milling machines, machining centers with ability to work on several parts in series, plus complete FMS (flexible manufacturing systems) including large parts stockers and transfer robots. About 7,000 to 8,000 machines are made each year.

The company is known for its advanced computer-controlled manufacturing systems. Many of the partsmaking facilities operate unmanned overnight and require little tending during the day, a major reason why output per employee is so high. Unlike most machine tool builders, especially in the United States, Mazak is both a pioneer in FMS and a large scale user of them. Mazak is also responsible for a number of important machine tool design innovations, such as the integral cutter spindle and drive motor. The integral spindle runs faster and with less vibration since there are no gears or belts between the spindle and its drive motor. The machine's accuracy is thus greatly improved.

Given this background, it is surprising to learn that Mazak is relatively primitive in its use of CAD and CAE. Design of new machines "takes a long

time. This is a problem for our president." Like other Japanese companies, Mazak relies heavily on the experience of its senior employees, and some senior managers do not trust computers in design roles. In this sense, Mazak is typical of conservative machine tool makers world-wide. Mazak recognizes the importance of CAE but is not satisfied with software currently available from commercial sources.

Mazak would like to sell more FMS but finds customers reluctant. Initial investments are large, and senior customer executives must approve such purchases. So the Tokyo office has been set up with plush sitting rooms and a fancy auditorium so that customers' executives can be wooed in style. A showroom of machines used to be enough, since the buyers of lower cost single machines were engineers who just drooled and bought.

Design of FMS

An FMS is a group of 5 to 15 NC machine tools connected by a parts conveyor. Each machine is equipped with many cutting tools and can change tools automatically. Parts to be machined typically visit several of the machines in a system in order to receive all the required cuts. Each part may have widely differing cutting requirements, which the machines accommodate by using their NC and tool-changing capabilities.

Mazak is such a heavy user of FMS that it is much more sophisticated about their proper use than most of its customers. Deciding what range of product types and production volumes is most suitable for FMS has been the main technical and sales challenge since the idea was born at Cincinnati Milacron and the University of Stuttgart in the middle 1960s. The goal was to meet the needs of diverse manufacturing that lies between mass production of almost identical parts and low volume piecework production of single individuals. Economics and efficiency completely

determine the choice since any of the methods is technically capable of making the parts.

However, the choice has never been easy to make and is a subject of intense ongoing research at universities and head-scratching at machine tool makers and users. The FMS easily solved the first-order problems faced by the alternatives. NC made it possible to change from one part to another, which ordinary mass production machine systems cannot do. But stand-alone machine tools, NC or not, are utilized only about 5% of the time since the machine bed is used to set up fixtures and cutting tools for the next part. The FMS solved this by permitting fixtures and tools to be set up in a separate facility equipped with good measuring tools.

With the first-order problems solved, the FMS now faces second-order problems that make the difference between economically successful and unsuccessful installations. The major issues are keeping all the machines busy when a variety of parts is moving through the system and keeping the machines provided with all the different cutting tools that such a range of parts needs. A poorly designed system will have too many machines, some of which are idle, or too many parts waiting for a machine to become available. Typical economic criteria include cost per part, including labor cost and payback of the initial investment and, in Japan, return per square foot since land is so expensive.

The traditional approach in research and most FMS makers was to look for scheduling methods that would sequence the parts into the system so that workloads on the machines were balanced. Another approach was to develop FMS Another approach was to develop FMS design software that would survey candidate sets of parts and decide the right number and mix of different kinds of machines that should comprise the system. The criteria were that all the required cuts could be made, there was space for all the required tools, and all the parts could be processed in the

required span of time. Such problems are typically solved using complex math programming methods (Ref 11).

When the required number of tools could not be made to fit in a machine's tool storage racks, "tool management systems" were proposed. Since a part could need 10 or 20 tools, the logistics of tools far exceed the logistics of the parts themselves. Tool management systems thus can cost more than they are intended to save.

As far back as 1981 Mazak took a completely different approach toward FMS for in-house use. It decided to make partially specialized FMS comprising only three or four machines. These machines were chosen to be identical and were capable of machining a small set of parts, and perhaps only a small fraction of the cuts those parts needed. Scheduling and sequencing problems essentially disappeared. A part entered the system, visited one machine where it got all its work done, and left. It then visited another small system and received more cuts.

The range of required tools was limited by the "given tool method." That is, the part designers were given a stable of tools to use and told to design the parts so that set of tools would be sufficient. For many FMS, this method eliminated the tool logistics problem. For other situations, group technology was used to find a group of parts that used 80% of the defined tool set for a particular FMS, and a tool management system was used to provide the rest. Portions of this story appear in Reference 12.

These two efficiencies have permitted Mazak to employ FMS very effectively in-house without needing solutions to the long-term scheduling and tool management problems. However, the discipline required to use the "given tool method" cannot be forced onto customers, so the easy design and highly efficient FMS operation achieved by Mazak is not always available to

customers.

CAD, CAE, CAM

The company relies on CADAM and microCADAM, which runs on about 72 total terminals, of which a small number are IBM 5080 graphics terminals. Most of the CAD work is drafting, preparation of 2D drawings for part manufacture, making shop floor instructions and user manuals, and so on. There is essentially no CAE, in spite of obvious potential applications. It is up to Awane in his new post to introduce CAE.

Design Methodology for Machine Tools

The rhythm of the machine tool industry is driven by the occurrence of the major machine tool shows. These occur every 2 years in Tokyo, Chicago, and Hannover, effectively meaning a show every year or less. New machines cannot be created that fast, and the typical cycle is about 2 years. Totally new machines or technological innovations take longer. The shows are used to get customer input, look over the competition, and show your latest. In addition, at Mazak, most designers have visas up to date for most countries where the company sells and are ready to fly the moment a customer needs something. Customer input and minor changes are thus the main forces governing typical design cycles.

A major design strategy at Mazak, and probably most other companies, is "series design," meaning a series of machines based on one principal design with many variations such as number of tool storage places, size of bed, and so on. New designs are apparently not too hard to create within a series family. Information about existing machines, their drawings and their performance, is kept in computer files. Five mechanical and one electrical engineer can turn out a new lathe design in 2 years. Improvements may include faster tool

The integral spindle-motor took five engineers 2 years to design, during which five prototypes were made. A similar one can now be turned out in a manmonth (200 hours). The main problems in such units are dissipating the heat from the motor and obtaining the correct preload on the bearings. Motor heat will cause the spindle rotor to distort, causing vibration and poor machining accuracy. Incorrect preload either causes poor accuracy or low bearing life.

Predictions of heat, distortion, vibration, bearing wear, and so on are obvious candidates for CAE. However, Mazak does none of them. Motor heat

and its effects are predicted by the motor and its effects are predicted by the motor manufacturer using its own CAE, and Mazak merely designs a cooling system to take away the heat. Bearing design is done at Mazak and checked by the bearing manufacturer, who has extensive CAE for this purpose. This method of farming out the hard parts is used often in Japan.

Advanced bearings such as magnetic or ceramic are used sparingly or not at all. Magnetic bearings are the subject of some Mazak-sponsored university research, while ceramic balls are used research, while ceramic balls are used with steel races in some high speed applications.

Integral motor-spindle design is now so well understood that a routine has been established. The overall diameter of the rotor is decided by the size of chuck it will hold and the size of any hole inside the spindle. Spindle speed, motor horsepower, and required rigidity also contribute to sizing the shaft diameter. The front bearings and their preload are chosen almost straight from the handbook to support cutting loads. Required horsepower determines motor diameter, which determines the overall size and length of the spindle. The rear bearing is chosen to hold up the shaft, nothing else. This preliminary

Assembly of such a unit is driven by the need to install the rotor on the shaft and achieve the bearing preload. Balancing the rotor also is important and must be done at the right time during assembly. Mazak's engineers did not feel that there was much room for flexibility or innovation in this assembly sequence.

Cost estimating of new or series designs is important because the market and competitors often set prices. Their main cost estimating technique is to consider four factors: cost of purchased parts, amount of material (usually measured by its weight), machining costs, and assembly costs. The first two dominate the final total and are calculated very carefully using past data. The other two fluctuate too much to be of great use. CAD does not play much of a role in cost estimating except that CADAM can calculate volumes of parts easily. Cost of new entries in a series is determined by altering the data from the parent machine, which has been in production for a while.

Future Computer Design Needs

Mazak is unclear about what it needs in the future. It has a mild anticomputer frame of mind in the design department and relies on its experienced people, some of whom have been with the company decades. One identified need is to calculate machine bed and column deflections under cutting loads and (at my suggestion) under thermal distortion. Other companies sell machines with built-in temperature sensors and heaters that deliberately introduce compensating distortions to keep the machine accurate. Such companies must be ahead of Mazak in applying computers to design.

Mazak also would like an automatic design system that would take in a set of specifications for a machine tool and spit out a complete design. Such statements are heard at other companies and are not totally whimsical. However, they are totally out of step with the state of the art. Prof. Kimura notes that many companies do not have clear plans or coherent explanations of what they want. Visiting them every month for a year is often not sufficient to figure out what they are thinking, even when the company is Toyota, whose thinking would seem to be fairly systematic.

PROF. NORIO OKINO, KYOTO UNIVERSITY

26 July 1991

Background

In the 1960s and 1970s Prof. Okino developed one of the first solid modelers based on constructive solid geometry, called TIPS. He still works on CAD/ CAM and solid modeling but his main interest recently has been what he calls "bionic manufacturing."

Bionic manufacturing involves two elements: self-governing behavior and object-oriented hierarchical structures. Okino has constructed a conceptual model of computer-integrated manufacturing (CIM) that begins with all of society at the top and extends downward by subdividing object classes until the lowest levels are reached somewhere in a factory.

Each level is composed of elements that have the same structure in principle. These are called "modelons." Each modelon contains a common memory and a number of processes or methods that operate using that memory. In addition, a number of lower level modelons are attached to this memory. Modelons are independent actors, like daemons or, in UNIX, processes. Each

modelon looks out for the conditions under which it could run; if they are satisfied, the modelon runs and deposits its results in the common memory it is attached to. Modelons are also like methods in object-oriented programming in the sense that they can send messages to each other, activate each other, etc.

A major feature of this structure is, as said above, that each modelon acts autonomously. What the convergence problems of such a structure might be Okino does not say.

Several topics are being studied under this structure. I was shown software demonstrations of two: a robot modelon that interacts with prismatic peg and hole modelons to determine where to grasp the peg for the purposes of putting it in the hole and a hidden line removal algorithm in which modelons representing each of three prismatic parts inform each other of the locations of their vertices. We discussed a third for which there was no demonstration: a shaft modelon made of three shape features, each of which is represented by generic feature modelons. All the work is being done on Apollo Domain 10000 workstations.

1. Robot Grasp Planning: This is the recently completed work of Dr. Watabe. In this system, there are several software modelons representing the parts, the robot, and the environment. The the robot, and the environment. The user requests a task, such as that the peg be put in the hole. The top level modelon broadcasts this request to the lower ones, all of which attempt to respond with a solution. Finding that they cannot, they broadcast in turn to any of their subelements whose responses they need or could use, in a divide and conquer approach. This is repeated at lower levels recursively. For example, the robot needs to know several sets of faces on the peg, such as parallel faces, parallel faces free-tograsp before inserting, the same after

inserting, and so on. The peg and hole modelons respond. I could not tell how the assembled state was established for the purpose of identifying the free faces. Perhaps the user constructed it as a way of posing the problem.

There is no agenda structure in this search. Watabe had not heard the terms "forward chaining" and "backward chaining" before. (These are common techniques in expert systems for solving the kind of problem he is working on.) Such searches commonly are not very efficient since there is no gradient to follow and no metric to score how close one is to finding a solution.

2. Hidden Line Removal: In this demo, three prismatic blocks intersect each other. Before the hidden lines are removed, one cannot see what is what, even with three views, because several of the edges appear parallel and the front-back optical illusion interferes. These facts make the demo more interesting but of course have nothing to do with the intrinsic difficulty of the problem. Each block has its own hidden line removal method and seeks information from the other blocks concerning where its lines enter or leave their boundaries. Information is passed around this way for several minutes before a solution appears on the screen.

Okino points out that speeding up the algorithm or competing with existing algorithms is not the objective, but

rather it is to understand modelons.

In this regard, it is interesting to review the discussion we had about the shaft made of three shapes: a plain cylinder, a conical cylinder, and a threaded cylinder, all coaxial. Each shape is supported by a generic feature that contains methods for drawing the shape, calculating its mass, and a process plan for how to make it. Presumably the process plan for the shaft is made by combining the process plans for the three supporting features.

I asked a basic question that underlies the problems in all process planning of this kind: what do you do in regions where the plans touch or intersect and presumably interact? That is, how does one compose process plans from subplans? He agreed that this was a challenging question and replied that perhaps one must declare the shaft itself to be the primitive element. A student is beginning to work on the composition problem.

Okino's reply indicates that there are many problems yet to be solved by this approach. The value of having generic elements at the leaves of the structures is clearly large and would give the approach considerable power. Requiring each specific shaft to have a representation of its own is not efficient. However, he is a software person rather than an engineer and is pursuing the structural issues first.

Prof. Kimura notes that objectoriented structures are good for some kinds of data but not others, especially those that have strong interactions as well as, or instead of, hierarchical, decomposable structures. Mechanical design and manufacturing may not be separable enough to permit objectoriented approaches to cover them completely.

NIPPONDENSO 29-30 July 1991

Background

Our hosts for this visit were Mr. Fukaya, General Manager of the Production Engineering Department; Mr. Tsuchiya, Fukaya's R&D Manager; a young engineer Mr. Harada; and at dinner Mr. Ito, the Executive Managing Director of Production Engineering.

Nippondenso Co. Ltd. is a former subsidiary of Toyota Motor Co. that makes a wide variety of automotive components, such as alternators, motors and actuators, air conditioning systems,

engine components and controls, radiators, dashboard displays, brake control systems, and so on. It has manufacturing plants worldwide to satisfy many automotive manufacturers. Handling the wide diversity of product models and responding to the Just in Time (JIT) ordering system philosophy have heavily affected how Nippondenso designs and manufactures products.

Nippondenso is a high technology company. Basic and applied research cover materials, vacuum apparatus, semiconductor fabrication methods, ceramics, robotics, vision systems, factory automation software, simulation systems for testing driver reactions, and CAD/CAM. Major design thrusts over the past 10 to 15 years include "managing diversity," designing new products faster, overlapping design tasks while managing risk, and dramatically reducing the size and weight of products while increasing quality and performance.

The company has 41,000 employees. Of these, about 5,000 are design engineers and 1,500 are production engineers. At the new R&D center where the visit took place, there are currently 150 researchers. A major characteristic of Nippondenso, which I have noted in many previous reports dating back to 1977, is that it makes much of its own automation equipment and nearly all of its 3,500 robots. This commitment to manufacturing equipment excellence is one part of its commitment to manufacturing excellence in general.

The typical working year at Nippondenso and in Japan generally is 2,200 hours, compared with 1,800 in the United States. The Government is trying to get this reduced to 2,000 by 1994 and 1,800 by 1996. Each Japanese company is attempting to reach this target, facing various problems.

Nippondenso characterizes its products as follows:

• Quality first

⚫ Wide range of product variety

An

important feature of Nippondenso is an obvious long term enterprise-wide strategy for how to grow the company into a master of manufacturing products with these characteristics. Nippondenso has evolved a systematic approach to managing design processes, designing carefully to support JIT operations, and developing larger and larger systems of automation. In the 1950s they had "spot” automation (what we would call islands); in the 1960s they had lines; in the 1970s "areas," meaning presumably several lines of the same type or several lines connected; in the 1980s and early 1990s "cube" or "totality." Such increasing automation creates serious dangers for a company whose customers switch specifications, alter model production volumes, demand instant response to orders, and increase variety of products.

Nippondenso has only gradually realized how deep the dangers can be and has instituted several procedures for combatting them. These include simultaneous product-process development, a classification of levels of necessary flexibility in production, and a classification of degree of innovation in design projects.