Kos Ishii
Stanford University
Department of Mechanical Engineering
Stanford, CA
Cheryl Juengel and C. Fritz Eubanks
The Ohio State University
Department of Mechanical Engineering
Columbus, OH
ABSTRACT
This study develops a method to capture the broadest customer preference in a product line while minimizing the life-cycle cost of providing variety. The paper begins with an overview of product variety and its importance in overhead costs: supply chain, equipment and tooling, service, and recycling. After defining the product structure graph as a representation of variety, the paper introduces an approximate measure for the customer importance and life-cycle cost of product variety The cost measure utilizes the concept of late point identification which urges standardization early in the manufacturing process and differentiation at the end of the process. The variety importance-cost map allows engineers to identify cost drivers in the design of the product or the manufacturing system and seek improvements. The refrigerator door example illustrates the concept. On-going work seeks to validate and enhance the method with several companies from different industries.
1. INTRODUCTION
1.1 Customer Preference and Product Variety
When designing a new product, there are many product options that a manufacturer must consider. The customer preference among product features necessitates the variety. Typical options are the door configuration of a refrigerator or the engine type in a car. The designer needs to know what options are most important to customers, so that an appropriate amount of attention is given to those options in the design of the product and manufacturing process. Quality Function Deployment (QFD) (Clausing, 1991; Bascarn, 1991; Steiner, 1991) addresses the identification and analysis of customer values. One can utilize QFD to characterize the product structure suitable for a company and create a hierarchical graph that enumerates all the product features in order of importance to the customer. Other work on customer preference charaterization include Locascio and Thurston (1994) and Haugue and Stauffer (1993).
On the other hand, engineers must design the product line so that it can be efficiently manufactured. The product and process must then be designed so that they readily accommodate the variety on models with different features. The product structure graph, a tree describing the product variety, will help engineers to determine what constitutes the "standard" model, and to design a production line that efficiently produces the variety. The designer must determine which components to standardized and which subassemblies to modularize. Then, the manufacturing facility can best utilize its resources to attract the broadest customer base. This strategy is related to the concept of `mass customization' (Shingo, 1981).
Figure 1 is a plot of total cost vs. degree of integration of product architecture (Steiner, 1994). Product architecture is measured by the number of functions per component in the product. The term "modular architecture" describes a low number of functions per component, and "integral architecture" describes high functions per component. Ulrich and Eppinger (1994) point out that as functions per component increases, volume related costs decrease, while complexity related costs increase. The complexity related cost includes post manufacturing costs related to life-cycle service (Eubanks and Ishii, 1994) and product retirement and material recycling (Ishii et al, 1994). At some point, the two cost curves intersect, which should be the point of lowest cost. Questions to be answered about this theory include: "What should be included in the cost of complexity?", "What are the functions relating cost to architecture?", and "Is there an optimal level of integration which results in the lowest cost?". For example a refrigerator door has a great deal of product variety. This stems from customer choices, including different shelf types, with or without countermaker door, with or without ice dispenser, and color.
Figure 1. Volume related cost vs. complexity (variation) related cost
Designers can greatly benefit from a systematic methodology that allows them to estimate the cost of providing variety to attribute the cost to design and manufacturing decisions. Designers can use this information to balance the design with the customer preference and simultaneously minimize the cost of variety and achive profitable product line structure.
This paper seeks to first define the product structure graph, then develop methods to quantify the graph and to utilize it for product and process design. Section 2 details the importance of design for product vareity (DFPV) method. Section 3 introduces a representation scheme for product line structure, and section 4 describes the proposed measure of variety significance and cost. Section 5 illustrates the use of the measure, and section 6 concludes the paper.
2. PRODUCT VARIETY: CUSTOMIZATION AND COST
2.1 Mass Customization
Until the early 20th century, products were handmade by highly skilled craftsmen. Each part had to be individually fitted to the other components, since there were no standard gauges by which to manufacture the parts. The resulting products were all unique, even those made from the same blueprint, due to the variations in component dimensions. The production costs with this method were high, and quality was inconsistent.
In contrast to the craft production system, the goal of the mass production system is to manufacture products at the lowest cost possible, by using interchangeable parts, specialized machines and division of labor. The products manufactured by this method are identical and produced very high volume. The most important features of this system were interchangeable parts and simple methods to attach them together. Henry Ford made the parts interchangeable by enforcing a standard gauging system throughout the manufacturing process. The next change made by Ford was to have each worker perform one step of the assembly process, instead of assembling large portions of the car. The cars remained stationary and workers walked around from car to car. In 1913, Ford introduced the moving assembly line, in which the car moved to the worker, who stood in place and repeated the same assembly operation over and over again. This system made the workers interchangeable, since anyone could be trained for the job in a few minutes. This division of labor spread to the engineering departments as well. This system has become unwieldy and inefficient for today's incredibly complex products.
The goal of mass customization, is to combine aspects of the craft and mass production systems, to create a high volume of varied products, with high quality at low cost. This production method is best shown by the Toyota Production System, developed mainly by Taiichi Ohno beginning in the late 1940s. This system has been well documented in several sources (Monden, 1983; Shingo, 1981; Womack et al, 1990). Toyota achieves high quality by allowing every worker to stop the production line if a defect is found. Teams of workers discover the reasons for the defect using a problem solving system called "the five whys", also developed by Ohno. The workers then correct the problem so that the defect does not occur again. Once Toyota implemented this system, workers soon eliminated flaws so that the production line now hardly ever stops. Workers feel more involved in the creation of the product and are more highly motivated than typical American mass production workers. Inventory levels are very low compared to those in mass production facilities, allowing for rapid reaction to changes in the marketplace and more customized products.
2.2 Late Point Identification
As pointed out earlier, the product and process must then be designed so that they readily accommodate the necessary variations. The key is to identify the "standard" model and strategically design the product and process that leads to short in-process time, low inventory, and high product line flexibility. The concept of "late point identification" (Steiner, 1994) tells designers to implement the variety towards the end of the manufacturing process and standardize elements that require long lead time. Modularity of assemblies is another useful tool here. The term "modular architecture" describes a low number of functions per component, while "integral architecture" describes high functions per component. Ulrich and Eppinger (1994) point out that as functions per component increases, volume related costs decrease, while complexity related costs increase. The complexity related cost includes post manufacturing costs related to life-cycle service (Eubanks and Ishii, 1994) and product retirement and material recycling (Ishii et al, 1994). Part commonality will also play a role here (Gerchak and Henig, 1989). Product and process designers must find the optimal balance among all the costs.
3. REPRESENTATION OF PRODUCT STRUCTURE
3.1 Product Structure Graph
The product structure graph takes as its inputs the customer wants as determined in the House of Quality (Clausing, 1991). Specifically, the graph focuses on the choices of different product features offered to customers. The overall goal is to optimize the manufacturing process for producing the variety in the given product line, while minimizing the investment required. The output of the graph is the product structure that designers should employ in the design of the product (Figure 2).
The root node of the graph is the product name, for example, door. This node branches into the choices available for one given type of variety, for example, size. Each node then branches into the choices available from it for another type of variety, and so on. The leaf nodes show exactly how many different products there are in the product line.
Figure 2. Product Structure Graph Information Flow
3.2 Product Structuring Examples
Figure 3 shows the product structure graph for the fresh food door of a refrigerator. The first set of branches lists the sizes of doors to produce, by the size in cubic feet of the refrigerator the door is attached to. The next variation is whether the door will have a countermaker (abbreviated CMKR) or not. "Y" (yes) indicates that it does have a countermaker, "N" (no) that it does not. A countermaker is a smaller door within the door to allow access to one shelf of the refrigerator without opening the whole door. Note that this option is not available for the 20 cubic feet model. There are three types of shelves: fixed shelves (FIX) cannot be moved, modules (MOD) that have adjustable positions, and countermaker modules (CM MOD), which are modules adapted to fit with the countermaker. The only type of shelf available for the 20 cubic feet model is fixed.
For the 22 and 24 cubic feet models, if there is a countermaker in the door, the shelf type must be countermaker modules. Otherwise, the customer has a choice between modules or fixed shelves. Finally, each of the shelf type node branches into the three available colors: Black (B), White (W), and Almond (A). The graph shows that there are 27 different fresh food doors.
Figure 3 . Fresh Food Door Product Structure Graph
4. MEASURING IMPORTANCE AND COST OF VARIETY
4.1 Motivation
The product structure graph clarifies the product variety in a hierarchical tree and helps engineers to focus on pertinent features and to prune unnecessary varieties, i.e., focused customization. The graph also helps the designer to see which components can be standardized, and which subassemblies modularized, to most efficiently produce a wide variety of products.
Whereas the graph and the late point identification strategy provide a qualitative guide to design for variety, engineers need a more systematic method and a quantitative measure of the cost of providing the variety. A cost model is essential since only with a quantitative measure can designers balance the variety issue with other manufacturing concerns such as assembly and fabrication costs. Engineers also need a method to attribute the cost of variety to product and process design decisions. Then, the engineer can compare the measure of variety importance with variety cost and either prune the option or improve the product/process design to efficiently implement the variety. Such a measure must be applicable to the early stages of product development when engineers may not have detailed design information.
Our proposed measure is of the form 
, where X
characterizes how important the choice is to the customer, and Y
represents the cost to produce the variety. The refrigerator example revealed
three main factors that affect the cost of providing variety: the number
of options, how late in the manufacturing process the variety is implemented,
and how "painful" it is to change over from one variety to another.
We proposed a very rough measure of cost of variety Y on a scale
of 
as:
(1)
where:
4.2 The basis for DFPV
The goal of the product structure graph is to encourage variety, but
only where important to the customer, and cost-effective. The 
ordered-pair provides a measure of whether the choice is
important enough to the customer to justify the cost. A design team can
use QFD to rate the product features in terms of how important they are
to customers (the section called the "WHYs" of the House of Quality.)
This measure is a number between zero and ten, with a higher number indicating
a greater importance. We may adopt a other methods such as conjoint analysis
(Elrod et al, 1992) to measure customer preference. While the number drawn
from house of quality is rather crude, we used this method for our intiatial
study due to its simplicity.
The three cost factors, 
, 
, and 
, quantify the factors contributing to the cost of manufacturing
variety. They are "discounts", with better designs getting a greater
discount off the cost of manufacturing. The variable 
represents the
number of options available at the level of a node. The fewer variations
there are, the less it will cost to manufacture them. While more sophisticated
evaluation schemes are possible, this paper adops a rough assignment of
to 
. The more options, the smaller
the discount: 
is 
when there are
more than five variations, 
for three to five variations,
and 1 for less than 3 variations (Figure 4).
Figure 4. Assigning Values to 
The stage in the manufacturing flow where the variation occurs affects
the cost in the following manner. When an option occurs early in the manufacturing
process, the differences must be managed through a greater number of subsequent
operations, increasing the complexity and cost of the process. The variable
represents the stage of the process where the variety
develops. Whereas a finer evaluation is possible, this paper divides the
manufacturing flow into three stages: component, subassembly, and final
assembly. If the option separates at the component stage, 
is 1/3. If the
option separates in the subassembly stage, 
is 2/3. When
the variation occurs in final assembly, 
is 1. A higher
number indicates lower cost to manufacture the variation (Figure 5):
Figure 5. Assigning Values to 
quantifies how "painful" it is to make the
changes required to manufacture the different variations in the product
line. Assigning a value to 
also requires candidate
process plans, for each of the variations. The main determinant of the value
of 
is the amount of time it takes to change from producing
one type of variation to producing another. The longer it takes to make
this change, the more costly it is to produce that variation. Again, the
higher number indicates lower cost. If the change is very quick, on the
order of one minute, then 
is 1. If it takes somewhat
longer, around one hour, then 
is 2/3. An example
of this could be changing a die, or a paint color. If the change requires
a different plant layout, an additional production line, a different vendor,
etc., and takes approximately one day or more, then 
is 1/3 (Figure
6):
Figure 6. Assigning Values to 
The number of divisions, and the values assigned to each 
can be increased
or decreased to better describe a particular design. For example, 
could be further divided to include 1/2 hour, and 4 hours
(1/2 day), with the values 1/5, 2/5, 3/5, 4/5, and 1. If a cost metric is
available, 
could be computed as a continuous normalized
cost.
4.2 The variety importance and cost measures
In the ordered pair, 
, X measures the customer
importance of a particular option. The The value of X assigned to
the product structure graph is normalized to the interval [0, 1] by dividing
the QFD importance measure, which is between 1 and 10:
(2)
Y rates the cost of manufacturing the variation on a scale from
0 to 1, where 0 means low cost and 1 represents very high cost. The value
of Y is calculated by multiplying the discounts 
, 
, and 
and subtracting this product from 1 (Equation 1):
The product of 
, 
, and 
gives the total discount for the design. This measure takes
a similar form as Thurston's customer preference utility measure (1990).
This paper uses this form as the estimate of the cost of providing the variation
in question.
The parameters X and Y are calculated for each node in the graph. By following the path of choices that leads to a particular model, averaging the X values gives an importance measure for the model, and averaging the Y values gives a cost measure for that model. Once the importance and cost measures for each model have been calculated, the model measures are averaged to give an overall measure for the entire product line. The averaging of the measures is rather crude but serves our purpose for this preliminary study. Our on-going work is investigating a more appropriate method of aggregating the measures.
5. EXAMPLE APPLICATION OF DFPV
Figure 7 shows the product structure graph for a freezer door, with the
measure for each node. The first branch shows the different
sizes of doors available. The next branch is whether the refrigerator has
an ice dispenser, "Y" for yes, or "N" for no. The third
node is for shelf type, "MOD" for modular shelf that customers
can adjust, "ID MOD" for modular shelf that accommodates the ice
dispenser, "FIX" for one-piece thermoformed shelf that customers
cannot adjust. Finally, each of the ice shelf nodes splits into the three
possible colors. Given these choices, there are 36 different freezer doors.
The X importance measures are the result of an informal survey.
The 
values are shown in the sample spreadsheet in Figure
8. In the upper left corner, the size 27 box shows 
as 2/3, approximated
by 0.67. This is because there are four options at the level of this node,
and 
is 2/3 when there are from 3 to 5 options. 
was given a value of 1/3 because the components have to
be of different sizes to make different size refrigerators. Thus, the option
occurs at the component stage of the manufacturing flow. The changeover
between sizes takes approximately one hour, so 
is 2/3. The 
values for the rest of the nodes were assigned similarly.
The Y column calculates 1 - 
. The Model Average
column averages the 
measures over the path leading
to that model. For example, the top model average measure is for the black,
modular shelf, no ice dispenser, size 27 refrigerator freezer door. The
X and Y model average columns are then averaged to give the
overall 
measure for the product line at the bottom right corner
of the spreadsheet.
Figure 8. Product Structure Graph Ratings
The overall 
measure for the freezer door shown
in Figure 8 was (0.70, 0.81). This places the product line into the "Improve
Design" region of the Cost-Importance plot. The size, shelf type, and
color nodes all have high Y measures. One way to reduce the Y
measure for the shelf type is to make all adjustable shelves instead of
offering a choice between fixed and adjustable. The product structure graph
for this design is shown in Figure 11. This design reduces the number of
variations at the shelf type level, and 
for the shelf
type nodes goes from 2/3 to 1. Assuming that this design change makes it
possible to thermoform all the door liners on the same mold and just drop
in different shelves in final assembly, 
goes from 1/3
to 1. 
remains 2/3 because the changeover would still take
approximately one hour. With these changes, the Y measure for the
shelf type nodes decreases from 0.85 to 0.33, and the overall Y measure
decreases from 0.81 to 0.68. The shelf node is then in the "Good Design"
region, but the overall product line still could use some improvement. The
next area the design team should focus on improving is either the color,
or the size variation.
Figure 9. Spreadsheet Computation of the Variational Cost Measure
Figure 10. Cost-Importance Plot
Figure 11. Product Structure Graph for Redesigned Freezer Door
6. CONCLUSIONS AND FUTURE DIRECTION
This paper proposed Design for Product Variation (DFPV) as a methodology to capture the broadest customer preference in a product line while minimizing the life-cycle cost of providing the product variety. The major points of the paper were as follows:
The measure can lead to improvements in both product and process design. Engineers can improve the layout design by addressing assembly efficiency of multiple models to be produced on the same manufacturing line or seek standardization of assembly modules and components. If the customer importance of variation justifies the investment, engineers can use the proposed cost measure to identify the target areas of innovation for flexibility and agility. Examples include quick die change in the automotive body stamping process, automation capable of on demand program changes, strategic manual assembly, and flexible jig and fixture designs.
Our industrial collaborators stress the enormous benefits of capturing variational cost. The total impact goes beyond manufacturing cost. Factors affected include: streamlined suppliers, reduced inventory, fewer work in process, faster process time, reduced service parts, streamlined service logistics, reduced drawings, etc. The authors have used examples from automotive and appliance industries in developing the proposed measure, but we have recently found that DFPV is extremely relevant in the computer and electronics industry as well. In the electronics industry, the demand for customization is even greater requiring an intelligent modularization of the products. The challenge also comes from the rapid advances and short product life-cycle. An engineer from the electronics manufacturing machinery company cites that they must introduce new products every year to respond to the advances in semi-conductor devices. Such rapid advances necessitate careful attention to product variation to ensure not only low manufacturing cost, but also service cost and to promote recycling and reuse at the end of the machine's useful life.
While the authors believe the proposed methodology is useful in its current form, the validation of its effectiveness requires substantial work. Our cost measure is empirical, and needs a quantitative calibration. On going work includes:
The fifth challenge is important. The decision on the product line structure usually lies not with a design engineer, but with people higher up, like the director of a product division. A product line structure involves several teams responsible for various design options. Further, not every option or variation will be under development simultaneously. We intend to collaborate with management experts in pursuing this issue.
ACKNOWLEDGMENTS
The authors sincerely thank Dr. Mark Steiner of GE Appliance who gave us insights into the intriguing issue of product variety. We also thank Mr. Hank Harada of Nippondenso who shared us his experience with manufacturing of multiple models. Funding for this work came from National Science Foundation Design and Manufacturing Division and General Electric.
REFERENCES
Bascarn, Eduardo (1991). Design Through Selection: The Use of QFD in the Attribute Generation Process, Design Theory and Methodology , ASME DE-vol. 31, pp. 195-200.
Clausing, Don (1991). Concurrent Engineering, Design and Productivity Int. Conf., Honolulu, Hawaii, February 6-8, 1991.
Elrod, T., Louviere, J. and Davey, K., "An Empirical Comparison of Ratings-Based and Choice-Based Conjoint Models," Journal of Marketing Research, Vol. XXIX, 1992, pp. 368-377.
Eubanks, C. F. and Ishii, K. (1993), AI Methods for Life-cycle Serviceability Design of Mechanical Systems, Artificial Intelligence in Engineering (Elsevier), Vol. 8, pp. 127-140.
Hauge, P. and Stauffer, L. (1993). ELK: A Method for Eliciting Knowledge from Customers. Design Theory and Methodology, ASME DE-vol. 53. pp.73-81.
Homem de Mello, L. and Sanderson, A. (1991). Representations of Mechanical Assembly Sequences, IEEE Transactions on Robotics & Automation, vol. 7, no. 2, April 1991.
Homem de Mello, L. S. and Arthur C. Sanderson (1990). AND/OR Graph Representations of Assembly Plans, IEEE Trans. on Robotics & Automation, vol. 6, no. 2, April 1990.
Ishii, K., C.F. Eubanks, P. Di Marco. (1994), Design for Product Retirement and Material Life-cycle. Materials and Design. Vol. 15, No. 4, pp. 225-233.
Krinninger, A. and Clausing, D.P. (1991). QFD for Production, Working Paper of the Study QFD For Production, Laboratory for Manufacturing and Productivity of MIT, Cambridge, August 1991.
Locascio, A. and Thurston, D. (1994). Quantifying the House of Quality for Optimal Product Design. Design Theory and Methodology, ASME DE-vol. 68. pp.43-54.
Monden, Yasuhiro (1983). Toyota Production System, Industrial Engineering and Management Press, Norcross, GA, 1983.
Pine, B. Joseph II (1993). Mass Customization: The New Frontier in Business Competition, Harvard Business School Press, Boston, MA, 1993.
Schubert, Michael A. (1989). Quality Function Deployment: A Comprehensive Tool for Planning and Development, IEEE Transactions no. 1498, National Aerospace and Electronics Conference, vol. 4, New York, 1989.
Shingo, Shigeo (1981). Study of `Toyota' Production System from Industrial Engineering Viewpoint, Japan Management Association, Tokyo, 1981.
Stalk, George Jr. (1988). Time - The Next Source of Competitive Advantage, Harvard Business Review, 1988 7-8.
Steiner, Mark W. (1991). In Search of QFD, Design Theory and Methodology, ASME DE-vol. 31, pp.191-200.
Steiner, Mark W. (1994). Product Line Structuring. Presentation notes at GE Quality Symposium, GE CR&D, Sept. 1994.
Thurston, D. (1990). Subjective Design Evaluation with Multiple Attributes. DTM, ASME DE-vol. 27, pp.355-361.
Ulrich, Karl T. and Steven D. Eppinger, (1994), Methodologies for Product Design and Development, McGraw-Hill, 1994.
Womack, James P., Daniel T. Jones, and Daniel Roos, (1990), The Machine that Changed the World: The Story of Lean Production, Harper Perennial, New York, NY, 1990.
