Kosuke Ishii, Associate Professor
Department of Mechanical Engineering
Stanford University
Stanford, CA 94305
September, 1994
(Revised November, 1994)
A paper submitted to ASME Journal of Mechanical Design
ABSTRACT
Life-cycle engineering seeks to incorporate various product life-cycle values into the early stages of design. These values include functional performance, manufacturability, serviceability, and environmental impact. We start with a survey of life-cycle engineering research focusing on methodologies and tools. Further, the paper addresses critical research issues in life-cycle design tools: design representation and measures for life-cycle evaluation. The paper describes our design representation scheme based on a semantic network that is effective for evaluating the structural layout. Evaluation measures for serviceability and recyclability illustrate the practical use of these representation schemes.
1. INTRODUCTION
Design for manufacturability (DFM) has proven itself as a key concept in competitive product development. DFM helped many US manufacturers improve product quality, reduce cost, and shorten development cycles. More recently, life-cycle engineering design has emerged as an extension to DFM that covers not only manufacturability, but issues related to the entire product life-cycle (Figure 1). With increased attention to the environment, the definition of life-cycle now covers not only that of a single product, but resources that result in the life-cycle of a manufacturer's line of products: solid materials, fluid and gas emissions, and energy.
Figure 1. Product life-cycle
Life-cycle engineering seeks to maximize a product's contribution to the society while minimizing its cost to the manufacturer, the user, and the environment. We focus on design and manufacturing decisions that significantly impact the product life-cycle. Most researchers agree that decisions made during the early stages of design determine more than 80% of the life-cycle cost. Among the most significant issues are the structural layout of a product and the materials used. Life-cycle engineering requires designers to estimate the life-cycle cost and attribute it to the design and manufacturing decisions. This paper focuses on 1) the current methodologies and tools in life-cycle engineering design, 2) significant research issues to further develop the field, and 3) the author's own research results over the past several years.
2. RECENT DEVELOPMENTS IN LIFE-CYCLE ENGINEERING
Many prior studies exist in the area of design for manufacturability (DFM). Perhaps the most successful methodology is design for assembly (DFA; Boothroyd and Dewhurst, 1983). Their computer program asks the user a series of questions about the handling, orientation, and insertion of parts during assembly, and evaluates the design in terms of the assembly time, its breakdowns, and assembly efficiency. Other prominent DFA methods include those of Westinghouse (Sturges and Kilani, 1992) and Hitachi-GE (Miyakawa, et al., 1990).
There is also a wealth of research on component design for producibility. Poli (1988) developed a methodology to evaluate a plastic part design. The key question is the part complexity: the number of geometry features such as ribs, bosses, snaps, and cutouts. The orientation of features is also important since it influences the number of axes of draw. Poli's methodology essentially gives an early estimate for tooling cost and molding cost. The natural extension of these programs is to incorporate the manufacturability concept in the computer aided design environment. Dixon's group (1986) applied AI technology to accomplish redesigns. The purpose of this class of programs is to monitor the CAD data as the designers develop their candidate designs, find if any of the design rules are violated, provide reasons for the flaw, and suggest remedies. Design for robustness has also targeted component designs. Taguchi (1993) has been instrumental in proliferating this concept, which seeks a design that is insensitive to uncontrollable noise such as manufacturing errors and operational conditions.
Life-cycle issues during the product ownership period have also attracted attention. Ownership quality not only affect warranty costs, but also has a major impact on product image and repurchase intent. Reliability design (Birolini, 1992) and failure modes and effects analysis (FMEA; Ormsby, et al, 1991) are traditional methodologies that identify potential weaknesses in the design. However, engineers must not only consider reliability but also address ease of service and simultaneously specify support logistics. Hence, design for serviceability (DFS) has attracted significant interest as a method to enhance product ownership quality (Gershenson and Ishii, 1992).
Recent years have seen a surge of work in environmentally conscious design and manufacturing. Life Cycle Assessment (LCA) is a broad methodology for identifying environmental burdens that arise from a product. The US Environmental Protection Agency (EPA, 1993) developed documents that address life-cycle concerns from raw material acquisition to final product disposition and include total energy use and pollution impacts. LCA seeks to minimize the environmental impact of the manufacture, use and eventual disposal of products without compromising product functions. So far, most LCA studies have focused on single material products such as disposable drink containers and diapers. For complex products such as automobiles and appliances, LCA is often too time consuming for designers to implement themselves. Allenby's methodology (1991), commonly known as design for environment (DFE), ranks various environmental issues pertaining to each life-cycle stage. His method provides a more qualitative evaluation of designs and is more applicable to early stages of design. Product take-back laws in Europe and the recyclability laws in Japan provide a more focused goal. Many researchers have focused on product retirement (Burke, et al, 1992; Marks, et al, 1993). The key is the "simultaneous" planning for post-life use of the product in the early stages of design, i.e., design for product retirement (DFPR).
Each methodology mentioned above brings benefits to engineering design. Figure 2 classifies the methodologies in terms of the applicable stage of product life-cycle (horizontal axis) and design cycle (vertical axis). Obviously, life-cycle design requires the combination of all the viewpoints (Alting, 1992). However, combining the use of all the tools is not trivial. Quality Function Deployment (QFD: Hauser and Clausing, 1988) is a powerful tool for relating customer requirements, functional specifications, product design, and process characteristics. Whereas QFD guides design teams in achieving the integration, engineers can further benefit from a more quantitative methodology.
Figure 2. Life-cycle design methodologies, product life-cycle, and design cycle
3. FUNDAMENTAL RESEARCH ISSUES
An integrated life-cycle design methodology must help engineers estimate the life-cycle implication of a candidate design, identify cost drivers, and facilitate simultaneous design of the product, the manufacturing specification, service logistics, and product retirement plan. Such a task is most effective at the layout design stage, at which time the design is still preliminary and many decisions are uncertain. The evaluation measure must be flexible enough to accommodate this uncertainty. Another requirement is that the methodology is easy to use and does not pose a significant additional burden to the engineers. We identified the following issues as our current research challenge.
1. Design Representation Scheme: A life-cycle evaluation tool requires a flexible set of data that contains pertinent information about the candidate design. Typical information required includes structural configuration of the components, their fastening methods, the material of the components, and their size. One must design a data structure such that engineers require very little time to specify the necessary information. The key is to find the smallest set of data that facilitates the evaluation for the entire product life-cycle. The author finds the use of a semantic network to be an effective representation scheme for the evaluation of layout designs.
2. Identification of life-cycle evaluation knowledge: Previous research in DFM has developed a significant knowledge-base and techniques in evaluating life-cycle costs. DFA and component producibility evaluation methods are well established. However, many life-cycle issues remain unexplored. Ownership quality as perceived by the customers is still unclear, beyond the failure frequencies and serviceability costs. Environmental compatibility is still vast and difficult to evaluate. One must package this knowledge in a form applicable to the evaluation of early designs.
3. The evaluation measure: The author considers the life-cycle cost of achieving certain functions to be the most useful evaluation measure. Further, if one can analyze the breakdown of the cost, engineers can use that information to improve the candidate design. Unfortunately, early design data do not provide accurate estimate of the life-cycle cost. Hence, one must devise a measure for estimating the life-cycle cost from the evaluation knowledge identified above. Again, for DFA, a wealth of research and validation studies have resulted in useful and sufficiently accurate evaluation measures. We must continue to refine these measures and develop new measures for other life-cycle costs to seek an integrated evaluation tool. The task of life-cycle product structuring also poses a challenge. We must develop a measure of cost to achieve product variations, and compare that with the importance of the variations in pursuing customer base and return in profits.
4. Simultaneous design for product life-cycle: The above research challenges address design evaluations. The author believes that these methodologies can be adapted to facilitate an environment for simultaneous design of products, the manufacturing specifications, service logistics, and retirement plans. For example, assembly evaluation often involves qualitative simulation or "walk through" of the assembly process. Serviceability analysis can also lead to efficient design of service logistics and manufacturing plans for spare parts.
The following sections describe our challenges to these research issues. We focus on a unified representation scheme that accommodates layout design evaluation for assembly, service, and retirement. For serviceability and retirement analysis, we have identified the necessary evaluation knowledge and developed our original measures of cost estimates and procedures to compute and analyze the measures. The procedures also facilitate simultaneous design of service logistics and advanced planning for product retirement.
4. DESIGN REPRESENTATION FOR STRUCTURAL LAYOUT
To facilitate the evaluation of layout structure, we propose the use of LINKER, a hierarchical semantic network comprising components and subassemblies (nodes) and the relationships between the nodes (links). Figure 3 shows the LINKER representation of a common drip-type coffee maker. To accomplish automated reasoning about the design, we must define both the syntax and the semantics of the network notation (Woods, 1975). Our current scheme targets the analysis for assembly, service, and product retirement. We currently use four types of nodes for the design description, defined as follows:
Figure 3: System represented as a hierarchical network
N1) COMPONENT: A design element that cannot be disassembled without permanent damage to the resulting pieces, or loss of intended function following reassembly with the resulting pieces.
N2) SUBASSEMBLY: A design element that can be disassembled into 2 or more other elements and performs its intended function following reassembly with the original elements.
N3) FASTENER: A design element whose intended function or purpose is to maintain an assembled configuration of 2 or more components and/or subassemblies.
N4) FASTENING PROCESS: An action or operation, either physical or chemical in nature, whose function or purpose is to maintain an assembled configuration of 2 or more components and/or subassemblies.
The component, subassembly, fastener and process data comprise part or material cost, removal time, installation time, tools and training required to perform the action, the name of the item or process, a user-defined part number or code, and the next higher assembly (if applicable). We currently use five types of links:
L1) COVERS: No physical connection exists between the two items, but the first item in the link must be removed to access the second. The structural implication is that the cover is attached to or supported by some other item in the system.
L2) ATTACHES TO: This represents a solid connection with no relative motion between the two items during operation. This link is broken by physically removing the first item from the second. When removing the second item in the link, the first item remains attached (i.e., the link remains intact). The structural implication is that the second item in the link is attached to or supported by some other item in the system.
L3) ATTACHES TO AND COVERS: This represents a solid connection with no relative motion between the two items during operation. This link is broken by physically removing the first item from the second. When removing the second item in the link, the first item in the link must be removed to access the second. The structural implication is that the second item in the link is attached to or supported by some other item in the system.
L4) ENGAGES: This represents a meshing-type connection with relative motion between the two items during operation. This link can be broken by disengaging either of the two items in the link. The structural implication is that the 2 items are attached to or supported by some other item(s) in the system.
L5) SUPPORTS: This represents a solid connection with no relative motion between the two items during operation. This link is broken by either physically removing the second (supported) item in the link, or by externally supporting the second (supported) item in the link and then physically removing the first (supporting) item. The structural implication is that the supporting item is attached to or supported by some other item in the system.
If a fastener or fastening process is required to maintain the link, we use a link modifier, called a sublink. It augments a link relation, such as "panel attaches to housing using screws." Sublink data contains the number of fasteners or process points, clearance around the fastener or process point, tool orientation and, for fasteners, removal and insertion direction. Figure 4 is a screen dump from our Linker design representation for the coffee maker implemented in ToolBook under Microsoft Windows.
The LINKER allows the user to evaluate a design from various stages of the life-cycle: assembly analysis, labor operation and labor step analysis for service, and product retirement analysis. Our experience with industrial collaborators indicates that this integrated feature is an essential key to promoting life-cycle engineering design. Each node or link has a data page that the user can access by double clicking on the graphical icon. Other data pages contain information for assembly and service analysis. We believe LINKER can serve as a broad tool for competitive product and process development and support ISO 9000 activities. LINKER, as a layout design representation, provides a front-end for our computer program for Life-cycle Assembly, Service, and Retirement (LASeR).
Figure 4. Life-cycle design tool showing a clumped coffee maker
5. DESIGN EVALUATION METHODOLOGIES
5.1 Life-cycle Serviceability
Service Mode Analysis (SMA) focuses on any form of service needs in estimating life-cycle ownership quality (Gershenson and Ishii, 1992). Service modes include regular maintenance, repair of failed components or systems, or service for undesirable side effects. The computer can use the LINKER to infer a sequence of labor steps needed to perform each mode of service. Given a set of cost driving service modes and their frequencies, the program can compute the total life-cycle service costs from the cost of each labor step.
The inferencing process starts inside the system and works its way out (Eubanks and Ishii, 1993). Starting with the malfunctioning component, the program examines all associated links. Depending on link type and direction, as in outgoing or incoming, the program will either: 1) generate a required labor operation pair (disassembly and assembly); 2) save the other component on a "component stack" for later processing; 3) do both 1 and 2; or 4) do nothing. When all links for the repair operation component have been examined, any components saved on the stack are processed in the same fashion. When the component stack is empty, the program searches for a link to a next higher assembly. If it exists, the linked component will be processed next, thus moving up one level in the component hierarchy. If a next higher assembly does not exist, the inferencing process terminates.
We compute the labor step cost based on the labor time, necessary tools and technician training required, and the part cost and part availability. The labor step cost (LSC) is:
(1)
where: tL = labor time (hours)
pL = labor time penalty (hours)
cLR = labor rate ($/hour)
cP = part or material cost ($)
pP = part or material cost penalty ($)
The labor time is the sum of handling time and either fastening or unfastening time. We add a labor time penalty to account for special tooling requirements, special technician training requirements, fastener clearance and tool orientation. Part replacement adds to the part cost and accesses a penalty based on part availability. We can now use equation (1) to estimate the life-cycle service cost LCSC:
(2)
where: fRj,k = frequency of labor operation j associated with
service mode k
LSCi,j = labor step cost i associated with labor operation j
l = number of labor steps associated with labor operation j
m = number of labor operations associated with service
mode phenomenon k
n = number of service mode phenomena being evaluated
(typically 5 to 10)
We compute the labor step costs for a set of repair operations and cost drivers displayed on a summary screen as shown in Figure 5. Values displayed are step cost, frequency of occurrence over the entire repair operation set, and total life-cycle cost (frequency x cost). Users can interpret the cost breakdowns and seek improvements by redesigning the structure or enhancing the reliability of components and systems.
Figure 5. Output screen for serviceability evaluation
5.2 Advanced Planning for Product Retirement
The product take-back laws in Europe mandate that manufacturers pay for product retirement. This trend urges engineers to make advanced plans for product retirement and seek recyclable designs. Again, LINKER provides an effective front-end for assisting in the advanced planning. Our method is based on a concept called "clumping." A "clump" is a collection of components and/or subassemblies that share a physical relationship, and some common characteristic based upon the end-of-life intent. Recycling requires that materials and fastening methods within the clump be compatible with existing reprocessing technologies (Marks, et al, 1993). For the coffee-maker example, one can group the product into two recycling clumps and one reuse clump. One would recover the plastic from the housing and the aluminum from the bottom cover and hot plate assembly. Since the carafe is an easily breakable item, it can serve as a service replacement. These clumps will not require further end-of-life disassembly. The issue is whether these clumps can be economically separated, reprocessed, and sold. Components can also be grouped for disposal. If the re-use or recycle value of a portion of the product is negligible, one might clump it for disposal and eliminate the disassembly cost. Of course, if the disposal clump contains a hazardous or toxic material, one must disassemble the system further to isolate and process the offending material.
Disassembly and reprocessing costs determine the system recycling cost. For a given system, as the number of individual clumps increases, the disassembly costs rise, and the reprocessing costs fall. Large, complex clumps, while easily removed from the system, require more complex reprocessing techniques. A larger number of simple, homogeneous clumps may require more time to disassemble, but are simpler to reprocess. The challenge in product retirement is to develop the most appropriate level of disassembly. The general retirement cost equation takes the form:
(3)
where: n = total number of clumps
System disassembly cost is a key factor in the analysis for product retirement. The total disassembly time for a system (with no clumps) is calculated by summing the individual disassembly times for each element in the system. (Equation 4).
(4)
where: Ds = system disassembly cost
Ci = time to remove component
Fj = time to remove fastener
Pk = time to remove or undo process
fnj = number of fasteners associated with one link
pn = number of process points associated with one link
l = total number of components in system
m = total number of links with fasteners
n = total number of links with fastening processes
After calculating the disassembly costs, one must evaluate the reprocessing costs for each clump. Unlike disassembly, reprocessing cost is extremely difficult to estimate at the layout design stage. By the time products are ready for retirement, which could be more than 10 years for durable goods, reprocessing technology and demand for recovered material could be very different from what it is today. In lieu of a reliable cost model, we apply a knowledge-based technique called the Design Compatibility Analysis (Ishii, 1992) to obtain a qualitative rating and then map it to a rough cost estimate. The analysis routine looks first at the components in the clump. It checks the knowledge base for any rules dealing specifically with components' material and post-life intent for the clump. The degree of compatibility maps to a [0,1] rating, as follows.
very compatible 1.0
compatible 0.8
limited compatibility 0.6
incompatible 0.2
hazardous 0.0
compatibility unknown 0.5
The compatibility rules (C-data) represent expert knowledge of the ease of reprocessing. A C-data contains its ID number, the associated design components/features, a compatibility descriptor such as "very good" or "poor," reasons and suggestions, and most important, the conditions for the data to be true. Here is an example describing material incompatibility.
C-data:
ID = dfr016
elements = material_A, material_B, intent
descriptor = incompatible: 0.2
reason = One ppm of PVC mixed with PET will cause discoloration of the PET.
suggestion = Try substituting polycarbonate for PVC.
conditions = material_A = "pet",
material_B = "pvc",
intent is primary_recycling. (5)
Then, DCA individually compares each component with every other component,
fastener, and process in the clump, creating the set [0,1]n, where n
is the number of matching compatibility data for the clump. We then map
[0,1]n into a single clump compatibility rating CC(s)
[0,1]
for each clump, s, using the following function.
1) the maximum in the set, if it consists only of numbers greater than or equal to 0.5.
2) the minimum in the set, if it contains at least one number less than 0.5.
3) 0.5 if the rule set is empty, indicating neutral compatibility.
To map the [0,1] rating to cost, we use equation (6). The cost decays exponentially as the compatibility increases. The cost curve is a result of a series of discussions with industry. If clump compatibility CC(s) = 1.0, we assume the cost to reprocess the clump is equal the market value of the recovered material. A clump with CC(s) = 0 indicates that there is a hazardous or toxic material in the clump and a reprocessing cost of infinity. If the clump has a rating of "incompatible," i.e., CC(s) = 0.2, then we assume that the clump is not worth reprocessing and it must be disposed of. Hence we assign a standard landfill cost for the clump, computed as a function of its weight or volume.
(6)
where:
CRC(s) = Clump Retirement Cost
LFC(s) = Landfill Cost
CC(s) = Clump Compatibility
Equation (6) substituted into equation (3) provides the total product retirement cost. Figure 6 shows the output of the retirement analysis for the coffee-maker example.
Figure 6. Retirement cost breakdown of ice dispenser assembly
5.3 Industrial Example
We applied our tool to two models of an in-door ice dispenser from GE refrigerators. The primary difference between these two designs is that the 1992 model dispenses ice using a primarily mechanical system of springs, wires, and an inertial damper, whereas the 1993 model dispenses ice using an electro-mechanical solenoid assembly. The 1993 model is a simpler design and has fewer moving parts. For assembly evaluation, we used the Hitachi-GE Assembly Evaluation Method, while the program used our original methods for serviceability and retirement analysis.
For all three areas of the life-cycle analysis, the new (1993) ice dispenser model shows a significant decrease in cost. Assembly costs decreased by 19%, service by 27%, and recycling costs by 23%. The fewer numbers of components in the new model contributed significantly to these decreased costs. Note that we assumed proportional clumping strategies for both ice dispensers, since we normally compare clumping strategies for a single design to improve its overall recyclability.
The case study established the high potential of our tool as a life-cycle design aid in the layout stages of product development. The key feature is the consolidated design representation LINKER, which allows rapid evaluation of various life-cycle costs. We do not claim our model to be an accurate cost estimator, but the tool does identify cost drivers and allow users to compare different design alternatives. To validate the cost model, we are currently tracking the actual cost of the new product. Early indications show that the reduction in manufacturing cost is close to our prediction, although some of the improvement comes from parts standardization and better product line structuring. Validation of the service and retirement costs would require continued monitoring.
6. CONCLUSIONS & FUTURE WORK
This paper began with a survey of methods aimed at improving the life-cycle quality at the early stages of design and identified the significant research issues in developing an integrated life-cycle design tool: design representation and life-cycle evaluation measures. We then presented a representation based on a semantic network and evaluation methods for serviceability and product retirement. These methods led to a PC-based computer program that allows the designer to quickly evaluate a layout design for life-cycle costs.
Our software addresses a rough model of the structure and obviously is not a comprehensive design tool. We view our model as a vehicle to develop practical methodologies, particularly in identifying the pertinent parameters and evaluation measures. Within our collaborating companies, the model has significantly raised awareness of life-cycle cost issues and led to practical training materials. Likewise, our cost model does not target accurate cost estimates, but rather seeks to identify cost drivers and capture relative differences among design alternatives. We believe such rough cost models are still useful in guiding the designers at the early stages.
Whereas our prototype shows promising results, there are many more challenges in life-cycle engineering design. Design representation and evaluation measures continue to be the central research issues in our future endeavors.
1) Addressing functional designs: LINKER only contains structural information about the design and thus cannot attribute the costs to functional intent. We also cannot readily incorporate failure modes and effects analysis into the current program. Some form of functional representation must accompany the corresponding structural layout.
2) Total life-cycle evaluation: The current program evaluates assembly, service, and product retirement separately. Whereas the designer can use the program iteratively to improve life-cycle quality, we ultimately want the total life-cycle cost including the environmental impact. Missing pieces include early evaluation of components, service logistics and support cost, and most importantly, the cost of environmental impact beyond product retirement.
3) Life-cycle design of product structure: Our previous effort focused on a single product. Most companies provide a line of products to cover the widest customer preference. One must look at the life-cycle of the entire product line and accommodate the necessary model variations in a most cost effective manner.
ACKNOWLEDGMENTS
Sponsors of this research include the National Science Foundation, NASA Lewis Research Center, General Electric, General Motors, and Ford. The author also acknowledges the members of the Life-cycle Design Laboratory at Ohio State University for their work that led to this paper.
REFERENCES
Alting, L. (1992) "Life-cycle Design of Products: A New Opportunity for manufacturing Enterprises," in Kusiak, A. (ed.), Concurrent Engineering: Theory and Practice, John Wiley, pp. 1-17.
Allenby, B.R. (1991) "Design for Environment: A tool whose time has come." SSA Journal, Vol. 12, No.9.
Birolini, A. (1992), "Design for Reliability," in Kusiak, A. (ed.), Concurrent Engineering: Theory and Practice, John Wiley, pp. 307-347.
Boothroyd, G. and Dewhurst, P., (1983), "Design for Assembly: a designer's handbook," Boothroyd Dewhurst Inc., Wakerfield, Rhode Island.
Burke, D., Beiter, K., and Ishii, K. (1992), "Life-cycle Design for Recyclability," Proceedings of the ASME Design Theory and Methodology Conference, September 1992, Scottsdale, AZ, DE-Vol. 42, ISBN 0-7918-0936-6, pp. 325-332.
Dixon, J.R. (1986). Designing with Features: Creating and Using a Features Database for Evaluation of Manufacturing of Castings. ASME Computers in Engineering 1986, Vol. 1., pp.285-292.
Eubanks, C. F. and Ishii, K. (1993), "AI Methods for Life-cycle Serviceability Design of Mechanical Systems," Artificial Intelligence in Engineering, Vol. 8, pp. 127-140.
Gershenson, J. and Ishii, K. (1992), "Design for Serviceability," in Kusiak, A. (ed.), Concurrent Engineering: Theory and Practice, John Wiley, New York, ISBN 0-471-55492-8, pp. 19-39.
Hauser, J. and Clausing, D. (1988) "The House of Quality," Harvard Business Review, Vol. 66, No.3, pp.63-73.
Ishii, K. (1992), "Modeling of Concurrent Engineering Design," in Kusiak, A. (ed.), Concurrent Engineering: Theory and Practice, John Wiley, pp. 19-39.
Marks, M., Eubanks, C., and Ishii, K. (1993), "Life-Cycle Clumping of Product Designs for Ownership and Retirement." Proc. of the ASME Design Theory and Methodology Conference, Albuquerque, New Mexico. ASME DE-Vol. 53, ISBN 0-7918-1170-0.
Miyakawa, S. et. al., (1990), "The Hitachi New Assemblability Evaluation Method (AEM)," Trans. of the North American Mfg. Res. Institution of SME, pg. 352.
Ormsby, A., Hunt, J. and Lee, J. (1991). Towards an automated FMEA assistant. Applications of Artificial Intelligence in Engineering VI, ed. G. Rzevski and R. Adey. Computational Mechanics Publications, Southampton, UK, pp. 739-752.
Poli, C., Graves, J., and Sunderland, J.E. (1988) Computer-Aided Product Design for Economical Manufacture. ASME Computers in Engineering, 1988. Vol. 1, pp.23-37.
Sturges, R., and Kilani, M. (1992) Towards an Integrated Design for an Assembly Evaluation and Reasoning System. Computer-Aided Design, Vol 24, No. 2, pp. 67-78.
US Environmental Protection Agency (1993) "Life-Cycle Assessment: Inventory Guidelines and Principles", EPA Report No. EPA/600/R-92/245, Office of Research and Development, Washington, D.C.
Taguchi, G. (1993). "Taguchi on Robust Technology Development: Bringing Quality Engineering Upstream." ASME Press, New York, NY.
Woods, W. A., (1975), "What's in a Link: Foundations for Semantic Networks," Readings in Knowledge Engineering, (R. Brachman and H. Levesque, eds.), Morgan Kaufman Publishers, Los Altos, CA.