A Distributed Design Methodology For Extensible Product Life Cycle Strategy Jianzhi Li, Puneet Shrivastava, Hong C. Zhang Industrial Engineering Dept., Texas Tech University. Lubbock, Texas, USA, 79409 Email:
[email protected] already started the product take-back operations for the end of life computers from their customers. One alternative available to handle this huge pile of E-waste is to recycle all the products and recover the precious materials and reduce the harm to the environment which occurs due to landfill. But recycling needs infrastructure, logistics support, capital investment and investment in development of environmentally benign technologies. Recycling should therefore be the last resort for EOL products. The components which still have functional life left should be incorporated in other new or refurbish products. There is also an added incentive of lowering purchasing/inventory costs due to reuse of components from End-of-Life products [6]. Returned products may enter the production process again as input resources, either in the original form or as components and modules after disassembly. Many firms have recently reported increased profits by reselling returned products in secondary markets after refurbishment [4,7]. Reference [4] claims that remanufacturing operations account for sales in excess of $53 billion per year in the US. With this view researchers have also conducted studies in manufacturing, design and recycling of products trying to find an answer to the problem. But all these work are within current business settings. Particularly, the opportunity for us is to identify the possible system that can formulate a product chain and extend the life cycle of product and service in a multi-agent environment that requires negotiation in product design variables, while satisfying the requirement of performance from consumers’ perspective. By successful implementation of this strategy, we can not only reduce the EWaste, but also reduce material, manufacturing, and recycling requirement in the electronic industry.
Abstract- In this paper, the problem of life cycle mismatch in electronic product is discussed. An extensible product life cycle (EPLC) strategy is proposed and studied, which is a possible answer to the problem of E-Waste and helps to lower product costs and improve environmental performance. EPLC strategy is recognition of the mismatch between product performance life cycle and its functional life cycle. In essence an EPLC strategy is to reincarnate a product at component level and share them with other products in a proactive manner, thus form a product chain that will extend and continue the life of the otherwise “obsolete” product in different industrial applications. Implementation of this strategy is different from the current practice of product reuse. It requires the manufacturers to schedule a product’s extended life cycle early in the design phase and optimize the design parameters considering the requirements in the extended life span. Since the design problems involved in EPLC are interrelated and designers are also distributed in different departments, a distributed design framework is developed to find the Pareto optimal set for the design problem involved in EPLC. Conclusion is also given out in the last part. Keyword: End-of-life; Extensible product life cycle; Decision based design; Distributed design.
I. INTRODUCTION Over the last decade, there has been a rapid development in technology, especially in information technology. While this has significantly improved product performance, the market is flooded with electronic gadgets, which quickly become obsolete with the unveiling of new innovative technology. People do enjoy these products and services, but they also have to update or renew their products, although the products may still be in good condition. This situation is more prominent in the personal computer industry. The lifespan of a computer system is shortened as hardware and software companies constantly develop newer components and programs that fuel the demand for more speed, memory and power. Along with increase in number of products consumed, a huge amount of “end of life” products are accumulated and waiting to be recycled, although they have not really reached the end of their life. A recent US study by the National Safety Council found that over 315 million computers will become obsolete by the year 2004 (an underestimate) and only 16 percent of them will be recycled by 2005 [1]. Environmental concerns, on the other hand, have expanded the list of customer expectations to include the environmental criteria, which requires manufacturers to be more responsible for the post-consumer management of their products [2-3]. OEMs such as Dell, HP and Compaq have 0-7803-8250-1/04/$20.00 © 2004 IEEE.
II. EXTENSIBLE PRODUCT LIFE CYCLE A. Product Life Cycle Mismatch The reason why life cycle of a product can be extended is because there is a mismatch of performance life cycle and functional life cycle as shown in Fig. 1. Whenever technology innovation happens, the product will be discarded while the product or components of the product can still be used. In modern industries, product life cycle determined by its performance life cycle is much shorter than its functional life cycle. Average performance lifecycle for computer systems is 2 years but its functional life cycle can be more 10 years.
214
computers. But under the current business settings this end-oflife strategy is highly flawed as the end of life component reuse philosophy needs to be incorporated during the design stage with the participation of the all the manufacturers, recyclers and suppliers. This is the essence of the extensible life cycle philosophy.
10
rm
an
ce
life
c y c le
cyc
B. Extensible product life cycle and its requirements The idea of extensible life cycle is to fully utilize this mismatch as illustrated in Figure 2. Life cycle of the components in product 1 is no longer limited by life cycle of product 1. Instead, it will be used in another product by component sharing in product 2 or product 3. By doing this, the components will reincarnate itself as a part of a new product with an extended life cycle.
le
00
rfo
li f e
95
85
80
Pe
ti o n
90
Life cycle
c Fun
Year
Figure 1. Mismatch of product life cycles
The second type of life cycle mismatch is the difference among the life cycle of the components that compose the product. This is because the life cycle of the components, either functional or performance, are essentially different. For example, the life cycle of a motherboard may be longer than the processor. Reference [8] highlights the mismatch in the electronics product as shown in Table.1 for their study for End-of-life Design Advisor (ELDA) for Phillips electronics products. TABLE I FUNCTIONAL AND PERFORMANCE LIFE CYCLE MISMATCH IN ELECTRONIC PRODUCTS Functional Life Performance Life Cycle Products Cycle Cycle (years) Desktop Computer 10 2 LCD Monitor 5 2 CD-Recordable 7 2.5 Audio System 9 3.5 Television Set 11 4 Cordless phone 10 5 Washing Machine 10 6
Figure 2. Illustration of multiphase extensible component life cycle
The extensible life cycle strategy is in essence a proactive method that should be conducted in the early stage of a product in product design and other activities. It will schedule and predetermine the life-cycle stages and product chain that a component will pass, thus determination of the design parameters will consider the requirement in the extended product life cycle. Under extensible product life cycle strategy, recyclers will know where remanufactured components should be delivered, and they will not have to worry about sales of the remanufactured component by remarketing, since this has been predetermined. The remanufactures will also know other requirements of the remanufactured components, such as quality and packaging. Firstly, a revolution in product design is required by EPLC strategy. Product design is no longer limited to only one product, designers have to consider all the possible products, in which a component might be a part, select the optimal design parameters, segregate and allocate their extended life cycle to each product. To do this, suitable information sharing and decision support tools should be required for decision makers to collaborate in the design process.
As per their analysis of the electronics products the recommended end-of-life strategy for 77% of the Philips products was remanufacturing. Only 15% of the products had suggested end-of-life strategies of recycling for material value with moderate disassembly. The remaining 8% of the products should be recycled for material value without any disassembly, according to the ELDA analysis. Tibben-Lembke discussed the issue of component life cycle as a part of the product life cycle [9]. According to their analysis when product is disassembled and components reclaimed, the life cycle of the components it contains is an important factor in the value of the product. For example, consider the components that may go into a personal computer: RAM, hard drive, CPU (model and speed), video card, sound card, modem, SCSI cards, ZIP drives, etc. Reference [10] illustrates, sales of computer memory chips follow very nicely the product life cycle curve given in most marketing texts [11]. After sales stop for their original use, some products will live on in new uses. Even though a product may be considered obsolete in the personal computer market (such as an Intel 386 processor), it may be powerful enough for use in many devices, and there may continue to be demand for this component, long after it has stopped being sold in new
0-7803-8250-1/04/$20.00 © 2004 IEEE.
III.
DECISION MAKING FRAMEWORK AND INFORMATION SHARING
Collaboration of decision-makers from different departments with different domain knowledge and information sharing between them are crucial in successfully implementing this strategy. Table II lists some activities in which the secondary manufacturer, supplier and manufacturer’s participation and collaboration in decision-making and information sharing are required. Each of these activities also
215
When environmental issues are also in our consideration, product modularity will be determined not only by functional requirements and production issues, but also by product environmental objectives, such as usage, service, recycle and disposal. When further considering this extensible life cycle strategy, decisions generated in design will definitely be affected by reverse production center and requirements and demands of the secondary manufacturer as illustrated in Figure 3. Integrating decision-making in product development with decision-making in down stream chain is essential for extending the product to another round of life. A. A distributed agent based decision support framework The distributed and multi-attribute characteristics of the decision problems in the EPLC strategy requires a special decision framework that can help the decision makers interact, and communicate, negotiate with each other. The decisionmaking framework designed in this research can fulfill the requirement of distributed decision making in EPLC is illustrated in Figure 4. The decision framework for EPLC includes following components: a. Plug and play information-sharing infrastructure The plug and play information sharing provide the deployment environment for the decision agents and information fusion required by decision agents when decision tasks are conducted. It supports dynamic integration among various agents and the mechanism to communicate with sources of information systems and applications by the strategy of messaging and service oriented integration. The information required by decision makers or agents is managed and fused together in a structured format by the product and material management agent (PMM Agent). It provides a distributed environment where the agents are deployed and situated. The agents can also provide input to and change the applications they are bounded through the information-sharing infrastructure. b. Decision tasks To implement the EPLC strategy, some key decision tasks should be conducted. These tasks are related to operations such as product design, manufacturing, supply chain management, reverse logistics and end of life management. Although these tasks are presented here as separated, from a concurrent engineering’s point of view, they are interrelated to each other. Decisions made in product design phase have to consider performance of a product in its whole life cycle including production, logistics, using of product, recycling. These decision tasks may include: • Design for extensible life cycle (DELC) • Extended green supply chain management • End of life management
has to be coordinated in related departments of each organization. Because of the interdependence of these activities, some decision parameters will unavoidably lead to conflict among different departments and organizations. TABLE II ACTIVITIES REQUIRING COLLABORATION IN DECISION-MAKING Manufacturer Supplier Reverse Secondary (Mfr) Mfr Mfr Pricing
X
X
Design
X
X
Production
X X X X
X X
Forward Logistics Reverse Logistics Forward inventory Reverse manufacturing Reverse Inventory
X X
X
X
X
X X
X X
X
In this paper, a particular focus is given on collaborations that can fill the gap between product design and down stream supply chain. Design decisions formulate the future of the product life cycle, whereas the supply chain network is the place where management and control of product’s life really happen. Decisions made in the product development phase have to concurrently consider other issues that cannot be answered by the designers. Integrating design with down stream supply chain will also provide design with tools such as statistical and stochastic analysis. People have put their effort in understanding design as a decision-making process for a while. These efforts had yielded result such as Decision-Based Design (DBD), which views design as a decision-making process with values, uncertainty and risk. Herrmann also proposed that product development organization could be viewed as a Decision Production System and tried to identify the suitable information flow in this system [12]. These research activities provide us a sound basis of applying available decisionmaking theories to this strategy. On the other hand, modular design has been widely practiced by the electronics industry to get flexible configurations, ease of maintenance and upgrading, standardization, and shortening design and production time [13].
Figure 3. Two OEMs share the same components in two products
0-7803-8250-1/04/$20.00 © 2004 IEEE.
216
Business Model for Extensible Product Life Cycle
D e c is io n M a ke r 1
D e c is io n M aker 2
D e c is io n M aker n
D is trib u te d
N e g o tia tio n P ro to c le
M u lti-C rite ria
M u lti-a ttrib u te U tility A n a ly s is
D is trib u te d D e c is io n -M a k in g
L ife C y c le A s s e s s m e n t
P ro d u c t D e s ig n
E x te n d e d S u p p ly C h a in M anagem ent
M a n u fa c tu rin g
E n d o f L ife M anagem ent
PM M P lu g a n d P la y In fo rm a tio n S h a rin g F ra m e w o rk
Figure 4. Overall Decision Framework
c.
Life cycle assessment As environmental performance is one important attribute of every decision task, a life cycle assessment layer is developed. This layer provides the methodology in environmental assessment for each of the decision tasks. Because of the distributed nature of decision tasks, life cycle assessment is separated and embedded into smaller assessment tasks that will only evaluate the environmental performance of one or several processes relevant to the decision makers. By embed, we refer to the fact that this segregated life cycle analysis is embedded in the process of decision-making, unlike traditional LCA procedure, which is conducted after the decision has been made. d. Distributed Multi-Agent Decision-Making Support Layer No matter what the decision task is, there will be many decision makers (or agents) participating in the decision tasks, which are interrelated, in the Distributed Multi-Agent Decision-Making Support Layer. Autonomous agent and agent interaction mechanism such as negotiation protocol are implemented in the distributed decision-making support layer. This layer will be build upon the plug & play information sharing infrastructure that provides the needed data and communication mechanism for the agent situated in the support layer. e. Decision makers Decision makers are the people or software agents that have their own goal and participate in decision problems of their interest. It is by the decision agents that the decision tasks are completed in the distributed decision support layer. The focus of this research will be the development of the distributed multi-agent decisionmaking support layer, and the embedded LCA layer. In
0-7803-8250-1/04/$20.00 © 2004 IEEE.
217
the following sections, we will discuss more on these issues in this decision-making framework. B. Modeling of the decision problem A general m-agent decision-making problem is to find the optimal value of the decision variables that maximize a vector of utility functions of the m participating agents over a set of feasible resource constraints [17]. The problem is to determine a set of non-dominated (Pareto optimal) solutions and make compromise among these decision agents. The standard form of the problem formulation for loosely coupled situation (resource allocation) can be expressed as (modification based on [14]): K Maximize: U ( xG ) = f { u 1 ( xK ), u 1 ( xK ), " , u m ( xK )} (1)
G
Where ui ( x ) is the utility function of agent i. Subject to: x 1i + x i2 + " + x im ≤c j for i=1,2, … ,n j
j
ui ≥xi ≥li
j
or i=1,2, … ,n and j=1,2, … ,m
C. Pareto Optimal Conception To solve this distributed optimization problem, the convenient way is to first find a Pareto optimal solution. Definition: the conception of the Pareto optimality can
K*
be stated as: a vector of x is optimal, if there exists no K feasible vector x which would increase some agents’ utility without causing a simultaneous decrease in any agent’s utility”[15]. Mathematically, the Pareto K* optimality is defined as follows: a decision vector x is optimal if and only if for any x and i,
K* K u j ( x ) ≤ u j ( x ),
(2) K* K j = 1, " , m; j ≠ i ⇒ ui ( x ) ≥ ui ( x ) D. The negotiation process to reach a Pareto point Suppose two agents have to negotiate on a multivariable allocation problem. A proposal is an allocation of resources to both agents and is expressed as p ∈ P , where P is the set of possible allocations. Figure 5 shows a simplified two decision variable problem in an Edgeworth box [16]. A single proposal is composed by two decision variables representing two goods. The allocations of the goods are assignments of values to decision variables. Thus, in this figure two agents a and b have to reach an agreement over the allocation of two commodities 1 and 2. The dimensions of an Edgeworth box represent the constraint of quantities of the good. Each agent is assumed to have an initial endowment of both goods. Suppose both agents have different utility preferences over the resources. This is shown by the convex indifference preference curves (or iso-curves) of the two agents in Figure 5, where each curve represents the indifference an agent has over the increasing/decreasing utility of one commodity versus the simultaneous decrease/increase in utility of the other commodity, and the tangent line to the indifferent curve can be modeled by MRS (marginal rate of substitution) which is expressed as: dx MRS → Slope of IC S = − 2 (3) dx 1 E. Determination of the entire Pareto sets As mentioned before, the Pareto point generated in this manner will form a so-called “Pareto frontier” which is generally a surface or line that all the Pareto points are located. Another problem for us is to pick up the optimal point that will give the system the best performance. If the Pareto surface is monotonic, we can design a kind of algorithm that will lead us to the optimal Pareto point. But if it is not, we have to find all the Pareto points and compare among them. It is obvious that the increase in the number of decision makers, the dimension of the Pareto frontier will increase linearly. It will require a lot computing power in order to locate all the point in the Pareto sets by the method discussed before. A better way is to use the Pareto points generated to predict the possible Pareto points, and then put them to the negotiation process. The point generated in this way will be very near to the Pareto point after it is put to the negotiation process, which will save the time and rounds required in the negotiation process. Another consideration in this process is that it is better to find the grid that covers the Pareto surface instead of all the possible points. Comparison of Pareto point in the grid will give us a basic idea where the best solution is located. We may further try to fine-tune the point on that grid only. This will also save the computing power. As illustrated in Figure 5, the joint utility level of 0-7803-8250-1/04/$20.00 © 2004 IEEE.
218
both agents is increased at the point where the convex indifference curves of each agent intersect at one point. A hypothetical set of such points is shown as solid black ovals. These allocations are said to be Pareto optimal over the initial allocation. At this point, because their tangent line overlaps with each other, the MRS of both agents is the same.
Figure 5. Illustration of Pareto optimal point
All the Pareto points generated in this manner is referred to as the Pareto frontier or the Pareto set as showed in Figure 6. As illustrated in the figure, the joint gain of joint welfare of both agents on these Pareto point are more than 1. This is resulted from the negotiation process that allocated more resources to the agent that has more preference to one of the two goods. So, In essence, negotiation is a sort of tradeoff between conflicting or interdependent agents. If both agents have same preference to the same commodity, the joint gain will be reduced since there is not tradeoff between these two agents regarding the allocation of good 1 and 2. The points on the Pareto frontier are Pareto optimal. 1
Pareto Optimal Frantier
Ub
0
Ua
1
Figure 6. Illustration of Pareto optimal frontier
F.
Pareto point prediction Suppose we have some Pareto points available that were generated in previous rounds of negotiation process, these points can be used in prediction of the new Pareto point around one Pareto point generated. The Pareto point prediction uses second order Taylor series in prediction of new Pareto points.
Suppose the Pareto frontier expressed in decision variables is modeled by the following equation: pn = frontier ( p1 , p2 , " pn−1 ) (4) Where p1, p2 , " , pn designate the decision variables or proposal vector. * Suppose we have one available Pareto point p , the second order Taylor series for the frontier around Pareto * point p is:
1 n−1 * n−1 dpn p n = pn + ∑ ∆pi + ∑ ∆pi H ij ∆p j 2 i = j =1 i =1 dpi
Where
(5)
dpn
designates the partial derivative of dpi pn to pi , ∆pi designate a delta increase of the new Pareto point to the old Pareto point in dimension i < n 2 d pn and H ij is given by : H ij = dpi dp j If OEM wants to improve the precision of the Optimal Pareto point, he/she can further conduct negotiation within the grid composed by the Pareto points around the point selected as illustrated in Figure 7.
Figure 7. Illustration of the selection of the optimal Pareto point
IV. CONCLUSIONS In this research, a new business strategy of extensible product life cycle is proposed. It advocates product component sharing by recognition and utilization of the mismatch between product performance life cycle and functional life cycle. This research will be applicable where there are multiple decision variables and multi agent environment. In order to optimize the design parameters for this extended life cycle, a distributed multi-agent negotiation based decision framework is developed and proved effective in solving the decision problem associated with EPLC strategy. Decision
0-7803-8250-1/04/$20.00 © 2004 IEEE.
219
problems in EPLC are modeled and solved using Pareto point analysis. ACKNOWLEDGMENT
This research was supported by the Texas Higher Education Coordination Board’s Advanced Technology Program (ATP) award under grant contract #0036440241-2003 and National Science Foundation (NSF) grant # DMII-0225927. REFERENCES [1] "Computer Display Industry and Technology Profile", EPA 744.R.98.005, December 1998 and based on figures presented by the National Safety Council, Washington DC, May 1999 "Electronic Product Recovery and Recycling Baseline Report" [2] C. Schempf, Noellette and Lave, Lester. “Green Design and Quality Initiative.” Handbook of Environmentally Conscious Manufacturing. [3] J.R. Matthew, J. C. Ammons, D. Newton, “Strategic Design of Reverse Production Systems” Computer & Chemical Engineering, vol. 24, 2001,pp. 991-996 [4] V.D.R. Guide, V. Jayaraman, R. Srivastava & W.C. Benton, “Supply chain management for recoverable manufacturing systems”, Interfaces, 2000. [5] D.L.Rogers, R.S. Tibben-Lembke, “Going Backwards: Reverse Logistics Trends and Practices”. Reverse Logistics Executive Council, 1998. [6] M. Fleischmann, H.R. Krikke, R. Dekker, S.P. Flapper, “A characterization of logistics networks for product recovery”. The International Journal of Management Science, vol. 28, 2000, pp.653666. [7] R.B.M. Koster, M.P. de Brito & M.A. van de Vendel,“Return Handling: an exploratory study with nine retailer warehouses”. International Journal of Retail & Distribution Management, vol.30, no. 8, 2002. [8] C. M. Rose and Ab Stevels, K. Ishii, “Method for formulating product end-of-life strategies for electronics industry”, Journal of Electronics Manufacturing, vol. 11, no. 2,2002, pp. 185-196 [9] R. S. Tibben-Lembke, “Life after death: Reverse logistics and the product life cycle”, International Journal of Physical Distribution & Logistics Management, vol. 32 No. 3, 2002, pp. 223-244 [10] N. Bollen,. “Real options and product life cycles’’, Management Science, vol. 45,no. 5,1999, pp. 670-84. [11] P. Kotler, Marketing Management: The Millennium Edition, 10th ed., Prentice-Hall, Upper Saddle River, NJ,2000. [12] Herrmann and Schmidt “Viewing Product Design as a Production System.” NSF Conference, Birmingham 2003. [13] J. Li, Y. Yue, H.C. Zhang, “A Fuzzy Graph Approach to Improve Product Modularity in Support of Environmentally Conscious Design”, Industrial Engineering Department, Texas Tech University. Unpublished. [14] Ehtamo, Harri, Verkama, Markku, and Raimo P. H¨ am¨ al¨ ainen, “How to Select Fair Improving Directions in a Negotiation Model over Continuous Issues.” IEEE Transactions On Systems, Man, and Cybernetics—Part C: Applications And Reviews, vol. 29, no.1, 1999, pp. 26-33. [15] Tappeta, V. Ravindra and J.E. Renaud,“Interactive Multiobjective Optimization Procedure” AIAA-99-1207, 1999, pp. 27-41. [16] Heiskanen, Pirja. “Decentralized Computation of Pareto Solutions in Multi-Party Negotiations.” Thesis for the degree of Licentiate of Technology, Delft, May 1998, Helsinki University Of Technology, Department of Engineering Physics and Mathematics, Systems Analysis Laboratory. [17] V. J. Neumann, O. Morgenstern, Theory of Games and Economic Behavior (2nd edition) Princeton, N.J.: Princeton University Press, 1947
