1995 ASME Advances in Design Automation Conference, Boston, Massachusetts, Sept. 17-20, 1995.

THE USE OF ACTIVITY-BASED COSTING, UNCERTAINTY, AND DISASSEMBLY ACTION CHARTS IN DEMANUFACTURE COST ASSESSMENTS

Bert Bras and Jan Emblemsvg The Systems Realization Laboratory The George W. Woodruff School of Mechanical Engineering Georgia Institute of Technology Atlanta, Georgia 30332-0405

ABSTRACT In this paper, the development of an Activity-based Cost (ABC) model is presented for use in design for demanufacture under the presence of uncertainty. Demanufacture is defined as the process opposite to manufacturing involved in recycling materials and product components after a product has been taken back by a company. The crux in developing an ABC model is to identify the activities that will be present in the demanufacturing process of a product, and afterwards assign reliable cost drivers and associated consumption intensities to the activities. Uncertainty distributions are assigned to the numbers used in the calculations, representing the inherent uncertainty in the model. The effect of the uncertainty on the cost and model behavior are found by employing a numerical simulation technique - the Monte Carlo simulation technique. The additional use of disassembly action charts allows the influence of the uncertainty to be traced through the cost model to specific demanufacture process and product design parameters. OUR FRAME OF REFERENCE The growing importance of including environmental issues in design has amplified the impetus for companies to more formally consider the entire life-cycle of a product, from cradle to grave or even to reincarnation through recycling and reuse. A crucial issue is the assessment of costs (or profit) related to pursuing environmentally benign products and processes. We believe that in order to provide efficient and effective decision support in life-cycle design, costing methods should: 1) Assess and trace costs and revenues; 2) Handle both overhead and direct costs; 3) Handle uncertainty; 4) Provide decision support for the process of designing. Attempts to generate formal and systematic design approaches which include environmental considerations, as well as economical, mechanical and other design considerations run aground when confronted with the need to quantify environmental

properties and requirements [1]. Some hurdles to developing environmental impact and cost models are: a) there is a lack of both hard test data and past experience, b) new and different technologies are integrated with unknown effects, and c) designs are incomplete and evolving at the stage where the largest reduction in environmental impact and cost can be made. In addition, the world s environmental processing capabilities are largely uncertain. Hence, the following two questions still remain largely unanswered: What is the cost associated with pursuing environmentally benign products and processes? Which aspects of both the product and process design have the largest influence on these costs? In this paper, we present a method for developing cost models which aid designers in answering these and similar questions in the context of designing for the life-cycle. The core of our method is the combined use of Activity-Based Costing and uncertainty. Several costing approaches have appeared in the literature in the context of designing environmentally benign products and processes. However, when it comes to assessing costs to lifecycle and ecological issues, Activity-Based Costing (ABC) is gaining ground rapidly on conventional costing systems [2-5] . Based on our review of relevant life-cycle costing approaches, we believe that emerging Activity-Based Costing approach has the best potential for efficient and effective cost assessments in the context of designing for the life-cycle [6] . In this paper, we extend the work presented in [6] by including the following issues: The uncertainty associated with the information used in product cost assessments. In particular, we highlight the modeling and propagation of these uncertainties through cost models using commercially available software. The capability to identify those process and product design aspects that contribute most to the cost using so-called action charts. This enables a design group to quickly spot the most

285

cost inefficient parts of the design which allows the group to concentrate the redesign effort and improves design efficiency. The approach will be shown in the context of a subset of lifecycle design, i.e., design for demanufacture1 of telephones. ACTIVITY-BASED COSTING Activity-Based Costing (ABC) has received its name because of the focus on the activities performed in the realization of a product. Costs are traced from activities to products, based on each product's consumption of such activities. Activity-Based Costing differs from conventional costing systems in two distinct ways: 1) In conventional costing systems, the assumption is made that each unit of a given product consumes resources (e.g., energy, material and direct labor), while in ABC the assumption is made that products or services do not directly use up resources, but consume activities. Hence, in ABC, the cost of a product equals the sum of the costs of all activities that must be performed in the realization of the product [4] . 2) Conventional cost systems are based on unit-level cost drivers (these unit-level cost drivers are often referred to as allocation bases in conventional cost systems) of the product that are directly proportional to the number of units produced. Direct labor hours, machine hours and pounds of material are examples of such unit-level allocation bases. An ABC system, on the other hand, uses cost drivers that can be at the unit-level, batch-level, and/or product-level. Examples of batch-level cost drivers are setup hours and number of setups. Examples of product-level cost drivers are number of parts, number of times ordered, and number of engineering charge orders [7] . Because of the assumption that a product uses activities and the allowance for batch and product level cost-drivers, it is generally agreed that ABC systems are superior in modeling and tracking costs (see, e.g., [4, 7] ). Mostly noted is ABC s capability to separate direct from indirect costs. In [4] it is noted that traditional cost systems systematically undercost small, lowvolume products and overcost large high-volume products. This is due to the inability to trace overhead costs correctly, which in turn results from the use of only unit-level cost drivers and the focus on resource consumption. In depth discussions of ABC can be found in, e.g., [4, 5, 7-9] . We have chosen to use ABC because of the noted superiority in cost-tracing, separation of direct and indirect costs, higher accuracy, and its capability to blend into the Activity-Based Management (ABM) systems that more and more companies are employing (see, for example, [7] ). A motivating example for its use in an environmental context can be found in [3] where it is described how Activity-Based Costing and environmental aspects can be combined to give companies the ability to identify more accurately those plants and products which are driving up their environmental expenditures. However, it should be noted that, although many have focused on ABC, the issues of a) how to provide efficient and effective decision support in design, and b) how to best include the uncertainty involved

are still largely unaddressed. Especially the uncertainty cannot be ignored and should be included when one seeks to assess costs associated with aspects of a multi-year product life-cycle without much historical data. INCLUDING UNCERTAINTY IN COST MODELS Although a number of researchers outline methods for assessing and reducing environmental impact (e.g., [10] ), hardly any discussion is given to the accuracy of the data used and the sensitivity of the outcome to variations in the inputs. Especially when dealing with ecological issues uncertainty must be included due to a predominant lack of hard data. Uncertainty can be modeled in a variety of ways depending on what kind of uncertainty is to be modeled. Generally speaking, we have the following possibilities: We can model uncertainty based on historical data. This will typically involve statistical analysis along the line of Gaussian Statistics. We can model uncertainty based on experience, qualified guessing, etc. One way of doing this is by modeling the uncertainty analogous to fuzzy numbers, but solving the model numerically rather than using fuzzy theory. In design, and especially in original design, good historical data are often impossible or difficult to get, thus methods based on Gaussian statistics will soon become inappropriate and even impossible to apply. Our method should therefore be designed to deal with fuzziness . This means that we guess, for example based on experience, the type of distribution to use as well as the mean, the left deviation and the right deviation. The uncertainty is simply modeled by assigning distributions to every number in the model for which there exists uncertainty. Given that the uncertainty is modeled in a cost model, we must determine the effect of these uncertainties on the cost. We have found it useful to use the Monte Carlo simulation technique to find the cost uncertainties resulting from our assumptions.
Assumption cells
Chosen statistical distribution for an assumption cell

Forecast cells
Statistical distribution of a forecast cell

Trials
4 12 20 etc.

Results
10 20 35 etc.

Direct Labor

Propagation through model

6 8 15 etc.

Product Cost = Direct Labor + Material The forecast cell is shown here as a histogram for simplicity

Material

FIGURE 1 MONTE CARLO SIMULATION EXAMPLE This technique is a very simple, but powerful numerical approximation method that is simply based on performing a controlled and virtual experiment within a model. Although numerous different simulation methods exist, we have found it advantageous to employ a software called Crystal Ball for this purpose. It allows the definition of assumption and forecast cells in a spreadsheet computer model. The Crystal Ball software adds into Microsoft Excel spreadsheet software, hence we talk about cells . A forecast cell can be looked upon as a response variable, while an assumption cell can be viewed as a

Demanufacture: the process opposite to manufacturing involved in recycling materials and product components after a product has been taken back by a company.

286

source variable. Consider the example in Figure 1 where product cost is modeled as a simple linear function of material and direct labor cost. Based upon our assumptions w.r.t. material and direct labor, we want to forecast the associated product cost. In each assumption cell, an uncertainty distribution is defined as is appropriate for the particular value in that cell. In our example (see Figure 1), the Direct Labor assumption cell is distributed as a triangular distribution while the Material assumption cell is distributed elliptically. The Monte Carlo simulation provides random samples of numbers in the assumption cells (material and direct labor). The random numbers propagate through relationships/equations in the model and the value of the associated forecast cells (product cost, in our example) is calculated by means of the appropriated relationships/equation. In our example, the value of the forecast Product Cost is a simple summation of the random numbers for Material and Direct Labor. When all the trials have been performed, the calculated values of a forecast cell will form a new statistical distribution (see the Product Cost distribution in Figure 1). Due to the randomness, the numbers that have propagated through the model can be used in ordinary statistical analysis as if we were running a real experiment, e.g., to construct confidence intervals, perform T-tests, etc. DEVELOPING AN ABC MODEL INCORPORATING UNCERTAINTY FOR DESIGN SUPPORT Our method for developing ABC cost models that includes uncertainty for decision support in design uses consists of six steps. A flow chart of our method is given in Figure 2. We illustrate our method by developing an ABC model for assessing and tracing the cost of demanufacturing a telephone and show how to guide detailed design changes based on the assessments provided by the model. Before discussing the details, we point out that our method has the following core components (see Figure 2): Formulation (steps 1 through 3) These steps deal with the actual formulation of the model. Solution (step 4) The formulated model is solved, i.e., a cost assessment is obtained. Validation (steps 5 and 6) The results from the solution process are used to assess and verify the model and reiterate the process if necessary. Step 6 will not be addressed in the paper. More information can be found in [6] .

Step 1 Create an Activity Hierarchy and Network

Step 2 - Identify and Order all Necessary Cost Drivers and Consumption Intensities Step 6 Iterate, if necessary Step 3 - Identify the Relationships between Cost Drivers and Design Changes

Distinguish: Design Dependent Cost Drivers Design Independent Cost Drivers Distinguish: Selection of best design no explicit relationships needed Modification of a design: mathematical functions, or action charts Use commercially available MS Excel and Crystal Ball software

Step 4 - Find and Minimize the Cost of the Consumption of Activities Step 5 Evaluate the Solution

Use profitability distributions, sensitivity analyses, and trend charts.

FIGURE 2 FLOW-CHART FOR DEVELOPING ABC MODELS Step 1 - Create an Activity Hierarchy and Network. The purpose of this step is to break down that part of the life cycle which forms the design focus into a hierarchy of activities. A demanufacture process can be broken down into (sub)activities as in Table 1. This is not the only possible hierarchy of demanufacture activities. As depicted in Table 1, three different levels of activities are present. For example, activity A1 ( Collect ) consists of the level 2 activities Buyback , Transport , and Store (A11, A12, and A13, respectively). Of those activities, only Transport has lower level activities, namely, Load and Move (A121 and A122). The purpose of an activity hierarchy is to ensure that all the activities in the part of the life cycle to be studied are considered. If an objective is to identify the effect of changes in design parameters on cost, then it is essential to form activities detailed enough that cost drivers can be assigned and that the lowest level of the activity hierarchy can be assigned directly to the design parameters through the cost drivers. After identifying all the activities, a network indicating the relationships between the activities is constructed. In Figure 3, the activity network is shown for a demanufacture process corresponding to the activity hierarchy in Table 1. In the demanufacture case study we are interested in two different process scenarios: Dismantling; this scenario includes only activities related to the process of dismantling a product as much as possible/feasible in order to recover reusable components. Shredding; this scenario includes only activities associated with shredding a product. There is no dismantling. Both process scenarios are included the network. Another approach could have been to develop separate models. It is important to note that, in general, an ABC activity network does not have a one-to-one corresponds with a process network. In an activity network, connections and relationships between process activities are given. A single activity may, however, consist of several process actions. Consider activity A311 manual dismantling . This activity contains all manual dismantling actions, no matter where or in what sequence they occurred in the demanufacturing process. We use the network to identify:

287

 what effect a change in the design parameters will have on the consumption of activities, and what effect a change in consumption of an activity will have on the other activities. The network also provides the designer a graphical view on how different decisions will affect the activities required. TABLE 1 DEMANUFACTURING ACTIVITIES2.
Level 1 activity Collect (A1) Level 2 activity Buy back Transport Store Non-hazardous dismantling (destructive/ non-destructive) Level 3 activity Load Move Manual dismantling Dismantling using handtools Dismantling using light equipment Dismantling using special equipment Manual dismantling of haz. comp. Dismantling of haz. comp. using handtools Dismantling of haz. comp. using light equipment Dismantling of haz comp using special equipment Sort non-haz. reusable comp. Sort non-haz. recyclable mat. Sort haz. reusable comp. Sort haz. recyclable mat. Inspect visually non-haz reusable comp. Test non-haz. reusable comp. Inspect visually haz. reusable comp. Test haz. reusable comp. Non-haz. loading Non-haz. moving Haz. loading Haz. moving Non-haz. loading Non-haz. moving Haz. loading Haz. moving Collect non-haz. waste Collect haz. waste Keep records Keep storage Keep max. storage Keep records Keep storage Keep max. storage Notation A11 A121 A122 A13 A2 A311 A312 A313 A314 A321 A322 A323 A324 A411 A412 A421 A422 A51 A52 A611

Landfill Incinerate

Non-haz. landfill Haz. landfill Non-haz. incinerate Haz. incinerate

A1e51 A1e52 A1e61 A1e62

Pre-Clean Dismantling (A3)

Hazardous dismantling (destructive/ non-destructive)

Sort (A4)

Non-hazardous sort Hazardous sort

Clean reusable comp. (A5) Inspect reusable comp. (A6)

Clean non-haz. reusable comp. Clean haz. reusable comp. Inspect non-hazardous reusable comp.

Inspect hazardous reusable comp. Shredding Collect reusable comp. (A8) Collect recyclable mat. (A9) Store reusable comp. (A1a) Store recyclable mat. (A1b) Transport reusable comp. (A1c) Transport recyclable material (A1d) Manage waste (A1e) Collect non-haz.reusable comp. Collect haz.reusable comp. Collect non-haz. recyclable mat. Collect haz.recyclable mat. Keep records Keep storage Keep max. storage Keep records Keep storage Keep max. storage Non-haz. transport of reusable comp. Haz. transport of reusable comp. Non-haz. transport of recyclable material Haz. transport of recyclable material Collect waste from disassembly stations Store waste for landfilling Store waste for incineration Transport waste to final destination

A612 A621 A622 A7 A81 A82 A91 A92 A1a1 A1a2 A1a3 A1b1 A1b2 A1b3 A1c11 A1c12 A1c21 A1c22 A1d11 A1d12 A1d21 A1d22 A1e11 A1e12 A1e21 A1e22 A1e23 A1e31 A1e32 A1e33 A1e4

Step 2 - Identify and Order all the Necessary Cost Drivers and Consumption Intensities The purpose of this step is to identify the cost drivers and corresponding consumption intensities that are necessary to find the cost of the consumption of activities with the desired accuracy. The cost of the consumption of a specific activity is the cost driver(s) multiplied with the consumption intensity. The total costs is found as the sum of the costs of all the activities that the design solution would impose. The properties of ABC depend on the cost drivers chosen. Bad cost drivers may give bad cost estimates. An example of a bad choice would be if a unit level cost driver (e.g., mass per unit) was chosen to keep track of a batch level activity (e.g., inspection). Having identified the cost drivers, the consumption intensities for each cost driver should be determined next. Furthermore, uncertainties in cost drivers and consumption intensities should be modeled at this stage. It would be beyond the scope of this paper to list all cost drivers and associated consumption intensities with the uncertainty distributions used in the demanufacture cost model. Illustrative examples are listed in Tables 3 and 4. From Table 1 we see that activities A411 and A412 are sorting of non-hazardous reusable components and recyclable material, respectively. As can be seen in Tables 3 and 4, normal and triangular uncertainty distributions are assigned, respectively, to both the cost drivers and consumption intensities in terms of a mean and left and right deviation. The Crystal Ball software allows twelve different distribution types. Among them the triangular and normal distributions as used in Table 3 and 4, but also other types such as the uniform, Weibull, exponential, and user-defined custom distributions. Step 3 - Identify the Relationships between Cost Drivers and Design Changes. The next step in our method is to identify the relationships between cost drivers and design changes. The relationships between cost drivers and design parameters are the crux of a design decision support model, because they capture how much a change in one or more design parameters will affect the consumption of the activities, i.e., the cost. TABLE 3 COST DRIVER EXAMPLES.
Activity A411 A412 Cost driver Distribution Direct labor Direct labor Normal [h/batch] Normal [h/batch] Mean 20.0 15.0 Left dev. 10.0 10.0 Right dev. 30.0 20.0

TABLE 4 CONSUMPTION INTENSITY EXAMPLES.

Activity A411 A412 Cost driver Direct labor Direct labor Consumption intensity Distribution Triangular [$/h] Triangular [$/h] Mean 20.0 20.0 Left dev. 18.0 18.0 Right dev. 23.0 23.0

Gray cells represent lowest level activities. The notation refers to the shaded cells. haz. = hazardous, comp. = components, mat. = materials

A key objective for using an Activity-Based Cost model in design is to identify how changes in different design parameters affect the cost and consumption of the activities. The level of detail and sophistication needed in modeling the relationships between cost drivers and design parameters depends on the purpose and usage of the cost model. In Figure 4, different uses

288

and ways of modeling the effects of design changes on the cost are illustrated. We identify two distinct uses of an ActivityBased Cost model: 1) Evaluation of a number of discrete designs in order to identify the economically best design, that is, the cost of a number of alternative designs is determined and a selection of a design is made based upon the result. Therefore, we do not have to model the relationships between design parameters and cost drivers explicitly. This approach works at the activity level (the top level in Figure 4) and selection of the most cost effective design is the primary purpose. 2) Identification of optimal values for continuous design parameters, that is, a given design is modified through, e.g., mathematical optimization in order to identify the ideal values of a number of design parameters. Rather than selection of designs, modification of an existing design is the primary purpose. This is shown in the gray area of Fig. 4. In order to improve design parameter values (usage 2), it is necessary to identify the effect the design parameters have on the cost drivers. The highest amount of detail is obtained if these effects are quantified in detailed mathematical Cost
A 311 A 11 A 412 C A 121
Dismantling

Driver/Design Parameter relationships which link design changes at the property/dimension level (bottom level in Figure 4) to the cost drivers. This approach allows us to modify designs on a very detailed level and enables cost minimization directly by computing the set of the most cost effective design parameter values using, e.g., optimization algorithms. However, the usage of such mathematical Cost Driver/Design Parameter relationships can become extremely cumbersome in the design of complicated systems where there are a) many relationships and b) many changes made over time in the relationships. In Figure 4, a solution is represented by introducing a level in the middle - where we do not keep track of design properties and dimensions, but rather keep track of how the design properties and dimensions affect specific actions. The aggregated effect on these actions is then transformed into an aggregated effect on the activities and the cost drivers. In the next section, this concept of using action charts to identify the aggregated effect of design changes on activities in the ABC model is discussed in detail.

A 312

A 411

A 51

A 611

A 612

A 81

A 1a2

A 1c11

A1c12

A 91
Landfill

A 1b2

A 1d11

A 1d12

A 313

B A1e22 A 1e4 A1e51 A 1c22

F A
314

122

Incinerate

A1e32

A 1e4

A 1e61 A 1c21

B
Shredding

A7 A 1a1
No Yes

A 13

Non-hazardous

A 1e11 A 321 A 421 A 52 A 621 A 622 A 82

A 1a2 A 1a3

A2

A 322

No

D
Hazardous

E
No

Yes

A 323

A 422

A 92 A 1e22

No

A 1b3

A 1b1

A 1b2 A 1d21

Dismantling

A 324 A7
Shredding No

B Decisions nodes: A B Hazardous or non-hazardous material or component Type of demanufacturing process Type of disassembly tools Less than 100 kg/month of hazardous waste?

D A 1e12
Landfill

Yes

Yes

A 1e23

A 1e4

A 1e52 A 1d22

No

A 1e21 F
Incinerate

A 1e32
Yes No

Store for less than 90/180 days?

No

E A 1e31

Yes

A 1e33

A 1e4

A 1e62

Landfill or incinerate

FIGURE 3 DEMANUFACTURING ACTIVITY NETWORK. THE ICONS ARE THE SAME AS USED IN [11] . TABLE 2 DEMANUFACTURING ACTIVITIES AND COST DRIVERS.
Activity A11 A121 Cost drivers Buy back No. of batches Activity A52 A611 Cost drivers Direct labor; Tooling time; No. of set-ups Direct labor Activity A1c22 A1d11 Cost drivers No. of batches; Fuel; No. of set-ups No. of batches

289

A122 A13 A2 A311 A312 A313 A314 A321 A322 A323 A324 A411 A412 A421 A422 A51

Number of batches; Fuel Volume Direct labor; Tooling time Direct labor Direct labor; Tooling time Direct labor; Tooling time Direct labor; Tooling time Direct labor; No. of set-ups of safety equipment Direct labor; Tooling time; No. of set-ups Direct labor; Tooling time; No. of set-ups Direct labor; Tooling time; No. of set-ups Direct labor Direct labor Direct labor; No. of set-ups Direct labor; No. of set-ups Direct labor; Tooling time

A612 A621 A622 A71 A81 A82 A91 A92 A1a1 A1a2 A1a3 A1b1 A1b2 A1b3 A1c11 A1c12 A1c21

Direct labor; No. of tests Direct labor; No. of set-ups Direct labor; No. of set-ups; Number of tests Tooling time Tooling time Tooling time; No. of set-ups Tooling time Tooling time; No. of set-ups Direct labor Volume Direct labor; Volume Direct labor Volume Direct labor; Volume No. of batches No. of batches; Fuel No. of batches; No. of set-ups

A1d12 A1d21 A1d22 A1e11 A1e12 A1e21 A1e22 A1e23 A1e31 A1e32 A1e33 A1e4 A1e51 A1e52 A1e61 A1e62

No. of batches; Fuel No. of batches; No. of set-ups No. of batches; Fuel; No. of set-ups Tooling time Tooling time; No. of set-ups Direct labor Volume Direct labor; Volume Direct labor Volume Direct labor; Volume No. of batches; Fuel Volume; Mass Volume; Mass Volume; Mass Volume; Mass

Evaluation of Alternative Designs Increased Level of: Detail Model complexity Design change traceability Activity Network Modification of Design Activity Network Action Charts

Focus of interest: The effect the actions have on the activities.

The effect the design properties, (e.g. Mass) have on the actions.

Activity Network Action Charts Cost Driver/Design Parameter Relationships

Properties/dimensions etc. of the design.

FIGURE 4 DIFFERENT LEVELS OF MODELING Tracing Costs to Design Using Action Charts What are action charts? Activities are formed by grouping actions that have a logic connection together [4, 5] . This is done mainly because it is impossible to achieve credible cost information for every step in a process. It would require an enormous amount of cost drivers and consumption intensities, associated with an even larger amount of possible uncertainty. Thus, actions that occur in a process are grouped into activities. Forming the activities in a way that roughly describes the process is advantageous because this will make the ABC model much easier to understand and use as it coincides with our perception of the process. The grouping of several process actions into a smaller number of activities opens up the possibility of designing a model with a generic set of low level activities which has design specific inputs - the actions. This is a powerful approach since any process can be described with a set of activities that will always be present, no matter what product we are dealing with. The definition of activities is a function of the desired degree of generality, accuracy and traceability. Increased generality will in general give decreased accuracy and traceability. The usage of aggregated actions is a way of capturing how design changes affect the costs and revenues. It is an approach in between a) not modeling any relations at all and merely assessing cost of designs and b) modeling the relationships in detailed mathematical relationships and computing the most cost effective values of design parameters. By aggregating the actions in so-called action charts, we keep track of how the design properties and dimensions affect the actions. The aggregated effect on the actions is then transformed into an aggregated effect on the activities and the cost drivers. In Figure 6, a sample from a dismantling action chart of a telephone is shown, derived from disassembly charts outlined in [12] . Note

the level of details, which is typical for a good action chart. The less detailed an action chart is, the less suitable it is for design modifications. What purpose does the dismantling action chart serve? The action chart in Figure 6 allows the detailed documentation of a disassembly process. Manual disassembly can be considered as a single activity in an ABC cost model and a reduction of overall disassembly time is clearly advantageous. But on which product component should a designer focus? Time is not the only cost driver in disassembly. A different material or fluctuations in material prices also affect overall revenue. This kind of product design related information is embodied in the dismantling action chart. In essence, an action chart forms an interface between detailed product information and a general demanufacturing ABC model, in our case.

290

Notes: Qualitative assessments: Nondestructive: CE: Connecting Element, 1 => Bad, Yes => assembly can be dismantled without totally or SA: Sub Assembly, 2 => Below average, partially destroying any part of the (sub) assembly. SP: Single Part, 3 => Above average, No => assembly cannot be dismantled without totally or A/C: Accessibility; (4 is good, 1 is bad), 4 => Good partially destroying any part of the (sub) assembly. Std: Standard part, Mtl. Recycl.: Material Recyclability, Quantity, Time and Mass are assessed quantitatively.

Fatigue Dirt/ NondestCorr. structive

yes

yes

yes

yes

yes

yes

No

the consumption of the demanufacture activities. To support this step, we have implemented the entire model in Microsoft Excel 4.0 spreadsheet files on a Macintosh platform. A detailed description can be found in [13] . As stated before, we have chosen to use software called Crystal Ball which allows the definition of uncertainty distribution in spreadsheet cells and finds resulting uncertainty distributions numerically using a Monte Carlo simulation. The results obtained in this fashion for telehone demanufacture are discussed next. Step 5 Evaluate the Solution There are many assumptions made in the model (135 assumption cells in 10 model files). An important assumption is that the plant capacity is less than market demand. Another assumption worth noting is that a $1 buy back price is paid per telephone. Once the model is implemented in spreadsheet structure, several tools are available for evaluating the solution, results, and the effects of the assumptions and design decisions made. We illustrate two tools available, i.e., profitability distributions and sensitivity charts. Profitability Distributions. The resulting profitability distribution for the Dismantling scenario is presented in Figure 7. These distributions provide a good indication of what the effects of the uncertainties in the assumptions are and, in our opinion, provide more valuable information for designers than merely a single number. The mean is estimated to be -$2.30 if we pay, on average, $1.00 for the telephones. In other words, it is not economically feasible to dismantle the present telephone design. The results from the Shredding scenario are presented in Figure 8. The shredding option is also economically infeasible, but we see that the revenues nearly balance the costs if we would not have paid $1.00 for old telephones.
Cell F415
.033

0.0030

0.0060

0.0140

0.0580

0.0001 4 4 4 Std SP 1 From 26; spring Steel

Recycl.

Materials Recyclability

Mtl.

Material

Reusability (Product-Recyclability)

Time Tool for Abrasion Stnd./ [sec.] Assembly Special Pt.

And so forth

Std

Std

Std

Std

Snap on

Snap on

30

15

Snap on

Glue on

Force

Disassemblability

Crowbar

Crowbar

Quan- Type A/C tity

Tool

Crowbar

CE

SP

SP

SA

SA

SA

Handset Disassembly

Disassembly activity and object

Snap off basehandset cable

Remove mic. cables and CB

Disassemble circuit board

Remove top from bottom

From 26; circuit board

Remove Mass

From 26; cover

Name

SP

Std

Thermoset

copper, plastic

Lead

ABS

Mix

0.0001

Mass

[kg]

Forecast: Dismantling Unit-profitability Frequency Chart 999 Trials Shown

33

No.

27

28

29

30

.02

24.7

FIGURE 5 A PHONE DISMANTLING ACTION CHART.

.01 16.5

In order to support the ABC model, each action must be associated with a sufficient set of information. In our opinion, the following set of action information seems to be sufficient input for each activity in a demanufacturing model: All actions related to the specific activity, and For each action: the number of units, the mass and material composition for each unit, the time to perform an action, the tools used, the process efficiencies, the hazardousness of the units, and the danger in performing the actions. Uncertainty distributions can be assigned in the action chart, e.g., for specific disassembly times, and the effect of variations in a product design can be traced. In the next section, we discuss the results obtained from using a demanufacturing action chart in our demanufacturing ABC model. Step 4 Find and Minimize the Cost of the Consumption of Activities. Having created the ABC model, defined the relationships between design properties and cost using action charts, and modeled the associated uncertainty, we now proceed to step four of our method, i.e., find (and minimize) the cost associated with

.00 .00 -3.50 -2.94 -2.38 [$/unit] -1.81 -1.25

8.25 0

FIGURE 7 DISMANTLINGPROFITABILITY DISTRIBUTION

Forecast: Shredding Unit-profitability Frequency Chart

Cell F416
.032

998 Trials Shown

32

.02

24

.01

16

.00 .00 -2.00 -1.50 -1.00 [$/unit] -0.50 0.00

8 0

FIGURE 8 SHREDDINGPROFITABILITY

291

DISTRIBUTION Taking into account that many assumptions have been made and the fact that the upper limit of the distribution is close to a break even situation, we should not rule out the possibility of a break even situation in the real world. Identifying Largest Cost Contributors Using Sensitivity Charts. Assuming that we want to pursue dismantling of telephones, what changes should we make in order to boost profitability? One might argue that a product design should be improved according to Design for Disassembly guidelines, but it may be that process aspects such as labor and interest rates are far more significant than the product design. We employ sensitivity charts to gain insight in these issues. In Figure 9, a (shortened) sensitivity chart for the dismantling process model is given. Such a sensitivity chart is generated for each simulation by the Crystal Ball software and is based on the so called Spearman Rank Correlation and measures the degree to which assumptions and forecasts change together. The larger absolute value of the correlation coefficient, the stronger is the relationship. Positive coefficients indicate that an increase in the assumption cell is associated with an increase in the forecast cell.
Sensitivity Chart Target Forecast: Dismantling Unit-profitability
Telephone 2 Pay-back Price A612 Number of Tests [$/test] A13 Volume [$/ft^3] No. 2 Time [sec.] A311 Direct Labor [$/h] No. 11 Unit-revenue [$/unit] Buisness days per week Motor Control House [$] No. 20 Time [sec.] -.83 -.46 -.11 -.09 -.09 .09 .09 -.08 -.07

No. 11 Unit-revenue [$/unit], associated with reusable speakers, and Business days per week. As can be seen, product design as well as process factors affecr the cost. Most gain can be made by reducing the buy-back price, but a handset redesign is arguably a pressing product design issue. Sensitivity analyses for other scenarios and using perfect process information are documented in [13] . The more trials performed, the more useful is the sensitivity chart. The reason is that the variability of the correlation coefficient estimates decreases as the number of trials increases because the number of degrees of freedom increases, and that the probability for correlation by chance decreases, see [14] . CLOSURE AND FUTURE WORK In this paper, we discussed how an Activity-Based Cost model can be developed and used in design under presence of uncertainty. We highlighted the modeling and propagation of these uncertainties through a cost model using commercially available software. The inclusion of uncertainty and usage of Monte Carlo simulation provides the capability to identify those process and product design aspects that contribute most to the cost using so-called action charts. An action chart represents a group of associated actions which together form an activity. We used disassembly actions as an example. The subsequent use of sensitivity charts facilitates identification of the most cost inefficient parts. Profitability distributions and sensitivity charts assist in identifying: where we should focus our data collection efforts, e.g., because payback prices and direct labor have such significant influence (see Figure 9) we should collect more accurate data in these areas. where we should focus our design efforts, e.g., with respect to the telephone design, we should focus on the removal of the top from the base because it is the largest cost contributor. It should be emphasized that especially in the early stages of design, the identification of the largest cost contributors and critical factors is more important than the actual cost. With respect to validity, we have attempted to make our demanufacturing model as realistic as possible, e.g., the EPA regulations for storing hazardous waste have been incorporated. The model presented in this paper seems to give reasonable results. In fact, the same model has been applied to assess and trace the cost of demanufacturing a car and the results were compatible with real world experiences [13] . We have most likely underestimated the total overhead cost for a demanufacturer, so we would expect that the true costs are higher than estimated. Another aspect to take into account is that most of the plastic revenue information, is based on [15] and reported to be from 1990. Our focus in future work is on the following key aspects. We are continuing to validate, improve, and expand our method and models. One of our objectives is to utilize the tractability of the costs and uncertainty through an activity network and identify which life-cycle activities are truly critical in a product s lifecycle. In the long term, we seek to utilize the ABC method not only for monetary cost assessments, but also for life-cycle assessments of environmental impact in terms of matter and energy consumption. A model which provides an environmental

.....

.....

A622 Direct Labor [$/h] Brass [$/kg]

-.06 -.05 -1 -0.5 0 0.5 1

Measured by Rank Correlation

FIGURE 9 SENSITIVITY CHART FOR DISMANTLING PROFITABILITY As can be seen in Figure 9, the chart allows us to pin point the factors/assumptions for the Dismantling scenario that correlates most with the forecast cell. This facilitates studying the importance of the different assumptions in the model. It also indicates what cells should be updated with better, more accurate information. The major cost and revenue triggers are: Telephone 2 buyback price, A612 Number of tests [$/test]3, number of tests to check if components are reusable or not, A13 Volume [$/ft^3], storage consumption intensity for units in to the demanufacturing plant, No. 2 Time [sec.], time to perform action number 2 from the phone action chart - Remove top from bottom of the handset disassembly, a product aspect to be redesigned, A311 Direct labor [$/h],
3

From the unit, [$/test], we understand that this is the consumption intensity of activity A612 and not the cost driver.

.....

292

impact assessment in terms of energy and matter consumption and emission can be used for exploring the global (societal) environmental impact of engineering products and processes. E.g., recycling products is nice, but recycling may cost more energy than a disposal process. The use of an ABC approach may overcome some of the difficulties associated with conventional Life-Cycle Assessment/Analysis tools, e.g., the cumbersome amount of work involved and the lack of common standards. In our opinion, an activity-based Life Cycle Assessment is most easily done for energy where we have a single unit. It will be more difficult for materials where, e.g., we need to distinguish different grades of toxicity. Although we have focused on demanufacturing as the area of application, It should be noted that our method, as most ABC based methods, is generic in that it can be applied whenever the activities are described in sufficient detail to have cost drivers assigned. REFERENCES 1. Congress, U. S., Green Products by Design: Choices for a Cleaner Environment, OTA-E-541, Office of Technology Assessment (1992), Washington, D.C. 2. Keoleian, G. A. and Menerey, D., Sustainable Development by Design: Review of Life Cycle Design and Related Approaches, Air & Waste, Vol. 44, May (1994), pp. 644-668. 3. Brooks, P. L., Davidson, L. J. and Palamides, J. H., Environmental compliance: You better know your ABC s, Occupational Hazards, February (1993), pp. 41-46. 4. Cooper, R., ABC: A Need, Not an Option, Accountancy, September (1990), pp. 86-88. 5. Cooper, R., Five Steps to ABC System Design, Accountancy, November (1990), pp. 78-81. 6. Emblemsvg, J. and Bras, B. A., Activity-Based Costing in Design for Product Retirement, Proceedings 1994 ASME Advances in Design Automation Conference, DE-Vol. 69-2, Minneapolis, Sept. 11-14, ASME, (1994), pp. 351-362. 7. Turney, P. B. B., How Activity-Based Costing Helps Reduce Cost, Journal of Cost Management for the Manufacturing Industry, Vol. 4, No. 4 (1991), pp. 29-35. 8. O Guin, M., Focus The Factory With Activity-Based Costing, Management Accounting, Feb. (1990), pp. 36-41. 9. Raffish, N. and Turney, P. B. B., Glossary of ActivityBased Management, Journal of Cost Management for the Manufacturing Industry, Vol. 5, No. 3 Fall (1991). 10. Navin-Chandra, D., ReStar: A Design Tool for Environmental Recovery Analysis, 9th International Conference on Engineering Design, The Hague, August 17-19, Heurista, Zurich, Switzerland, (1993), pp. 780-787. 11. Greenwood, T. G. and Reeve, J. M., Activity Based Cost Management for Continuous Improvement: A Process Design Framework, Journal of Cost Management for the Manufacturing Industry, Vol. 5, No. 4 (1992), pp. 22-40. 12. Beitz, W., Suhr, M. and Rothe, A., Recyclingorientierte Waschmaschine (recycling-oriented washing machine), Institut fr Maschinenkonstruktion - Konstruktionstechnik, Technische Universitt, Berlin (1992). 13. Emblemsvg, J., Activity-Based Costing in Designing for the Life-Cycle, MS Thesis, G.W.W. School of Mechanical Engineering, Georgia Tech, Atlanta, Georgia (1995).

14. Hines, W. W. and Montgomery, D. C., Probability and Statistics in Engineering and Management Science, John Wiley & Sons, Inc., (1990). 15. Dieffenbach, J. R., Mascarin, A. E. and Fisher, M. M., Modeling Costs of Plastics Recycling, Automotive Engineering, October (1993).

293