Development of an AHP–CBR evaluation system for remanufacturing: end-of-life selection strategy

Abstract

This paper presents our research works on developing an evaluation system for remanufacturing. This paper has two objectives. The first objective is to evaluate the product end-of-life (EOL) by integrating an analytical hierarchy process (AHP) with case-based reasoning (CBR). The nearest neighbourhood (NN) algorithm was used to retrieve a past case that is closest to the current case under consideration. The AHP, which allows the pair-wise comparison and consistency judgement, was used to determine the weight in NN. The second objective is to evaluate EOL of parts and components. The integration of an economical cost model and an environmental cost model was used to determine the EOL of parts and components. An example is presented and discussed. The predicted results of the complete method were compared with those of the previous study. The results showed that the integration of the AHP–CBR method has reached a good correspondence with the established methods.

Keywords:

• case-based reasoning
• AHP
• remanufacturing
• end-of-life
• economical cost
• environmental cost

1. Introduction

The environmental impact and ecological issues are always in the priority of many governments worldwide. Public concerns about diminishing natural resources, limited landfill space and hazardous waste disposal have prompted the legislators to place the end-of-life (EOL) product recovery issues to the manufacturers (The European Parliament and the Council of European Union 2000, 2003a,b, Herrmann et al. 2006). To survive the competition, manufacturers have to produce products that are safe and friendly to the environment. Hence, many manufacturers have developed several methods mainly dedicated to disassembly and recyclability of a product such as design for environment, design for disassembly, design for recycling and so on (Kaebernick et al. 2002).

Several studies have focused on the selection of EOL strategies. Rose (2000) developed a decision tree model called end-of-life design advisor (ELDA) to help us determine the EOL strategies. The ELDA contains the characteristics that influence the EOL. These characteristics are wear-out life (WOL), technology cycle (TC), level of integration (LOI), number of parts (NOP), reason for redesign (RFR) and design cycle (DC). Zhang et al. (2000) adopted an analytical hierarchy process (AHP) to find the best EOL strategy. The AHP-based evaluation considered environmental impact, cost and reclaimed materials as major criteria for strategy determination.

Several studies applied multicriteria decision analysis (MCDA) approach on the selection of EOL strategies. Bufardi et al. (2004) applied MCDA to select the best EOL alternative. Jun et al. (2007) applied multi-objective evolutionary algorithm to optimise the EOL selection strategy. Staikos and Rahimifard (2007) integrate AHP, life-cycle analysis and benefit–cost analysis to find the EOL strategies. Fernandez et al. (2008) developed a fuzzy approach to find the EOL strategies. Iakovou et al. (2009) developed ‘Multicriteria Matrix’ to evaluate the EOL selection strategies of a product. The disadvantage of these methods is that it required even experienced decision makers to understand the problem, the feasible alternatives, different outcomes, conflicts between the criteria and level of the data uncertainty in finding the best EOL strategy. Moreover, a few studies provide a systematic evaluation method for inexperienced decision makers to evaluate the EOL strategy. To cope with these limitations, we consider the case-based reasoning (CBR) approach to find the best EOL strategy at product level. The advantage of CBR is that it provides a quick approach in finding the best EOL strategy of EOL products regardless of the experience level of decision makers.

CBR is one of the artificial intelligence (AI) tools, which involves the activities of storing previously solved problem in a case base, retrieving a similar previous case based on a set of product characteristics and finally adapting the solution of the previous case to the new case (Kolodner 1991, Aamodt and Plaza 1994, Chougule and Ravi 2005). Zeid et al. (1997) and Veerakamolmal and Gupta (2002) employed CBR in planning for disassembly (PFD) and disassembly process planning (DPP). The episodic memory organisation packets were used to store and index the PFD and DPP plans so that it can be retrieved by comparing it with the new case. However, the trial and error approach to match a similar case requires a lot of time and produces poor accuracy in finding a similar case. To overcome this limitation, we adopted nearest neighbourhood (NN) algorithm, the advantages of which is to enable us to assess similarity numerically, in finding a similar case.

The nearest neighbourhood (NN) algorithm is the most popular method used to retrieve a similar case. It involves specifying the parameters and weights of product characteristics, based on which the nearest case is identified and retrieved (Kim and Han 2001). The most important and difficult task in NN is setting the weights (Chougule and Ravi 2005). Shih et al. (2003, 2006) applied a trial and error approach in setting the weights in NN algorithm to find the closest stored cases to the input cases. However, there is a need to develop a systematic way to determine the weights for similarity function in the CBR approach. This problem can be solved by applying the AHP. Park and Han (2002) used the integration of AHP with CBR to predict bankruptcy in financial institutions. However, there has been a lack of studies considering the integration of AHP with CBR in selecting the EOL alternatives. The AHP provides a structural approach for assigning the weight. In AHP, only two attributes are compared to calculate the weight (Saaty 1994). Another advantage of AHP is that it provides a consistency of judgement during pair-wise comparisons.

On the other hand, many literatures have presented results on disassembly planning approach for estimating the cost of EOL. Gungor and Gupta (1997), Feldmann et al. (2001) and Desai (2002) systematically investigated the disassembly sequence and related operations so that disassembly cost can be accurately estimated. Lee et al. (2001) incorporated both cost and environmental impact estimation into EOL disassembly charts model. The authors tried to find the alternatives of EOL that can both maximise profit and minimise environmental impact. Additionally, the authors introduced EOL values that compare the costs of new parts or components cost of EOL.

In summary, the studies that are concerned with environmental issues such as product recovery, remanufacture and disassembly are increasing rapidly. The manufacturers tend to treat these issues separately (Ilgin and Gupta 2010). Unfortunately, few studies have dealt with the product recovery, remanufacture and disassembly simultaneously. Additionally, many manufacturers have adopted recycling as their strategy for treating the EOL of products. However, remanufacture of components, parts or entire products is a more efficient EOL strategy than recycling (Steinhilper 1998, Hata et al. 2000, Zussman 2000, Kaebernick et al. 2002). Although some previous works have addressed the issues for selecting the EOL strategies, less attention has been paid to integrate AI tools for remanufacturing. Many literatures have focused on disassembly planning to evaluate the EOL at the part and component level. However, few studies have been done on considering the environmental cost factors to evaluate the EOL at this level.

These issues motivate us to develop an integrated system that considers the EOL at both product and parts levels. At the product level, we propose the integration of AHP and CBR. The AHP aims at providing the weight to be used in CBR. At the parts and components level, we propose the integration of economical and environmental costs of the remanufacturing, recycling and landfill values (EOL values). This integration aims at providing a relatively rough but quick estimation of an EOL option at the parts and components level. The economical cost deals with which EOL option is the most economic for a part or a component. The environment cost concerns about which EOL option is the most environmentally friendly option for a part or a component. We compare the EOL value of remanufacturing to the EOL value of recycling. The maximum EOL value is the most suitable EOL option for a part or a component of the remanufactured product.

This paper is organised as follows. Section 2 presents an overall picture of the integrated AHP–CBR method. Section 3 presents examples. In this section, the results of the integrated AHP–CBR method are compared with those of an established traditional method of previous works. In Section 4, we discuss the results. Finally, in Section 5, we draw conclusions and provide the direction for further research.

2. An integrated approach for EOL strategy selection

2.1 Overview of the integrated approach

The integrated approach is categorised into two stages. The first stage is to aim at selecting the product EOL option using an integration of the AHP–CBR method. The second stage is to select an EOL option of its components and parts based on the integration of an economical and environmental costing. The outline of the AHP–CBR method is illustrated in Figure 1.

Figure 1 Overview of the integrated AHP–CBR approach.

As shown in Figure 1, we begin the evaluation process with the gathering of the information of an EOL product. The information consists of the parameters that will be used to evaluate an EOL option of the product at the first stage of the AHP–CBR method. In the first stage, we gathered the information on the classification of the EOL product. Next, the characteristics of the parts and components are identified. The characteristics are based on the ELDA developed by Rose (2000). Then, the weight for these characteristics was given based on the AHP. After that, the nearest neighbourhood (NN) algorithm was applied to find the smallest difference between the input case and the stored cases.

The stored case with a maximum percentage is considered as the closest case to the input case. The range of acceptable values to measure the degree of similarity is usually in the range between 80 and 100% (Sun and Wang 2006). However, we justified 85% as a cut-off value for measuring similarity based on the established studies by Park and Han (2002), Li and Ho (2009) and Li et al. (2010). These studies demonstrated that the error when retrieving similar cases in the database is significantly low in comparison with the new case when the similarity value is more than 85%.

If the EOL option of the retrieved case is remanufacturing, the input case will adopt the EOL option of the retrieved case. Next, this product will enter into the second stage. At the second stage, the evaluation of an EOL option at the part and component level is implemented based on the estimation of EOL value. The maximum value of any EOL value is considered the most suitable EOL option for the evaluated parts or components.

2.2 Product level EOL selection strategy

For selecting the strategy of an EOL product, we employed an integrated approach of AHP–CBR. Fifty cases of the successful product EOL from Rose (2000) are adopted as the stored cases. The indices of these cases are defined based on ELDA (Rose 2000). In ELDA, candidates for EOL strategies are reuse, repair, remanufacturing, disassembly with material recovery, shredding with material recovery and disposal. In the AHP–CBR method, the EOL selection strategy comprises the following steps:

• Identifying the attributes for case indices.

• Identifying the case indices for case retrieval.

• Determining the weights using an AHP.

• Retrieving the similar cases from the case base.

In this study, five cases are being treated as new cases. The remaining 45 cases are used as stored cases in AHP–CBR.

2.2.1 Identifying attributes of case indices

The attributes that influence the case indices are adopted from Rose (2000). These attributes are market demand (MD), technological changes (TCh), legislation and product design (PD). The descriptions of these attributes are given in Table 1.

Table 1 The attributes that influence the case indices (Rose 2000).

Attributes	Description
Market demand (MD)	Any demand or pressures from the consumers about how the product should feature. For instance, the consumer wants a durable product with a leading edge technology
Technological changes (TCh)	The TCh influencing consumer to replace their ageing product with new models. For instance, the cellular phone
Product design (PD)	Any design based on the consumer's voice. It has a major influence on the wear-out life, reason for redesign and technology cycle
Legislation (L)	Any legislation that concerns the environment and sustainability. It includes the EU directives for EOL products, EPA and so on

2.2.2 Identifying case indices

The new case indices are defined based on the ELDA (Rose 2000). These indices are WOL, TC, LOI, DC, NOP and RFR. The descriptions of these indices are given in Table 2.

Table 2 Case indices for product level EOL strategy selection (Rose 2000).

Case indices	Definition
Wear-out life (WOL)	Duration from which the product is purchased until it no longer meets the original functions. For example, the WOL of automobile is approximately 10–20 years
Technology cycle (TC)	The period during which the product will be on the leading edge of technology before the new technology makes the original product obsolete. For example, the TC of a computer is approximately 6 months to 1 year
Level of integration (LOI)	Interrelation between modules and functions. If there are many unique functions, for each module, the LOI is high
Number of parts (NOP)	Number of assemblies in the product relevant for EOL treatment
Reason for redesign (RFR)	depends on customer demand, competitor behaviour and scientific progress. It is divided into original design, evolutionary design (either function improvement or aesthetic change) and feature change (either function improvement or aesthetic change)
Design cycle (DC)	The frequency with which a design team designs a product. For example, the DC of an automobile is about 2–4 years

The weights for these attributes were given using an AHP evaluation. This is further discussed in Section 2.2.3.

2.2.3 Determination of weights

The purpose of weight is to define the importance of each case index relative to others in retrieving the most appropriate case. An AHP evaluation was employed for calculating the weight in CBR. It involves the following activities:

• Developing an AHP structure of the defined problem.

• Conducting questionnaires.

• Determining the weight of the case indices via pair-wise comparison.

• Checking the consistency of pair-wise comparison.

2.2.3.1 Development of an AHP structure

The attributes are structured in a hierarchical form such as MDs, TCh, PD and legislation. The individual attributes are the case indices defined in Table 2. The structure of AHP is illustrated in Figure 2.

Figure 2 The structures of AHP in AHP–CBR method.

2.2.3.2 Conducting a questionnaire

A set of questions is designed to determine the importance of each attribute and each individual attribute. This questionnaire was carried out by means of pair-wise comparison. The questionnaire consists of two parts:

1. Attribute level consists of four attributes in the hierarchy shown in Figure 2. The MD, TCh, PD and government policies are compared with each other in order to determine their relative importance. The questionnaire form is shown in Table 3.

   Table 3 An example of the questionnaire at the attribute level.

   i	Absolutely important	Intermediate value between 7 and 9	Very strongly important	Intermediate value between 5 and 7	Strongly important	Intermediate value between 3 and 5	Weakly important	Intermediate value between 1 and 3	Equally	Intermediate value between 1 and 3	Weakly important	Intermediate value between 3 and 5	Strongly important	Intermediate value between 5 and 7	Very strongly important	Intermediate value between 7 and 9	Absolutely important	j
   Market demand (MD)	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Technological changes
   MD	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Product design
   MD	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Government policies
   Technological changes (TCh)	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Product design
   TCh	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Legislation
   Product design	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Legislation
   Note: Please circle(○)/tick(✓) which of the two characters in column, i and j, you think is more important and to what extent. When you are considering the EOL of the product, which one do you think is more important and how many times more important than the other.

2. Individual attribute level consists of six attributes in the hierarchy shown in Figure 2. WOL, TC, DC, NOP, RFR and LOI are compared with each other to determine their relative importance. The questionnaire form is summarised in Table 4.

   Table 4 An example of the questionnaire at the individual attributes level.

   i	Absolutely important	Intermediate value between 7 and 9	Very strongly important	Intermediate value between 5 and 7	Strongly important	Intermediate value between 3 and 5	Weakly important	Intermediate value between 1 and 3	Equally	Intermediate value between 1 and 3	Weakly important	Intermediate value between 3 and 5	Strongly important	Intermediate value between 5 and 7	Very strongly important	Intermediate value between 7 and 9	Absolutely important	j
   Wear-out life (WOL)	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Technology cycle
   WOL	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Design cycle
   WOL	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Number of parts
   WOL	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Reason for redesign
   WOL	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Level of integration
   Technology cycle (TC)	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Design cycle
   TC	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Number of parts
   TC	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Reason for redesign
   TC	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Level of integration
   Design cycle (DC)	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Number of parts
   DC	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Reason for redesign
   DC	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Level of integration
   Number of parts (NOP)	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Reason for redesign
   NOP	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Level of integration
   Reason for redesign	9	8	7	6	5	4	3	2	1	2	3	4	5	6	7	8	9	Level of integration
   Note: For MD factor, when you design the product and you consider its EOL, which one you think is more important and how many times more important.

2.2.3.3 Calculation of the relative weight

The main problem in CBR is how to set the weight of each attribute relative to others in retrieving the most similar cases. This difficulty can be overcome by pair-wise comparison in AHP. It involves the construction of a square matrix A1 _n × n in which the set of attributes are compared pair-wise (Saaty 1994, Liu and Inooka 1998). A1 _n × n can be represented as follows:

where a _ij is the element in the pair-wise comparison matrix. It gives a comparative importance of a criterion i with respect to criterion j (see, Table 5). In matrix A1 _n × n , a _ij  = 1 when i = j and a _ij  = 1/a _ij when i ≠ j.

Table 5 Numerical rating for pair-wise comparison (Liu and Inooka 1998, Saaty 1994).

Intensity of importance	Definition
1	Equal importance
3	Moderate importance
5	Strong importance
7	Very strong importance
9	Absolute importance
2,4,6,8	Intermediate values between the two adjacent judgements

The weights associated with each attribute are calculated by geometric mean GM _i . The geometric mean can be expressed as follows:

where n = 1, 2, 3, …, i.

Then, the geometric mean is normalised in order to obtain the relative weight, w _i , of each attribute. The normalised weight can be expressed as follows:

2.2.3.4 Checking the consistency of pair-wise comparison

The consistency ratio is calculated using the maximum eigen value λ_max (Saaty 1994, Chougule and Ravi 2005). The consistency index and consistency ratio can be represented as follows:

where n is the size of a comparison matrix.

The consistency ratio is acceptable if the value is less than or equal to 0.10 (10%). Otherwise, the result is considered inconsistent and the pair-wise comparison has to be repeated again. The random index (RI) is shown in Table 6.

Table 6 Random index (Saaty 1994).

n	1	2	3	4	5	6	7	8	9	10
RI	0	0	0.58	0.9	1.12	1.124	1.32	1.41	1.45	1.49

2.3 Retrieving the similar cases

The software was developed using Visual C# for the purpose of retrieving similar cases in the integrated AHP–CBR approach. This approach involves measuring similarity between the retrieved case and the input case. It is based on comparing and calculating a weighted sum of each parameter between the retrieved cases and the input case. The Euclidean distance is usually used to compare similarity between the retrieved cases and the input case (Vong et al. 2002). The Euclidean distance can be represented as follows:

However, this is valid for the numerical values. As we adopt the non-numerical values for the case indices of LOI and RFR, the following equations for dist(P _i ,Q _i ) and are applied:

where P _i is the input case of the ith index, Q _i is the retrieved case of the ith index, w _i is the weight of the ith index, is the summation of all w _i and is equal to 1, Sim(P _i ,Q _i ) is the function of similarity of the index between P _i and Q _i , dist(P _i ,Q _i ) is the distance between P _i and Q _i .

The stored cases are screened to determine the cases that have a similarity with the new case of more than 85%. If the similarity value is less than 85%, or the EOL option is at the product level, it is assumed that the new case should be considered for recycling or landfilling of the selective components. Additionally, if the similarity value is more than 85%, or the EOL option is not remanufacturing, it is assumed that the new case should not be remanufactured. The new case should be considered for the redesign of a component so that it can be remanufactured.

2.4 Selection strategy of part level EOL

2.4.1 Overview of the EOL strategy selection at the part level

In this stage, the economical and environmental costs were included for selecting the EOL options at the parts and components level. The model of EOL selection strategy at this level is illustrated in Figure 1. The required information at this stage is shown in Table 7.

Table 7 Required information at part level EOL strategy.

Required information	Unit
Part material
Part weight	kg
Cost of new part	$
Disassembly time	s
Sorting time	s
Cleaning time	s
Refurbish time	s
Testing time	s
Assembly time	s
Labour cost	$/h
Transport cost	$
Storage cost	$

The EOL values, i.e. remanufacturing value, recycling value and landfill value are calculated using the information in Table 7. Next, these values are compared with each other. The maximum value among them is considered as the best EOL option of a component or a part.

If the remanufacturing value is equal to or less than zero, the parts under consideration should not be remanufactured. The part is evaluated for recycling and landfill. If the recycling value is equal to or less than landfill value, the parts are automatically recommended for landfill. On the other hand, if the remanufacturing and recycling values are equal to or less than zero, the product or the components should be considered for landfill.

2.4.2 The structure of cost model

The EOL values are defined for evaluating the EOL options at the parts and components level. These values are remanufacturing value, recycling value and landfill value (Lee et al. 2001). These values can be represented as follows:

where C _pi is the life-cycle costs to produce new part i, C _Rem is the life-cycle costs to remanufacture part i, C _Rec is the life-cycle costs to recycle part i, C _Lf is the life-cycle costs to landfilling part i, C _Misc is the life-cycle of the miscellaneous cost, C _trans is the transportation cost of the EOL part i, C _Process is the re-processing cost of the EOL part i, C _Handling is the handling cost of the EOL part i, C _Storage is the storage cost of the EOL part i, V _Rem is the remanufacturing value of the EOL parts i, V _Rec is the recycling value of the EOL parts i and V _Lf is the landfill value of the EOL parts i.

As an environmental aspect has to be considered in selecting EOL options of parts and components, we can further elaborate Equations (9-12) as follows:

where C _PEnv is the environmental costs for producing new part i, C _EnvRm is the environmental costs for remanufacturing part i, C _EnvRec is the environmental costs for recycling part i, C _EnvLf is the environmental cost for landfilling part i, C _EnvMisc is the environmental cost of miscellaneous and C _Envtrans is the environmental cost for transportation.

2.4.2.1 Economical cost of new part

The economical cost of producing new parts covered the costs of forming a raw material to a finished product. The economical cost of producing a new part can be presented as follows:

where C _pi is the economical costs for producing a new part i, C _material is the material cost of part i, C _{manufacturing} is the manufacturing cost of part i and C _operation is the operational cost of part i.

The cost of processes, assembly labour and overheads are included in manufacturing cost. The operational costs covered the costs of marketing, distribution, administration and so on (Kaebernick et al. 2002).

2.4.2.2 Environmental cost of new part

The environmental cost of producing new parts covered the costs from the energy consumption to forming a raw material into a finished product. The environmental cost of producing a new part can be presented as follows:

where C _EnvMaterial is the environmental cost of a material to produce part i and C _{EnvManufacturing} is the environmental costs to manufacture part i.

2.4.2.3 Economical cost of remanufacturing part

The economical cost of remanufacturing parts covered the costs of disassembly, sorting, cleaning, refurbishing, testing and reassembly cost. It can be presented as follows:

where C _Rem is the remanufacturing cost, t _Dis is the disassembly time, t _Sort is the sorting time, t _Clean is the cleaning time, t _Ref is the refurbish/reprocess time, t _Test is the testing time, t _Assy is the reassembly time and is labour cost.

2.4.2.4 Environmental cost of remanufacturing part

The environmental cost of remanufacturing parts involves the costs of energy consumed for disassembling, sorting, cleaning, refurbishing and testing an EOL part. It can be presented as follows:

where E _Rem is the total cost of the energy used to remanufacture part i and is the energy consumption rate per unit, kW h.

2.4.2.5 Economical cost of recycling part

The economical cost of recycling parts can be presented as follows:

where M _Rec is the material cost per kg, w is the material weight.

2.4.2.6 Environmental cost of recycling part

The environmental cost of recycling parts involves the costs of energy consumed for disassembling, separating and shredding an EOL part. It can be presented as follows:

where C _RecE is the environmental cost driver for recycling part i per kg.

2.4.2.7 Economical cost of landfilling parts

The economical cost of landfilling parts can be presented as follows:

where is the landfill cost per kg.

2.4.2.8 Environmental cost of landfilling parts

The environmental cost of landfilling parts can be presented as follows:

where C _LfE is the environmental cost driver for landfilling part i per kg.

3. Results–application of AHP–CBR method to electronic product remanufacturing

The weights are determined using the AHP evaluation. Fifteen participants took part in the questionnaire survey. These weights are compared with those of the traditional CBR (T-CBR) approach.

Table 8 shows an overall weight distribution of the AHP evaluation. The consistency ratio for all individual attributes and the attribute level are within the acceptable limit of 10% (Saaty 1994).

Table 8 Overall weight distributions for product characteristics.

	MD (%)	TCh (%)	PD (%)	L (%)	Weight (%)
WOL	4	4	3	5	16
TC	8	6	4	5	23
DC	5	4	3	2	14
NOP	3	1	2	2	8
RFR	7	3	6	5	21
LOI	6	2	3	3	14
Total	33	20	21	22
Consistency ratio at individual attribute level
Consistency ratio (%)	1.64	6.32	1.45	1.70
Consistency ratio at attribute level
Consistency ratio (%)	3.77

Table 9 shows the comparison of weight distribution between the AHP–CBR and the T-CBR.

Table 9 A comparison of weight distribution between AHP–CBR and T-CBR.

	WOL (%)	TC (%)	DC (%)	NOP (%)	RFR (%)	LOI (%)
AHP–CBR	16	23	14	8	21	14
T-CBR	20	20	10	30	5	15
Differences	4	3	4	22	16	1

Table 10 shows the comparison of EOL selection strategy between the AHP–CBR and the T-CBR.

Table 10 Comparison of EOL selection strategy between AHP–CBR and established methods.

	Similarity values		EOL selection strategy
New case	AHP–CBR	T-CBR	AHP–CBR	T-CBR	ELDA	Industry practice
Computer	0.951	0.920	3	3	3	3
Telephone	0.981	0.981	3	3	3	3
Fax	0.991	0.989	3	3	3	2
TV	0.880	0.879	3	3	3	4
Audio system	0.951	0.932	5	4	4	5
Note: 1: reuse, 2: repair, 3: remanufacturing, 4: recycling with disassembly, 5: recycling without disassembly, 6: disposal.

After EOL path at the product level was determined, the EOL path at the part and component level was evaluated. The EOL path of the retrieved case that has the highest similarity value is adopted as an EOL path for the new case. In this case, the telephone handset was adopted as an example. We used the data from Johnson (2002) to access the information required to evaluate EOL options at the part level. This information includes the weight and material of a part, time to disassemble, clean, sort, refurbish, test and assemble the part of a product (see Table 7).

Table 11 shows the EOL values and the comparison of the recommended EOL path between the cost model of AHP–CBR method and the conventional method. The agreement between these two methods validates the cost model of the AHP–CBR method.

Table 11 Comparison of EOL options at part level between AHP–CBR method and traditional method (Johnson 2002).

				EOL values			EOL options
Part#	Part name	Mass (kg/unit part)	Material	V Rem	V Rec	V Lf	AHP–CBR	Traditional method
A	Handset	0.1734	Mixed	0.760	0.000	− 0.033	Remanufacture	Disassembly for remanufacture
A1	Upper handset	0.0741	ABS	1.139	0.065	− 0.028	Remanufacture	Remanufacture
A2	Lower handset	0.0697	ABS	1.109	0.061	− 0.027	Remanufacture	Remanufacture
A3	Handset circuit board	0.0157	PB	0.054	0.047	− 0.013	Remanufacture	Remanufacture
A4	Connector	0.0087	Steel	− 0.134	0.010	− 0.011	Recycle	Demanufacture via recycle
A5	Receiver	0.0052	Mixed	0.152	0.000	− 0.005	Remanufacture	Remanufacture
A6	Microphone assembly	0.0009	Mixed	− 0.220	0.000	− 0.010	Landfill	Demanufacture via landfill
B	Telephone cord #1	0.0367	Mixed	2.886	0.086	− 0.113	Remanufacture	Demanufacture via recycle
C	Telephone cord #2	0.0234	Mixed	0.481	0.055	− 0.110	Remanufacture	Demanufacture via recycle
D	Telephone base	0.8189	Mixed	28.141	0.000	− 0.156	Remanufacture	Disassembly for remanufacture
D2	Upper plastic casing	0.2134	ABS	4.090	0.187	− 0.052	Remanufacture	Remanufacture
D3	Speaker foam	0.0045	PUR	0.064	0.004	− 0.010	Remanufacture	Remanufacture
D4	4 screws	0.0025	Steel	− 0.225	0.003	− 0.009	Recycle	Demanufacture via recycle
D5	Speaker	0.0756	Mixed	2.376	0.000	− 0.023	Remanufacture	Remanufacture
D6	Plastic plate	0.015	Pp	0.097	0.010	− 0.012	Remanufacture	Demanufacture via recycle
D7	PolyFlex circuit board	0.0075	PVC	3.172	0.005	− 0.010	Remanufacture	Remanufacture
D8	LCD panel#1 – subassembly	0.0171	Mixed	4.704	0.000	− 0.009	Remanufacture	Remanufacture
D9	Main circuit board	0.0442	PB	5.003	0.039	− 0.010	Remanufacture	Remanufacture
D10	Elastometer key pad #1	0.0136	Rubber	0.362	0.001	− 0.010	Remanufacture	Remanufacture
D11	Elastometer key pad #2	0.0076	Rubber	0.387	0.001	− 0.010	Remanufacture	Remanufacture
D12	Elastometer key pad #3	0.0068	Rubber	0.425	0.001	− 0.009	Remanufacture	Remanufacture
D13	Shift button	0.0005	ABS	− 0.191	0.000	− 0.018	Landfill	Demanufacture via recycle
D14	12 memory keys	0.011	PP	0.047	0.007	− 0.012	Remanufacture	Demanufacture via recycle
D15	3 display buttons	0.0027	PP	− 0.176	0.002	− 0.010	Recycle	Demanufacture via recycle
D16	Feature button	0.0015	ABS	− 0.184	0.001	− 0.010	Recycle	Demanufacture via recycle
D17	Release button	0.001	ABS	− 0.187	0.001	− 0.009	Recycle	Demanufacture via recycle
D18	Hold button	0.001	ABS	0.033	0.001	− 0.011	Remanufacture	Remanufacture
D19	12 dial keys	0.0065	ABS	0.214	0.006	− 0.009	Remanufacture	Remanufacture
D20	12 line keys	0.0089	ABS	0.032	0.008	− 0.009	Remanufacture	Demanufacture via recycle
D21	Volume control	0.001	ABS	− 0.171	0.001	− 0.009	Recycle	Demanufacture via recycle
D22	Display cover – LCD#1	0.0063	PVC	0.125	0.005	− 0.005	Remanufacture	Remanufacture
D23	Display cover – LCD#2	0.0047	PVC	0.326	0.003	− 0.010	Remanufacture	Remanufacture
D24	LCD panel # 2	0.0043	Glass	− 0.311	0.000	− 0.011	Landfill	Demanufacture via landfill
D25	Hang up lever	0.0041	ABS	0.494	0.004	− 0.009	Remanufacture	Remanufacture
D26	Metal clip	0.0014	Steel	0.674	0.000	− 0.010	Remanufacture	Remanufacture
D27	Lower plastic casing	0.3042	ABS	2.051	0.267	− 0.010	Remanufacture	Remanufacture
D28	2 rubber pads	0.0023	Rubber	0.357	0.000	− 0.010	Remanufacture	Remanufacture
D29	Plastic stand	0.1369	ABS	1.750	0.120	− 0.010	Remanufacture	Remanufacture
D29A	2 rubber pads	0.0023	Rubber	0.357	0.000	− 0.009	Remanufacture	Remanufacture

As some of the parameters were beyond our control, the following assumptions were taken from the previous literatures:

1. The environmental cost of a remanufactured part was assumed to be 1% of the cost of a remanufactured part (Kaebernick et al. 2002).

2. Since the miscellaneous cost was beyond our control, it was omitted from the calculation (Lee et al. 2001).

3. The environmental cost for C _Envtrans was assumed to be $0.146 per ton kilometre (Anityasari 2008).

4. The labour cost was assumed to be $20 per hour (Johnson 2002).

5. The landfill cost was assumed to be $0.044 per kilogram (Johnson 2002).

6. The recycling rate of steel was $0.90 per kilogram (Anityasari 2008), of lead was $0.211 per kilogram, of rubber was $0.01 per kilogram, of ABS, PVC and PP was $0.11 per kilogram each (Johnson 2002).

The remanufacturing value, recycling value and landfill value were calculated using Equations (13-24). The negative numbers in the remanufacturing value column means that the treatment for remanufacturing processes were more expensive than the cost of new parts. The environment cost driver for extracting material of a new part, recycling, landfill and transportation was obtained from Environmental Priority Strategies Version 2000 (EPS2000d; Anityasari 2008).

4. Discussion

Under the weight distribution of product characteristics, the consistency ratios of the attributes and the individual attributes are within the acceptable values (Table 8). The result indicates that these weights are consistent for use in CBR.

As shown in Table 8, the attribute of MD is the most important; then followed by the attributes of legislation (L), PD and TCh in this order. The individual attributes of TC and RFR are very important in selecting the EOL strategy of a product; then the individual attributes of WOL, DC, LOI and NOP follow. The attribute of MD is very important in the individual attributes of TC, DC, RFR and LOI. The attributes of TCh and L are important in the individual attribute of TC. On the other hand, the attributes of PD and L are important in the individual attribute of RFR. All attributes are not important in the individual attributes of NOP. This result suggests that the designer should focus on how to design a product rather than aiming at reducing the NOP and components of a product when it comes to the selection of EOL strategy.

Table 9 demonstrates that the weights in individual attributes of NOP and RFR in AHP–CBR are in contradiction to those of T-CBR as obtained in Shih et al. (2006). In T-CBR, the individual attribute of NOP is very important; then followed by the individual attributes of WOL, TC, LOI, DC and RFR in this order. However, in AHP–CBR, the individual attribute of TC is very important; then the individual attributes of RFR, WOL, DC, LOI and NOP follow. There are no significant differences of the percentage weight of the individual attributes between AHP–CBR and T-CBR, except NOP and RFR. One possible explanation is that the attributes of MD and PD are very important in RFR regarding EOL selection strategy during the design of a product (Table 8).

One of the objectives in this study was to examine the effectiveness of AHP in providing NN in CBR with the weights for identifying the EOL of products. As stated earlier, setting the weight in CBR is troublesome and requires a structural approach. Table 10 verifies the effectiveness of this approach. Our result demonstrates a good correspondence with that of previous study (Shih et al. 2006). The computer, fax and audio system give higher similarity values than the T-CBR. Additionally, the telephone and TV show no significant difference of similarity value between our approach and T-CBR (Table 10).

In the selection of EOL strategy, the result given by AHP–CBR corresponds to that given by T-CBR and ELDA for the computer, telephone and fax (Table 10). Our finding, with regard to the selection of EOL strategy of the audio system, is in contradiction to ELDA and T-CBR. On the other hand, fax and TV corresponds with industry practice. One possible explanation is that the industrial practice is based on Rose's (2000) survey on the most environmentally friendly strategy adopted by some industries. However, it is not recognised as a standard practice at that time (Rose 2000, Shih et al. 2006). Our approach adopts the strategy provided by Rose (2000) in order to retrieve a case for selecting the EOL strategy. Additionally, Shih et al. (2006) adopted the T-CBR approach for selecting the EOL strategy. Our approach and T-CBR are different in that the AHP approach provides a structured and systematic way to set the weight to CBR. As a result, our approach is very close to ELDA and T-CBR.

Table 10 demonstrates that the computer and audio system show the same values but, the EOL selection strategy is different between them. A possible explanation is that there is a difference in the EOL selection strategy of a similar retrieval case of these two products. For example, in a case of the computer, the similarity value is 0.951 and the EOL selection strategy of the retrieved case is remanufacturing. On the other hand, in a case of the audio system, even though the similarity value is 0.951, the EOL selection strategy of the retrieved case is recycling without disassembly. Hence, it shows the difference in the EOL selection strategy for the computer and audio system even though the similarity value is the same.

Another objective in this study is to find the most cost-efficient EOL option at the parts and components level. From the overall viewpoint, the EOL value of remanufacturing (V _Rem) outperformed other EOL values (Table 11). This result indicates that it is more cost effective to remanufacture most of these parts than to recycle or landfill from the economical and environmental aspects. The EOL value of landfill (V _Lf) is always in a deficit. As for the connector, screws, memory keys, display buttons, feature button and release button, the EOL value of recycling (V _Rec) is better than that of remanufacturing (V _Rem). In other words, it is more cost effective to recycle these six parts than to remanufacture or landfill them. As for microphone assembly, shift button and LCD panel #2, neither remanufacturing nor recycling these parts seems to be profitable. Hence, landfill is the best option.

Our cost model demonstrates a good correspondence with the previous study (Johnson 2002; Table 11). Our finding, however, shows that the EOL options of parts such as the telephone cords #1 and #2, plastic plate, shift button, memory keys and line keys are in contradiction to the conventional method. It has been recommended that the shift button should be dumped. This is because neither remanufacturing nor recycling this part seems to be profitable. As for the parts such as telephone cords #1 and #2, plastic plate, memory keys and line keys, our approach shows better EOL option than traditional methods. This must be due to that remanufacturing these parts seems to be more profitable than recycling or dumping them. Additionally, the costs to purchase new parts are more expensive than remanufacturing these parts from the economical and environmental aspect.

The results of selecting the EOL options in our cost model are very close to a traditional method (Johnson 2002; Table 11). A possible explanation is that our cost model adopts a strategy similar to that provided by Johnson (2002) for selecting the EOL option. Johnson (2002) evaluates the EOL options from an economical aspect. On the other hand, our cost model considers economical and environmental aspects to evaluate the EOL options. As pointed out by Lee et al. (2001) and Kaebernick et al. (2002), the integration of the economical and environmental costs seems to play an important role in determining the EOL options of parts and components of a product.

5. Conclusions

We proposed an integrated AHP–CBR-based method to determine the remanufacturing strategy at the product level and conducted an evaluation of EOL values to find the best EOL option for the parts and components of the product. The developed method was compared with previous studies which showed a good correspondence with the past literatures. However, some additional issues should be addressed in future work.

The characteristic of the survey is based on individual participants of a scatter population. It is possible that the survey should be extended to groups of engineers and executives in the manufacturing industry. Moreover, the environmental cost to remanufacture parts and components of a product is based on an assumption of previous studies. In future work, the rate of energy consumed to remanufacture parts and components of a product will be obtained from the industry.

As the EOL option at the part and component level ranges from remanufacturing to landfill, the remanufacturability has to be considered. This can be done by assigning the remanufacturing index to each part of a product. The future research will integrate the proposed AHP–CBR method with the design for remanufacturing tool so that the remanufacturing index at the product level can be measured.

Acknowledgements

The authors are indebted to the anonymous referees for valuable suggestions in improving the quality of earlier versions of the manuscript.