1 Introduction

In the general business process, the product design stage determines more than 70% of the product manufacturing cost, and this stage has a high degree of uncertainty. However, for new product development today, a more systematic approach is needed to ensure that the design output meets the needs of customers, production and assembly considerations, product life cycle expectations, and many other concerns [1, 2]. In the past, Boothroyd and Dewhurst [3] studied the assembly constraints, costs, and other factors that must be considered in product design and proposed the concept of Design for Assembly (DFA). This idea opened up development in this domain and has been recognized by follow-up researchers. Therefore, many studies have also explored related “Design for X” methods, such as Design for Disassembly (DFD) [4], Design for Mass Customization [5], Design for Modularity (DFM) [6, 7], and Design for Environment (DFE) [8]. Discussions in this area can be found in the related literature [9]. Among them, Design for Disassembly (DFD) is also a member of the DFX family. Usually, DFD will be combined with the topic of green design, which is why DFD research is attracting increased attention. This study focuses on the issue of disassembly in the area of DFD.

The purpose of green design is to ensure that a product complies with the requirements of environmental protection so as to reduce the impact on the environment. In addition, environmental protection, cost, quality and other factors must be balanced. An important basic condition of green design is easy disassembly of the product. Moreover, the disassembly sequence planning (DSP) method plays an important role in the framework of DFD. DSP refers to the process of disassembling the product, particularly the sequential order of the disassembly of components. In terms of DSP classification, disassembly can be divided into total disassembly and selective disassembly [10]. Total disassembly mainly occurs when the target parts are clearly defined. Therefore, the focus of total disassembly is to find the optimal order (such as the one with the least cost or maximum profit) under various constraints. In contrast, selective disassembly focuses on the search for parts or modules that are worthy of disassembly. At the end of a product’s life, the recycling manufacturer can execute the relevant recycling processes according to a predetermined recycling plan. This study focuses on total disassembly.

In the past, some researchers have developed qualitative or quantitative methods to evaluate DFD design. Kroll et al. [11] have proposed an evaluation method for disassembly design; Desai and Mital [12] have further proposed evaluation methods to check whether the disassembly plan is destructive disassembly, total disassembly or selective disassembly; Cappelli et al. [13] discuss the generation of the optimal disassembly sequence to meet the minimum disassembly time and cost; Favi et al. [14] attempted to establish life cycle assessment indicators to further echo the design of the life cycle considerations. Most of these past researchers focused on constructing DFD assessment methods and indicators, and usually designers need to calculate these evaluation indicators manually. However, the practical application of these indicators is a time-consuming task [15]. If an ideal method can be developed in the future, it will enable DFD to further develop in the direction of design automation.

This study attempts to develop interactive genetic algorithms based on the idea of genetic algorithms (GAs) in DFD. In traditional GAs, the encoding of the chromosomes is performed before the processing of the algorithm, after which the chromosomes are decoded and the fit value of each chromosome is calculated, and the mechanism is selected according to the value of the fitness function. A chromosome with a higher fitness value has a greater chance of being selected for the new offspring. After the selection, Crossover and Mutation are performed to generate the next generation of chromosomes. Such process is repeated until the end of the evolution. Because GAs can easily adjust the design according to problems and generate ideas, they are favored by many researchers. In the past, Kongar and Gupta [16] developed the so-called Priority Preservative Crossover (PPX) mechanism, and Giudice and Fargione [17] considered disassembly time, life cycle cost, and environmental impact for the fitness function. Hui et al. [18] studied the mechanism of the Disassembly Feasibility Information Graph (DFIG) to explore GAs; Go et al. [19] took mechanisms similar to those of Kongar and Gupta (2006) to develop genetic algorithms for DSP because GAs are fast and flexible, learning easily. Kheder et al. [20] also adopted the crossover mechanism of PPX and added the idea of maintenance in the exploration of the goal formula. Tseng et al. [21] proposed novel block-based genetic algorithms to escape the regional solution; in that approach, the searching process is based on the goal formula and the score matrix. During a search, the direction is set toward the formula with better scores, which will greatly reduce the search range and make it easier to find the optimal solution, enhancing the efficiency of the entire algorithm and making GAs applicable to complex cases. It has been found that Block-GAs can effectively promote the efficiency of GAs. Therefore, Block-GAs are used in this study.

To make GAs applicable to the field of design, the developed system must have an interactive mechanism. Interactive Genetic Algorithms (IGAs) comprise one of the branches that later derive a number of genetic algorithms. The main difference between IGAs and traditional GAs lies in the fact that IGAs integrate human thinking. For example, an IGA will add user preferences to the design of fitness functions. In the past, Takagi [22] systematically proposed related methods and the theory of IGAs. Yan et al. [23] replaced the user’s scoring method with an evaluation method of Boolean operation. Babbar-Sebens and Minsker [24] proposed the so-called IGMII IGAs, which can improve the evolutionary strategy from expert opinions. Another direction of research is how to make users understand and evaluate the genes in IGAs more accurately [25]. But the application of IGAs in the study of DSP is rarely seen.

One reason for this could be that it is very difficult to quantify the plans or attributes of product design. Therefore, the key to whether the design can initially produce a better concept solution for the next stage is the method of quantifying the product characteristics as much as possible when the product has not yet been clearly formed. To take into consideration the high degree of uncertainty and ambiguity encountered in the initial solution, this study adopts the viewpoint of the Fuzzy set as the basis for fit function design. Based on the above discussion, the study combines fuzzy semantics with Interactive Genetic Algorithms (IGAs) to build the DFD environment.

In this paper, the framework of interactive design is described in Sect. 2, and fuzzification penalty design is discussed in Sect. 3. The operation of Block-GAs is discussed in Sect. 4. Section 5 illustrates examples of design, and finally, Sect. 6 presents our conclusions.

2 Construction of Interactive GAs for Disassembly Sequence Planning

In DSP planning, the constraints of the priority order should be met. The engineering properties of the parts also need to be predicted. If the properties of the parts change during disassembly, such changes will cause waste and influence the evaluation of the disassembly sequence, representing the waste of cost by disassembly. Interactive design focuses on the adaptability of the fitness function. The designer can achieve the desired disassembly result through the calculation results by adjusting the fuzzy scores and priority relation information. The following are the basic ideas underlying this study:

  1. 1.

    Fuzzy scores By adjusting fuzzy semantic definitions, the user’s ideas and expected results can be brought into the algorithm in a more realistic manner, one in which the user’s ideas are added to the calculation of the fitness value.

  2. 2.

    Adjustment of the priority relations When the initial priority relation is incorrect, new restrictions need to be added, special requirements need to be defined, or a priority order of special parts may be deleted, then specific adjustments can be made. Computation will be manipulated to generate solutions that match the designer’s idea and produce a correct result.

Based on the above discussion, the framework of disassembly sequence planning in product design proposed in this study is shown in Fig. 1. To meet the needs of the designers, it is necessary to satisfy the considerations of interactive and fuzzy semantics. The major steps are explained as follows:

  • Step 1 The user enters the part attributes and the priority relations between parts. According to the user-defined part attributes, establish a penalty value matrix and a relation matrix. The user can also freely set the penalty value matrix to make adjustments. The fuzzy scoring procedure is explained in Sect. 3.

  • Step 2 Carry out Block_GAs (Sect. 4).

  • Step 3 The user makes decisions based on relevant data such as disassembly time, and the output calculation results and analysis are given to the user for reference. These analyses include the penalty values among parts, the proposed improved parts sequence and part attributes.

  • Step 4 Add the designer’s idea: The designer judges whether it meets the design requirements. If it meets the requirements, proceed to Step 5. Otherwise, the user can redesign or adjust the priority of parts, the penalty value matrix, control parameters, etc., and repeat Step 4.

  • Step 5 Adjust the disassembly sequence order. This is explained in Sect. 5.

  • Step 6 Stop the evaluation and generate satisfactory solutions.

Fig. 1
figure 1

Interactive GAs for DSP

Full size image

3 Building a Penalty Matrix by fuzzy Scoring Procedure

In this study, the stapler example in Fig. 2 is used to illustrate how to build a penalty matrix with the fuzzy scoring procedure. The stapler is made up of 18 parts. The priority of the stapler parts is shown in Fig. 3. The labels in the figure are the disassembly priorities, and the circular nodes are the parts. The direction of disassembly is set to be the six directions in the table. This is based on the professional knowledge of the designer. The disassembly information is divided into part directions, tools, and liaisons. There are 6 different disassembly directions: + X, − X, + Y, −Y, + Z, and − Z. The information of each part is shown in Table 1. In terms of disassembly tools, there are four levels of tools, T1, T2, T3, and T4, according to the degree of difficulty. The details of the parts information are shown in Table 1. The fuzzy items in this study include attribute weights, direction, tools, and liaisons.

Fig. 2
figure 2

The parts of a stapler

Full size image
Fig. 3
figure 3

The sequential priority of the stapler parts

Full size image
Table 1 The directions and tool information of the stapler parts
Full size table

3.1 The Weighted Semantic Variables

In this study, subjective weights are used, from which users can subjectively give fuzzy semantics directly. Instead of the numerical values of the traditional scores, the weights of the fuzzy parts attributes are adopted here. Generally speaking, the proportions of the distribution weights add up to 1. This study uses fuzzy weights, with the evaluation criteria of fairly important, important, very important, and extremely important. The semantic variables are shown in Fig. 4. According to the user’s judgment of the importance of the attributes, the semantic variables are shown in Table 2.

Fig. 4
figure 4

The weight of semantics for importance

Full size image
Table 2 The semantic variables
Full size table

3.2 Parts Directions

The direction attribute of a part is expressed in terms of changing the angle, which is mainly divided into changes of less than 45°, changes of about 90°, and changes of about 180°. Three kinds of semantic variables of 0 to 1 are used as the fuzzy numbers. Take the direction of the stapler as an example. Assume P3 → P1 (→: Part P3 takes priority over Part P1); P3’s direction is − Y, and P1’s direction is + Y. The direction changes from − Y to + Y, and the change angle is about 180°. Accordingly, the fuzzy numbers of (0.5, 0.75, 0.75) are assigned. The directions of the semantic variables are shown in Fig. 5 and Table 3.

Fig. 5
figure 5

The semantics for part direction in the stapler case

Full size image
Table 3 The semantic variables for the direction in the stapler case
Full size table

3.3 The Disassembly Tool

The disassembly tool attribute is based on the change of tools, with semantics of T1–T4 representing “No change”, “Small change”, “Big change” and “Very big change”. The semantic variables use 0–1 for the fuzzy numbers, as shown in the semantic variables map in Fig. 6. The semantic variables are shown in Table 4. For the direction attribute of the stapler, suppose P2 → P1; P2’s tool is T1 and P1’s tool is T3. The tool changes from T1 to T3, the change is big, and the fuzzy numbers (0.4, 0.6, 0.8) are given for the direction attribute of the stapler’s case.

Fig. 6
figure 6

The semantics for part tool in the stapler case

Full size image
Table 4 The semantic variables for the tool in the stapler case
Full size table

3.4 Liaison-Type Semantic Variables

The attribute of liaisons is mainly used to describe the degree of relationship between parts. The semantic variables include “With close liaison”, “With moderate liaison” and “No liaison”. The fuzzy numbers are set between 0 and 1. The semantic variables are shown in Fig. 7. For the stapler case, the semantic variables and their liaisons are shown in Table 5. The liaison graph of the stapler parts is shown in Fig. 8, where N indicates that there is no liaison between two parts, C indicates a liaison between two parts, and I indicates a close liaison between two parts. Designers can examine the relationships between parts from the liaison graph.

Fig. 7
figure 7

The semantic variables for liaisons in the stapler case

Full size image
Table 5 Definitions of liaisons in stapler case
Full size table
Fig. 8
figure 8

The liaison graph in the stapler case

Full size image

3.5 Calculate the Penalty Value

At this stage, integral operations are performed based on the fuzzy numbers and fuzzy weights of the fuzzy semantics determined. Assume that the overall order is P3 → P14 → P1 → P2 → P4 → P18 → P16 → P10 → P11 → P17 → P15 → P13 → P12 → P5 → P8 → P6 → P7 → P9. Take the part relation P1 → P2 for example. The change in direction is judged as about 180°, with the corresponding fuzzy numbers (0.5, 0.5, 0.75); the change of tool is a big change, with the fuzzy numbers (0.4, 0.6, 0.8); and the liaison belongs to only some liaison, with fuzzy numbers (0.25, 0.5, 0.75). From the judgment of relative importance, the direction attribute is considered extremely important; the tool attribute, very important; the liaison type, very important. From the user’s definition of importance, the corresponding semantic fuzzy numbers are (0.6, 0.8, 1) for extremely important and (0.4, 0.6, 0.8) for very important. The weights are then integrated with the engineering attributes as Formula (1). According to Formula (1), the fuzzy numbers of the P1 → P2 relation can be calculated as P1,3 = (0.61, 1.36, 2.14).

$$Fuzzy P_{j,seq} = \left( {\left( {W_{Dl} *D_{j,l} + W_{Tl} *T_{j,l} + W_{Ll} *L_{j,l} } \right),\left( {W_{Dm} *D_{j,m} + W_{Tm} *T_{j,m} + W_{Lm} *L_{j,m} } \right),\left( {W_{Dr} *D_{j,r} + W_{Tr} *T_{j,r} + W_{Lr} *L_{j,r} } \right)} \right)$$
(1)

where Fuzzy Pj,seq, the integrated fuzzy number for design attributes; W, the weights; D, the direction attribute; T, the tool attribute; L, the liaison attribute; (l, m, r), corresponding fuzzy numbers; j, number of current part; seq, number in current sequential order.

The main purpose of the defuzzification procedure is to convert the fuzzy numbers into explicit values after the fuzzy operation. In this study, these converted explicit values serve as the penalty values added when the parts are disassembled. As proposed by Bevillacqua and Petroni [26], the center-of-gravity method is used for the defuzzification procedure. The advantage is that a distinct value can be obtained without complicated operations. As shown in Formula (2), the P1 → P2 relation is calculated as P1,3 = 0.61 + 1.36 * 2 + 2.14)/4 = 1.3675).

$$P_{j,seq} = \frac{{l_{j,seq} + 2m_{j,seq} + r_{j,seq} }}{4}$$
(2)

where seq, number in the current sequential order; Pj,seq, the explicit value between the current part and the next part.

Through the standardization process, the distinct values from defuzzification are transformed into a value of 0–1. The standardization can be processed from Formulas (3) and (4): Min(P) = 0.665, Max(P) = 0.96. According to the output of Formula (2), the relation P1 → P3 can be calculated as P1,3 = (1.3675–0.665)/0.96 = 0.7318. Table 6 lists the total penalty values between each two parts.

$$P_{j,seq} = P_{j,seq} - Min(P)$$
(3)
$$P_{j,seq} = \frac{{P_{j,seq} }}{\hbox{max} (P)}$$
(4)

where P, the penalty value between each two parts; Max(P), the maximum value in the penalty matrix; Min(P), the minimum value in the penalty matrix.

Table 6 The penalty value matrix
Full size table

4 Block-Based Genetic Algorithms

In this study, the Penalty Function is used as the design of the fitness function. The less the attribute changes, the lower the penalty value. The value of the fitness function is used to judge the pros and cons of the disassembly sequence. Considering the interactive characteristic, the formula needs to make the calculation of the fitness value feature the advantages of editing, adjustment and expansion. Its calculation method is shown in Formula (5). This study aims to reduce the number of disassembly directions and tool changes in the disassembly activities, since unnecessary directions and tool changes would cause an inefficient disassembly process. This view is similar to the research of Li et al. [27]. In this study, the direction and tool change are integrated in the objective formula so as to find a disassembly sequence with lower numbers of changes in direction and tools on the same product. Consequently, a disassembly sequence with fewer changes is relatively more efficient than one with more changes in the disassembly procedure. According to studies related to Methods of Time Measurement (MTM), this goal can reduce the disassembly time [11, 28]. From the search for a disassembly sequence with fewer changes, a more efficient disassembly sequence can be obtained, as shown in Eq. (5).

$$MS = \mathop \sum \limits_{seq = 0}^{n - 2} P_{j,seq} , \quad \forall j,j = 0, \ldots , n - 1$$
(5)

where MS, the fitness value; j, number of the current part; seq, number in the current sequential order; Pj,seq, the explicit value between the current part and the next part; n, total number of parts.

Figure 9 shows the procedure of Block-based genetic algorithms, which is characterized by introducing the penalty value matrix in the crossover method (Step 4) and the mutation method (Step 5). This offers the direction for searching, thereby improving the quality of the algorithm’s solution.

Fig. 9
figure 9

The block-based GAs

Full size image

The steps for the Block-based genetic algorithms are described below:

  • Step 1 Process non-binary chromosome encoding: generate the feasible initial parent chromosomes with numbers of parameters according to the user-defined part attributes and the part priority matrix.

  • Step 2 Calculate the fitness value of each chromosome as a basis for the chromosome evaluation.

  • Step 3 According to the fitness value, duplicate the chromosomes equal to the parent number by the roulette method for the next generation.

  • Step 4 Perform the crossover method based on the blocks. The details will be described later.

  • Step 5 Perform the block-based mutation method.

  • Step 6 Determine whether the evolution should be terminated. In this study, the maximum generation number is set as the termination condition. If the termination condition is satisfied, proceed to Step 7; otherwise, proceed to Step 2.

  • Step 7 Stop the algorithm and generate the optimal or close-to-optimal solution to the disassembly sequence order.

Figure 10 illustrates the procedure for the Block-based crossover mechanism, the steps of which are listed below:

  • Step 4.1 Select two chromosomes as Parent1: P3 → P14 → P18 → P16 → P17 → P2 → P1 → P4 → P10 → P11 → P13 → P15 → P5 → P8 → P7 → P6 → P9 → P12 and Parent2: P17 → P18 → P16 → P14 → P3 → P10 → P11 → P15 → P13 → P12 → P2 → P1 → P4 → P5 → P8 → P7 → P6 → P9 for crossover, as shown in Fig. 11a.

    Fig. 10
    figure 10

    The procedure for the crossover mechanism

    Full size image

  • Step 4.2 Randomly generate a block whose size is 0–100% of the total block. The total block size is 18, and assume that the size of the block generated is 88.88%; then the block size is 18*88.88% = 16.

  • Step 4.3 Divide the block size of 16 into 3 blocks, as shown in Fig. 11b.

  • Step 4.4 Calculate the fitness values for all blocks. This is shown in Table 7.

    Fig. 11
    figure 11

    The conceptual diagram of the crossover mechanism

    Full size image
    Table 7 Fitness values for blocks in Parent1
    Full size table

  • Step 4.5 Choose the best block. In this case, the lowest fitness value is 7.8464, indicating that the block P3 → P14 → P18 → P16 → P17 → P2 → P1 → P4 → P10 → P11 → P13 → P15 → P5 → P8 → P7 → P6 should be reserved.

  • Step 4.6 The chromosome Parent2 is P17 → P18 → P16 → P14 → P3 → P10 → P11 → P15 → P13 → P12 → P2 → P1 → P4 → P5 → P8 → P7 → P6 → P9, and then duplicate Parent2 for Offspring1. The chromosome Parent1 is P3 → P14 → P18 → P16 → P17 → P2 → P1 → P4 → P10 → P11 → P13 → P15 → P5 → P8 → P7 → P6 → P9 → P12, and duplicate Parent1 for Offspring2.

  • Step 4.7 Insert the reserved block in Parent1 to the corresponding positions of Offspring1; Offspring1 is P3 → P14 → P18 → P16 → P17 → P2 → P1 → P4 → P10 → P11 → P13 → P15 → P5 → P8 → P7 → P6 → P6 → P9, as shown in Fig. 11c.

  • Step 4.8 Search for the repeated genetic codes in Offspring1, and find that P6 is repeated and P12 is missing. Remove the repeated P6; Offspring1 is then P3 → P14 → P18 → P16 → P17 → P2 → P1 → P4 → P10 → P11 → P13 → P15 → P5 → P8 → P7 → P6 → P9, as shown in Fig. 11d.

  • Step 4.9 Search the range for inserting P12 into Offspring1, and find the ranges from P15 to P9 and from P13 to P18. Calculate the penalty values for feasible insertion positions, as shown in Table 8. The lowest value is found in P14, and the output Offspring1 is P3 → P14 → P18 → P16 → P17 → P2 → P1 → P4 → P10 → P11 → P13 → P15 → P5 → P12 → P8 → P7 → P6 → P9, as shown in Fig. 11e.

    Table 8 The penalty values for insertion positions in Offspring1
    Full size table

For the block-based mutation mechanism, the detailed procedure is illustrated in Fig. 12 and described below:

Fig. 12
figure 12

The procedure for block-based mutation

Full size image

  • Step 5.1 Process the mutation mechanism in the selected chromosomes.

  • Step 5.2 Randomly generate the mutation amount (1 to number of total parts); the amount of mutations equals the number of mutated genes. It is assumed to be 10 in this case.

  • Step 5.3 Randomly select a gene for the mutation point and remove it from the chromosome, as shown in Fig. 13a.

    Fig. 13
    figure 13

    The conceptual diagram of the mutation mechanism

    Full size image

  • Step 5.4 Insert the mutation point to the most suitable position in the gene (Step 4.8 to Step 4.9), as shown in Fig. 13b.

  • Step 5.5 Repeat Step 5.2 to Step 5.4 ten times (the amount of mutations is 10) and complete the mutation mechanism.

To verify the performance of the block-based genetic algorithms, this study compared it with the GA of Kongar and Gupta [16] and the Ant Colony Optimization (ACO) of Dorigo and Gambardella [29]. The solution effect is tested with a stapler example. The results are shown in Table 9. From the convergence graph in Fig. 14, it can be found that the solution quality of the block-based genetic algorithm is better than those of the other two methods. In terms of the execution time, the block-based genetic algorithm is similar to Ant Colony Optimization but better than Kongar and Gupta’s GAs.

Table 9 Test results
Full size table
Fig. 14
figure 14

The convergence graph

Full size image

5 Sample Tests

5.1 Fine-Tuning Design

The fine-tuning design is herein defined as displacement of the part position. After the order is generated, the user can select the part to be adjusted. If the disassembly order is P18 → P4 → P5 → P10 → P3 → P2 → P11 → P1 → P8 → P6 → P7 → P9 → P16 → P17 → P14 → P15 → P13 → P12, the penalty value is 5.9152 from the fuzzy score calculation in this disassembly sequence order (Fig. 15a). If we want to move part P15, then P15 is input as the part to be moved, and all the movable results are confirmed after checking the priority sequential orders (Fig. 3). After calculation of the Block-based GAs, the result is shown in Fig. 15b. It can be found that sequential order No. 1 has the lowest penalty value of 5.7578, so it can be selected. Now P18 → P4 → P5 → P10 → P3 → P2 → P11 → P1 → P8 → P6 → P7 → P9 → P16 → P17 → P15 → P14 → P13 → P12 is the new sequential order for disassembly, as shown in Fig. 15c. The designer can adjust the position of the disassembly sequence of a single part until it meets the requirements. In analyzing the overall sequential orders, it is suggested that different attributes be changed to reduce the penalty value.

Fig. 15
figure 15

Tuning mechanism for the disassembly sequence order

Full size image

The related disassembly information for all possible results of the new design is summarized in Table 10. If the designers focus on P18 → P4 (disassemble P18 before P4), suggestions can be offered for the adjustment of part attributes. According to Table 10, it can be found that some part relations, those where there exist continuous changes of direction and tool in the part relationship, may cause increases in penalty value. Obviously, reducing such changes can reduce the penalty value, and it can be simulated to improve the outcome. The simulated results for these part attributes are shown in Table 11 and Table 12. The designer can evaluate whether the changed attributes are feasible and able to reduce the penalty value by adjusting the original attributes of the parts.

Table 10 Analysis of the results for all disassembly sequence orders
Full size table
Table 11 Suggestions for the part direction in disassembly
Full size table
Table 12 Suggestions for the tools in disassembly
Full size table

5.2 Redesign

According to the results of the research method and the part attribute analysis, the user can adjust the penalty value through the output part information. If the result does not meet the design requirements, the following three-part output information can be observed: (1) P15, P16 and P17 are consistent in direction (− Y) and liaison (close liaison) (see Table 1 and Fig. 7). (2) From the disassembly sequence order, it is found that these three parts, P15, P16, and P17, will be disassembled one after another; in other words, they will be removed in order. (3) According to the stapler diagram of Fig. 3 and the priority sequence diagram of Fig. 2, P15, P16, and P17 are located in the same area. Given these proposals and the check for priority in the disassembly sequence order (Fig. 3), these three parts can be combined with P12, as shown in Fig. 16, to form a one-piece component, as shown in Fig. 15d. Such a design has the following three advantages: (1) it reduces the complexity in the priority sequence diagram; (2) it reduces the number of changes of part attributes and the waste caused by disassembly as well as the penalty value; (3) it reduces the number of parts. The addition of the decrement design concept not only complies with the goal of green design but also mitigates the impact on the environment.

Fig. 16
figure 16

The stapler parts before and after redesign

Full size image

6 Conclusions

In this study, an interactive genetic algorithm is proposed and successfully applied to disassembly planning design. The fuzzy penalty scores are first offered for designers’ reference, and then Block-based GAs are used to solve the disassembly sequence order. Using the quantitative fuzzy penalty scores, designers can evaluate the quality of the design.

This study makes the following contributions: (1) according to their demands, designers can define fuzzy scores and part information and verify the solution by adjusting the disassembly sequence order; (2) the goal of solution quality and an interactive approach are integrated in the algorithm, enhancing the value of the quantitative evaluation; (3) design for disassembly (DFD) will help to transmit environmental awareness back to the source of product design. Based on green design considerations, if a certain part should be removed, all of the disassembly routes should be assessed to achieve the purpose of resource recycling.

In future research, the implementation of green design still needs further efforts. Other quantitative assessment factors, such as the economic valuation of parts or the recycling of different parts (or subassemblies), as in reuse and remanufacturing, can be taken into consideration in the near future.