2.1 SSCM and SRM

SSCM attempts to enhance economic, environmental and social capability from the industry and value chain perspective, thereby effectively improving sustainability [10]. Stakeholders, government agencies and regulatory agencies are gradually becoming aware of the profound impact that the sustainability has on the environment and society [2]. Companies are also beginning to realize that supply chain sustainability will affect their image and reputation [5]. Therefore, the development of SSCM has become a demand of both individuals and society. Through a literature review, we found that SSCM is often associated with green supply chain management (GSCM). It is worth noting that SSCM and GSCM seem to be similar concepts, but their scope is different. GSCM mainly starts from the production process and carries out controls on pollution in each subsequent process, usually including procurement, involvement, production, packaging, image, etc. [12]. SSCM not only promotes the sustainability of a company’s production practices but also discusses its effects on production process with the three comprehensive aspects [13]. Therefore, we believe that SSCM research needs to be based on the TBL and that research needs to be raised from the corporate level to the social level.

Companies are increasingly dependent on suppliers, and the importance of supplier relationship management (SRM) has been highlighted due to globalization [14]. SRM is an important concept in SSCM. Like SSCM, SRM is also an academic term that means to manage the relationship between the participants in the supply chain to help participants jointly perform a series of operations, such as planning, operation, and decision-making, to improve performance and sustainability. SSS/OA is a critical research problem in SRM. However, many studies on SRM mainly focus on the relationship between SRM and corporate performance [1, 14, 15], the level of implementation of organizational supply chain management [16], the impact on carbon emissions, and the relationship with business operations [17]. TBL can help purchasers / managers evaluate suppliers across the economic, environmental and social perspectives., which may bring continuous improvement to the enterprise. So we attempt to construct a MCGDM-FMOO research framework to solve the SSS/OA problem based on the TBL theory.

2.2 Supplier selection criteria

To a large extent, the construction of the criteria system affects the validity and reliability of decision-making results, while incomplete criteria will have an adverse impact on DMs’ ability to accurately assess supplier sustainability [8]. Dickson and Gary [18] constructed the first relatively complete supplier selection criteria system containing 23 items and argued that cost, quality and delivery time belong to the critical considerations for evaluating suppliers. Weber [19] analyzed the most recent 74 papers and found conflicts among the selection criteria, concluding that supplier selection falls under the research area of multi criteria decision making. Chen [20] ranked the importance of 23 criteria based on previous studies and argued that quality, delivery and historical performance belong to the fundamental criteria. Therefore, based on this early research, delivery time, price and quality are considered to be the most indispensable factors in supplier selection process. Rashidi, Noorizadeh, Kannan and Cullinane [8] also proposed that quality, transportation, and price are considered the most critical criteria in the assessment of sustainability by performing a quantitative and qualitative analysis of related literature in the past 30 years.

However, many scholars have ignored the above traditional criteria in their research on SSS [21]. Govindan, Mina, Esmaeili and Gholami-Zanjani [22] emphasized quality and timely delivery in the supplier ranking but ignored cost. In some studies, the important role of supplier delivery capabilities has not been explored [23]. Not only economic performance needs to be considered, but environmental and social factors must not be ignored. Environmental management systems, recycling, pollution control, eco-design and energy consumption are the most critical and common criteria among the 45 commonly used environmental factors identified by Rashidi, Noorizadeh, Kannan and Cullinane [8]. Linton, Klassen and Jayaraman [24] linked the economy and environment together and pointed out that SSCM must be based on product whole life cycle, so total cost becomes a necessary consideration.

The negative impact of resources used and the pollutants generated also need to be considered in SSCM. Therefore, energy consumption and pollution control (especially carbon emissions) are evaluation factors that cannot be ignored when evaluating suppliers’ sustainability. Carbon emissions during transportation are another critical and easily overlooked factor [25]. In terms of society, the sustainability of suppliers is mainly reflected in three aspects: the influence of enterprise operations on employees, the influence on social organizations, and the influence on social members [8]. However, according to a literature review on SSS [26], many current studies have not fully considered and investigated social factors.

2.3 Approaches in the SSS/OA

Solving SSS/OA problems usually includes four methods [27]: MCGDM methods, mathematical programming(MP), artificial intelligence(AI) and hybrid methods. Among them, hybrid methods support DMs by using a combination of two or more methods of the same type [28, 29]. It also allows decision makers to integrate different types of methods to make up for the shortcomings of one approach, thereby improving the accuracy of decision-making [5, 6, 30–32].

Several representative studies on decision-making and order allocation problems are showed in Table 1. It can be seen that although more scholars began to focus on the tools of AI and MP, MCGDM method is still the mainstream and important technique for studying SSS. Accordingly, improving and combining the sub methods to make up for the defects is an important research trend at present.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. A summary of representative studies on supplier selection and order allocation.

https://doi.org/10.1371/journal.pone.0271194.t001

MCGDM method integrates qualitative research methods and quantitative research methods, resulting in the balance between DMs’ subjective opinions and objective data. Therefore, it is widely used in multi industry for supplier selection and evaluation.

Through the research framework formed over the years, MCGDM is mainly discussed in three steps. Firstly, DMs select the appropriate language set to evaluate the criteria by scoring them based on the importance, so as to make the reasonable differences in weight distribution. DMs usually assess criteria with crisp number in the early stages of research. With the development of fuzzy set theory, experts can choose a language set more suitable for the scene to evaluation, which is an effective tool to solve the uncertainty and hesitation of DMs in this process. Secondly, another critical aspect in MCGDM is to extend language set into to weighting model. This procedure needs to take the DMs’ subjective preferences and the objective characteristics of data into account. Therefore, weighting model are usually divided into subjective and objective technique, which are also suitable for expert weighting. Finally, the aggregation operators are utilized for scoring and ranking each alternative. In general, MCGDM is the hybrid approach involving multiple management departments, then weighting DMs and criteria to gain a set of data for alternative evaluation.

MCGDM contains many subdivision methods, such as AHP, BWM, TOPSIS, DEA, TODIM, VIKOR etc. each single method has inevitable defects and irreplaceable advantages. Exploring the combination and improvement of different methods can realize the mutual complementarity, so as to gain more credible and reasonable criteria weighting and supplier ranking results. As an essential and concerned research field, MCGDM mainly focuses on processing of language set, weighting distribution and aggregation operator.TODIM, as a MCGDM method that can consider the psychological factors of DMs to avoid loss, is more in line with the application in real scenes. Qin, Liu and Pedrycz [35] expanded the TODIM method to deal with green supplier evaluation under IT2FS and performed sensitivity analysis on the results. Celik, Yucesan, & Gul [44] constructed the hybrid IT2FS-BWM-TODIM model to handle uncertainty in the decision-making of GSS, which is useful for textile industry stakeholders. Gomes, Machado, Santos and Caldeira [45] applied the original and extended TODIM approaches to supplier selection in the steel industry and proved that the two methods produced the same experimental results under certain circumstances. TODIM was extended to the unbalanced HFLTS language environment to solve the sorting problem of telecom service providers [46]. Feng and Gong [47] proposed a two-stage decision-making model that integrates the LEWM and MOP: the LEWM was used to ranking green supplier in the automobile manufacturing industry and to conduct order allocation from three aspects: total cost, carbon emission and purchase value. Dos Santos, Godoy and Campos [48] collected information on environmental standards proposed by 32 experts and applied the comprehensive MCGDM Entropy-TOPSIS-F approach to the furniture SSM. These methods chose green suppliers with the best environmental performance. Ecer and Pamucar [13] considered the important roles of the economy, society and environment in the sustainable supply chain and determined the weights of these three types of criteria by using F-BWM.

Through the integrated fuzzy CoCoSo method, electrical appliance manufacturers were ranked according to their pros and cons. From the perspective of the green innovation capabilities of SMEs, a framework for large organizations to select suppliers was constructed from three stages of criteria: construction, criteria weighting (BWM) and supplier ranking (fuzzy TOPSIS). Finally, correlation sensitivity analysis was obtained to prove the validity of the decision model [49].

By reviewing the literature, we found that the BWM, entropy and TODIM methods, three commonly employed MCGDM tools, have been widely used in the real world. However, the integrated BWM-Entropy model is used for weighting, and the decision model involving TODIM in the ranking process has not been developed. Therefore, we propose a hybrid MCGDM model to deal with the SSS process.

2.4 Research gaps and highlights

This literature shows that considerable studies have been performed on green supplier evaluation [50]; the works on SSS have demonstrated a rising trend of popularity in recent years, but the overall progress has been relatively small. In addition, in SSS-related research, social and environmental factors are often not given the same attention as economic performance. Meanwhile, some developing countries do not pay enough attention to SSCM due to lacking the formulation of relevant laws and regulations. Moreover, choosing a more sustainable supplier often means increased costs for companies.

Some single MCGDM approaches, such as AHP, BWM, Entropy, TODIM, TOPSIS are most commonly used to solve SSS problems. These methods need to be combined with other tools to solve problems more effectively [51]. However, many studies are based on a single approach under a fuzzy environment [2]. Furthermore, the prerequisites of bounded rationality for DMs were ignored in many MCGDM tools, while TODIM is an effective tool to take those factors into account.

Therefore, we construct the SSS/OA criteria based on the TBL and integrate the PL-BWM-Entropy-TODIM framework to promote SSS. Subsequently, an FMOO model for order allocation that can handle uncertain parameters is proposed. Then, two methods, AUGMECON and LP-metrics, are effective tools to gain a richer Pareto solution set. At last, we use the TOPSIS method to select the most appropriate Pareto solution.

Table 2 shows a comparison between this study and previous studies displaying the problems solved, the fuzzy environment, the sustainable criteria, and the MCGDM method used.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Comparison of studies sustainable criteria and approaches.

https://doi.org/10.1371/journal.pone.0271194.t002

3.1 Fuzzy logic

3.1.2 Probabilistic linguistic term set (PLTS)

Definition 4 [11]. Let is a linguistic term set. F^(l)pr^(l) is the LT F^(l) with probability pr^(l), #F(pr) represents quantity of LTs in F(pr). denotes a PLTS.

Definition 5 [11]. Let a PLTS with , sub^(l) denotes the subscript of LT F^(l). If are sorted through the values of sub^(l)pr^(l) in descending order, we call F(pr) an ordered PLTS.

Definition 6 [11]. Let a PLTS with . represents the standardized PLTS, where for all .

Definition 7 [11]. Let and be any two PLTSs. If #F₁(pr₁)>#F₂(pr₂), #F₁(pr₁)−#F₂(pr₂) LTs are added to #F₂(pr₂) until the numbers of LTs in F₁(pr₁) and F₂(pr₂) are equal.

Definition 8 [11]. Let a PLTS , and sub^(l) represents the subscript of LT F^(l). The score function E(F(pr)) of F(pr) is given by (3) where .

For any two PLTSs F₁(pr₁) and F₂(pr₂) :

1. If E(F₁(pr₁))>E(F₂(pr₂)), then F₁(pr₁) is superior to F₂(pr₂), denoted by F₁(pr₁)≻F₂(pr₂).
2. If E(F₁(pr₁))<E(F₂(pr₂)), then F₂(pr₂) is superior to F₁(pr₁), denoted by F₁(pr₁)≺F₂(pr₂).
3. If E(F₁(pr₁)) = E(F₂(pr₂)), then F₁(pr₁) is indifferent to F₂(pr₂), denoted by F₁(pr₁)~F₂(pr₂).

Definition 9 [11]. Let a PLTS . The score function E(F(pr)) of F(pr) is given by with . The deviation degree of F(pr) is given by: (4)

For any two PLTSs F₁(pr₁) and F₂(pr₂) with E(F₁(pr₁)) = E(F₂(pr₂)):

• , then F₂(pr₂) is superior to F₁(pr₁), expressed by F₁(pr₁)≺F₂(pr₂).
• , then F₁(pr₁) is superior to F₂(pr₂), expressed by F₁(pr₁)≻F₂(pr₂).
• , then F₁(pr₁) is equal to F₂(pr₂), expressed by F₁(pr₁)~F₂(pr₂).

Definition 10 [61]: Let and be any two normalized PLTSs. According to definitions 5, 6 and 7, the normalized PLTSs and are obtained. The distance between them is as follows: (5)

In which , , and T represents the quantity of LTs in S.

3.2 Weighting criteria: PL-BWM-Entropy

The MCGDM framework includes criteria weighting and suppliers ranking which is shown in Fig 2. In this framework, assigning the criteria weights is an important process, which is divided into subjective and objective weighting. The BWM, which simplifies the complex calculation process in terms of reference comparisons, is a subjective method. The entropy weight method is a typical objective approach in which the information entropy of each index is calculated according to a decision matrix.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The conceptual framework of the MCGDM approach.

https://doi.org/10.1371/journal.pone.0271194.g002

The value of information entropy is inversely proportional to the final weight. The criteria weights related to comprehensive methods are more effective and stable.

3.2.1 The PL-BWM model

The BWM approach was presented by Rezaei, which includes selection, reference comparisons and computation steps. The calculation process of the PL-BWM approach are given by [62]:

Step 1: C_j = (C₁,C₂,⋯,C_n) which denotes the set of criteria is selected by the DM, and the best criterion C_B and worst criterion C_W are also marked to represent the importance.

Step 2: The DM provides the preference of C_B over C_j and C_j over C_W, which are expressed by PLTSs: (6) (7)

Step 3: The expected values of F^Bj(pr) and F^jW(pr) are calculated as follows: (8) (9) where and . In addition, f denotes a linguistic scale function, and the calculation results of f are between 0 and 1.

(10)

Step 4: Next, A_B shows the comparison between the C_B and other criteria, while A_W shows the comparison between other criteria and C_W using PLTSs: (11) (12)

Step 5: Finally, the subjective criteria weight is calculated under the following linear programming model: (13)

The final subjective criteria weight vector is determined.

3.2.2 The PL-Entropy model

The entropy method is a dominant weight determination method based on the evaluation matrix given by the DMs [63]. We assume that all alternatives and criteria are A_i and C_j. Under the environment of PLTSs, the specific steps of calculating the criteria weight by the entropy approach are given by [64]:

Step 1: After normalizing the decision matrix S = (s_ij)_m×n by definitions 5, 6 and 7, the information entropy of the j th criterion is given by: (14) in which m represents the total number of alternatives.

Step 2: The criterion objective weight can be calculated by the obtained H_j. The formula is as follows: (15)

Step 3: The criteria final in the basis of subjective and objective comprehensive weighting method are as below: (16) in which a+b =1.

3.3 Ranking suppliers: PL-TODIM model

The TODIM model proposed by Ma, Fan and Huang [63] on the basis of prospect theory. The alternatives assessment is established by capturing the bounded rationality of the human. This ranking result reflects the psychological characteristics of the DMs in avoiding risks. The calculation process is given by [64]:

Step 1. We need to convert all criteria (including benefit and cost types) into the same type by using Eq (17). Then, we normalize the decision matrix S = (s_ij)_m×n into R = (r_ij)_m×n by using definitions 5-7.

(17)

Step 2. The comprehensive weight of criteria w_j is obtained in Eq (16), and we can compute the relative weight w_jp according to: (18) in which w_j represents the weight of criterion C_j, .

Step 3. We compute the overall dominance degree φ of A_i over A_o corresponding to each criterion C_j as follows: (19) where (20) where i,o = 1,2,⋯,m. θ denotes the attenuation factor of the losses provided by DMs and θ>0. When , ϕ_j(A_i,A_o) denotes a gain. If , ϕ_j(A_i,A_o) represents a loss.

Step 4. According to the following form, we can calculate overall prospect value of each alternative A_i. (21) where higher values of δ_i indicate that alternative A_i is better.

3.4 Formulating the mathematical model

The proposed MOO model considers multiple suppliers and products comprehensively and aims to assign the final order quantity to different sustainable suppliers with minimize costs and CO₂ emissions, maximize the purchase value.

The relevant indices, parameters, variables and mathematical models are as follows.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

https://doi.org/10.1371/journal.pone.0271194.t003

model (22) (23) (24) subject to (25) (26) (27) (28)

The objective functions in (22), (23), and (24) represent the maximization of the total purchase value, the minimization of the total cost, and the minimization of carbon dioxide emissions during transportation, respectively. (24) effectively reflects the sustainability of the supplier in the order allocation process. Constraint (25) ensures that the order quantity from supplier i should be within its supply capacity. In constraint (26), the order quantity is required to be greater than the minimum order quantity specified by supplier i. Constraint (27) limits nonnegative ordered quantity. Constraint (28) denotes sign constraint.

3.4.1 Constructing the FMOO model

Fluctuations in the supply chain will cause uncertainty in various costs. To deal with the dynamic nature of purchasing costs, transportation costs, management costs, and carbon dioxide emissions in the proposed model, we introduce fuzzy logic into the trapezoidal MOO model. The specific FMOO model is calculated as below [30, 31]: (29) (30) (31) subject to (32) (33) (34) (35)

The confidence value α(0≤α≤1) provided by the DMs is represented in the above model. The superscripts 1-4 of the fuzzy parameters indicate the most pessimistic, the most likely and the most optimistic values [5]. For example, and represent the lowest and highest possible costs, respectively, for purchasing t products from supplier i. and together represent the two most likely purchasing costs for t from supplier i.

The linear membership function of (29) is given by: (36)

The linear membership function of (30) and (31) are given by: (37)

3.4.2 The augmented ε-constraint (AUGMECON) method

The traditional ε-constraint cannot guarantee the efficiency of the solution and requires a large computational load and long computation time when solving multiple objective functions. Mavrotas [65] proposed the augmented ε-constraint (AUGMECON) which produces a weak Pareto optimal solution and speeds up the calculation process, to overcome the disadvantages of the traditional ε-constraint.

When using AUGMECON, FMOO is converted into (38)–(41). Where (38) is the only objective function. (38) subject to (39) (40) (41) subject to Eqs (32)–(35)

where S represents the surplus variable and η is within the range of [10^-3, 10^-6] commonly.

3.4.3 LP-metrics method

The LP-metrics method converts multiple objective functions in mathematical programming into a single objective function through the following formula [32]: (42) subject to Eqs (32)–(35) where denote the ideal values of the objective functions and are the weights assigned to the objective functions by the DMs to find more Pareto solution sets of FMOO through different weight combinations.

3.5 Selecting the final solution by TOPSIS

As a classic MCGDM method, TOPSIS selects the final solution by measuring the distance between the alternative and the optimal/worst alternatives. The detailed process is as below [66]:

Step 1. Normalize objective function value by following: (43)

Step 2: The objective function values with the integrated weights are given by: (44) where represents weight of the jth objective function.

Step 3: The separation measures for each alternative are obtained by: (45) (46) in which and represent the PIS and NIS for each function, respectively.

Step 4: The closeness coefficient is given by: (47) where a higher CC_i value corresponds to a better alternative i.

4.1 Stage 1: Calculation of the criteria weights and supplier ranks by MCGDM approach

4.1.1 Weighting the criteria.

After constructing the criteria system, DMs use Eqs (6)–(16) in Section 3.2 to comprehensively weight the criteria. In PL-BWM, DMs assigned to "pollution control" (C₂₂) was the most critical criterion and "staff development" (C₃₂) was the least critical criterion based on negotiation of decision group. The evaluation process is shown in the following two matrices. the three DMs jointly executed the reference comparisons with PLTS after consultation, Tables 6 and 7 show the fuzzy preferences of Best-to-Others and Others-to-Worst based on Table 8.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. The linguistic label for fuzzy preferences of the best criterion over all criteria.

https://doi.org/10.1371/journal.pone.0271194.t007

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. The linguistic label for fuzzy preferences of all criteria over the worst criterion.

https://doi.org/10.1371/journal.pone.0271194.t008

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 8. Linguistic terms and corresponding degree in PLTS.

https://doi.org/10.1371/journal.pone.0271194.t009

Then, we can transfer the fuzzy preferences showed in Tables 6 and 7 into matrices as follows:

Through Eqs (8)–(10), the following specific linear programming model are constructed based on Eq (13). The final subjective weight are obtained by solving the linear programming in Lingo 11.0. Finally, the are listed in Table 10.

PL-Entropy is used for objective weighting. First, we integrate the evaluation matrix according to the DM’s opinions based on Table 8 and normalize the matrix as shown in Table 9. Then, the objective weights are obtained by Eqs (14)–(15). Integrating and into the comprehensive weight w_j is the final step of weighting by Eq (16). Table 10 present subjective, objective and comprehensive criteria weights. In the weighting process, it is found that “technological capability” (0.05) and “energy consumption” (0.126) are regarded as the least important and most important practices in SSS.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 9. Normalized group decision matrix in PLTS.

https://doi.org/10.1371/journal.pone.0271194.t010

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 10. Subjective weight, objective weight, comprehensive weight and relative weight of sub-criteria c_j.

https://doi.org/10.1371/journal.pone.0271194.t011

4.1.2 Ranking the suppliers

After getting the criteria weights, DMs rank the suppliers by Eqs (17)–(21). C₂₃ is the most critical criterion (w₂₃ = 0.126). Hence, the weights w_23p of all criteria relative to C₂₃ are presented below.

Then, calculate the dominance of each supplier A_i over each supplier A_o with respect to criteria C_j. For example, in order to calculate ϕ₁₁(A₁,A₂), j = 11, i = 1, o = 2. Firstly, we can calculate the score function E(F(pr)) according to Eq (3) to get the comparison results F₁₁₁(pr₁₁₁)≻F₁₁₁(pr₂₁₁). Then, the distance d(F₁₁(pr₁₁)−F₁₁(pr₂₁)) = 0.913 between F₁₁(pr₁₁) and F₁₁(pr₂₁) are compute by Eq (5). w_11p = 0.83, . Finally, ϕ₁₁(A₁,A₂) = 0.31 are obtained by Eq (20).

Thirdly, the overall dominance degrees φ are obtained and shown in the following matrix by Eq (21). Then, calculate the overall prospect value of each alternatives A_i(i = 1,2,3) (suppose θ = 1), δ(A₁) = 1, δ(A₂) = 0.46, δ(A₃) = 0. For example, , where .

Finally, rank the three suppliers in according with the value of δ(A_i)(i = 1,2,3). The larger the value of δ(A_i), the better the scheme A_i, and get A₁≻A₂≻A₃.

4.1.3 Analyzing the effect of the parameter θ

In the PL-TODIM method, the influence of loss will increase when 0<θ<1. While θ>1 indicates that the influence of loss will decrease. Therefore, we believe that θ will affect the overall dominance degree [35]. We observe its influence on the ranking by changing the value of θ (0.25-100). Table 11 shows the results.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 11. Ranking results for different θ.

https://doi.org/10.1371/journal.pone.0271194.t012

As can been seen in Table 11, different values of θ between 0.25 and 100 will result in a different overall dominance degree. However, the supplier ranking always remains A₁≻A₂≻A₃. Consequently, the results are consistent with the changed θ. Since the result caused by the change of θ is more sensitive in the range of 1-10, Fig 3 shows that the different risk factors lead to the change of three alternatives.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Alternatives’ prospect value of different θ-value for illustrative example.

https://doi.org/10.1371/journal.pone.0271194.g003

4.2 Stage 2: FMOO model

The FMOO is solved via Matlab2020a with the optimization toolbox running on a personal laptop Intel(R) Core (TM)i5-8250U CPU at 2.5 GHz with 8 GB of RAM and the Windows 10 operating system.

4.3 Stage 3: Using AUGMECON and LP-Metrics to solve the FMOO model

AUGMECON was originally used to solve the FMOO model. We convert Z1 to the objective function through Eq (38), and Z2 and Z3 are converted into constraints by Eqs (39)–(41). Here, η = 0.0001, and ε is assigned by DMs. LP-Metrics requires that the three objective functions be converted into a single objective function by Eq (42). The weights of different functions are considered in the mono-objective model, so the DMs assign the 20 weights of the original objective functions as shown in Table 12. This will allow us to obtain more Pareto solutions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 12. Assigned weights of objective function in LP-metrics method.

https://doi.org/10.1371/journal.pone.0271194.t013

Table 13 presents the objective values and order quantity by AUGMECON after ten α iterations, where ε₂ = 184710 and ε₃ = 299020. Similarly, Table 14 includes objective values and order quantities determined by LP-Metrics with ten α levels for , , and . For example, the quantity of b products that should be purchased from the first supplier was 27,165 when α = 0.1. In addition, Fig 4 presents the values of the objective functions after LP-Metrics runs 20 weight combinations. Therefore, we believe that as the weight of Z1 gradually increases, the company’s purchase value will not increase significantly, but the value of the total cost function and CO² emission function will increase significantly. Fig 5 shows that when α = 0.5, the objective function values with 5 α levels are calculated by the above two approaches. Since the solutions obtained by the above two approaches are different, we take the results of both methods into account and provide more solutions for DMs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. reto front with twenty weights for α = 0.1 of LP-Metrics method.

https://doi.org/10.1371/journal.pone.0271194.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Pareto front for α = 0.5 of AUGMECON method.

https://doi.org/10.1371/journal.pone.0271194.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 13. Results for AUGMECON approach.

https://doi.org/10.1371/journal.pone.0271194.t014

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 14. Results for LP-Metrics approach.

https://doi.org/10.1371/journal.pone.0271194.t015

4.4 Stage 4: Selecting the solution by TOPSIS

Twenty-four effective Pareto solutions can be computed through AUGMECON and LP-Metrics to solve the FMOO model when α = 0.5 (Table 15). The DMs select the optimal solution by Eqs (43)–(47). Hence, we believe that the 21st solution is the optimal solution, and the corresponding order allocation is shown as Table 16:

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 15. TOPSIS calculation result for α = 0.5.

https://doi.org/10.1371/journal.pone.0271194.t016

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 16. The optimal order allocation from each supplier.

https://doi.org/10.1371/journal.pone.0271194.t017

5.1 Discussion

SSS/OA is a critical decision-making problem for contemporary enterprises. The proposed MCGDM-FMOO model takes into account comprehensive weights, criteria construction, ranking process, and multi-objective optimization. In the illustrated example, the PL-BWM-Entropy technique found "energy consumption" (0.126), "pollution control" (0.119), "recycling" (0.11), "product quality" (0.107), and "environmental management systems" (0.107) to be important practices in SSS, while "technological capability" (0.05) and "staff development" (0.057) are considered the least important practices. PL-TODIM considers psychological factors of decision process by parameter θ. However, we found that changes in the range of 0.25-100 did not affect the supplier ranking (A₁≻A₂≻A₃), so we believe that PL-TODIM has strong robustness and consistency. Finally, we discussed the results of OA in three scenarios: (1) AUGMECON operation results with ten α levels (0.1-1) are shown in Table 13, (2) LP-Metrics operation results with ten α levels (0.1-1) for , , and are shown in Table 14, and (3) LP-Metrics operation results with twenty objective weight combinations for α = 0.1 are shown in Fig 4. The results of (1) and (2) show that the Pareto solutions provided by the AUGMECON and LP-Metrics methods are different, but not significantly so. Therefore, it is believed that both results are reasonable, and they enrich the Pareto solution of the FMOO model. The results of (3) denote that the attitude of the DMs regarding Z1 will not significantly affect the purchasing value, and the trends of the three objective function values are not greatly affected by the weight change. In general, considering sustainability in supplier selection will lead to increased costs.

The proposed approach contributes the following advantages: (1) Within the MCGDM model, PL-BWM-Entropy combines linguistic operators and two weighting methods to minimize the loss of linguistic information. Moreover, the comprehensive weighting model can better integrate the evaluation opinions of DMs and objective evaluation information. So, it is more likely to provide accurate and comprehensive information than single method. (2) Under various situations of MCGDM problems, such as engineering, economy, management and military, it is necessary to rely on industry experts to make decisions. The PL-TODIM considers the loss aversion behavior of DMs, hence retains decision-making information of all DMs as much as possible under the premise of bounded rationality. (3) The proposed FMOO model can effectively depict the uncertainty of parameters by trapezoidal fuzzy number in four dimensions (pessimistic, optimistic and most likely values). As a popular method, AUGMECON denotes the augmented ε-constraint method, which reduces the computation load and computing time by improving the algorithm when solving FMOO. Then using AUGMECON and LP metrics provides a richer set of Pareto solutions of FMOO.

5.2 Qualitative comparison and analysis

The existing MCGDM methods can be roughly divided into three categories on the basis of their functions: 1. Weighting methods, such as AHP, ANP, BWM and Entropy method which are mainly used to weight criteria or DMs. 2. Ranking methods are utilized to sort the advantages and disadvantages of alternatives, including TOPSIS, VIKOR, TODIM, ELECTRE and PROMETHEE. Some weighting methods are also applicable to ranking, such as AHP. 3. Hybrid methods. Due to the complexity, a single method can’t solve the research problem. Therefore, the combined methods are proposed to deal with the defects of the single methods.

Many hybrid MCGDM methods for supplier selection have been developed. We selected some representative studies, from whether the OA problem was solved, whether the weight was calculated from the subjective and objective dimensions, whether the uncertainty in the decision-making process was handled, whether the psychological factors of DMs to avoid loss were considered, and the aggregation methods used in the literature were qualitatively compared with the proposed methods.

By qualitatively comparing the research scope of the literature in Table 17, the biggest highlight of this paper is that we consider the psychological factors of DMs to avoid losses, not just making decisions based on expected utility, and measure the criteria weight from both subjective and objective aspects. In addition, our model can express the degree of hesitation to avoid information loss as much as possible, which is not available in most other methods. More reasonable order allocation quantity can be obtained based on the results of SSS.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 17. Qualitative comparison of MCGDM techniques.

https://doi.org/10.1371/journal.pone.0271194.t018

We selected some studies related to supplier selection and order allocation for comparison, including whether economic, environmental and social sustainability is involved in OA process, whether cost minimization, procurement value maximization and carbon dioxide emission minimization are included in the objective function, and whether the model can deal with multi product order allocation. Through the summary of order allocation in Table 18, compared with other literature, this paper fully considers the application of sustainability in the OA process, and effectively reflects environmental sustainability into the carbon dioxide emission function in the FMOO.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 18. Qualitative comparison of MCGDM techniques.

https://doi.org/10.1371/journal.pone.0271194.t019

5.3 Quantitative comparison and analysis

To demonstrate the effectiveness and novelty of the model more clearly, the proposed ranking model is quantitatively compared with other methods, and prove the effectiveness of this method by analyzing the similarities and differences of supplier ranking.

In order to control variables and keep consistent with the situation of MCDM problem, we firstly deal with the language rating in the two articles. The language level of PLTS is divided into seven levels in Tong et al [42]. In this paper, we divide linguistic variables into eight levels. By comparing the corresponding degree of the language sets, we believe that there is little difference in accuracy between the two articles. Therefore, when the PL-TODIM method in this paper is applied to the cases in [42], we believe that ignoring the l₀ (none) language set will not affect the final ranking. Finally, the corresponding relationship between the linguistic terms and corresponding degree in PLTS of the two articles is shown in the Table 19.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 19. Linguistic terms and corresponding degree in PLTS in two papers.

https://doi.org/10.1371/journal.pone.0271194.t020

Table C. 3 in reference [42] provides the comprehensive evaluation matrix expressed in PLTS. The evaluation of supplier A1 by the decision-making group under criterion C1 is . We convert this evaluation into for convenience of calculation based on the Table 19.

Comparing different methods in the PLTS environment eliminates the error caused by language set. We compare PL-TODIM with PL-TOPSIS and PL-PROMETHEE II methods respectively, and the final ranking of five alternatives are presented in Table 20. In the final results, great changes have taken place in the ranking between PROMETHEE II method and the proposed method. We think the main reason for the large difference in ranking is that PROMETHEE method does not give expression to the psychological change process of DMs in the decision-making process although it fully considers the deviation of each supplier in the evaluation criteria. The ranking of PL-TODIM and PL-TOPSIS are similar. The best and worst alternatives are A₃ and A₂. The opposite ranking of A₁ and A₄ may be due to the fact that TOPSIS is ranked according to the distance between best/worst and each alternative. The bounded rational behavior related to the individuals are also not considered. Instead, TODIM method reflects the psychological behavior of DMs, especially the degree to which different DMs avoid losses, which makes the decision-making better reflect the way of human thinking in reality. We believe that difference of ranking results caused by different principles of methods and the degree of considering the psychological factors of DMs. Therefore, we still believe that the proposed method in this paper is feasible and innovative.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 20. The results based on the comparing methods.

https://doi.org/10.1371/journal.pone.0271194.t021

6.1 Research summary

Reasonable SSS/OA processes are an effective means to improve supply chain performance and are gradually becoming increasingly valued. In this study, we construct a TBL-based SSS criteria system and propose the hybrid MCGDM-FMOO approach to deal with the SSS/OA problem. Criteria are weighted by PL-BWM-Entropy from both a subjective and an objective perspective, and then suppliers are ranked considering the psychology of the DMs using the PL-TODIM method in the developed MCGDM framework. The goal of the OA problem is to reasonably allocate multiple products to multiple suppliers under the premise of maximizing purchase value, minimizing costs and minimizing CO₂ emissions, as shown in the FMOO model. Here, AUGMECON and LP-Metrics provide more Pareto solutions as tools to solve FMOO. Then, TOPSIS can choose the solution that are closest to PIS and farthest from NIS, which we call the ideal solution. Finally, validity and practicability of solving the SSS/OA problems is proved by applying the MCGDM-FMOO model to an illustrative example. The method proposed can also be used to solve SSS/OA problems in other industries.

6.2 Managerial implications

MCGDM method has been widely used as the decision-making framework of intelligent system and industry decision-making [73]. This study has the following Managerial implications for decision-makers in practice: (1) Make full use of the experience and knowledge of DM group. The proposed method can better translate the uncertainty and hesitation of DMs into visual data in the PLTS environment. In addition, the evaluation information of all DMs can be retained without considering the group size. (2) Make the decision result more reasonable. In the decision-making system, the advantages and disadvantages of alternatives are obtained by evaluate multiple criteria. So, weighting criteria and alternatives ranking are two critical processes. The combined method of subjective and objective can better eliminate the impact of unreasonable criteria weight on the final decision. In the process of ranking, the bounded rationality of DMs is taken as the premise. As a result, these two aspects make the decision-making process more reasonable and credible. (3) Provide support for enterprises to improve and reconfigure the SCM process. The FMOO model established in this paper considers the total cost and environmental factors, it makes the order distribution in line with the current concept of sustainability on the premise of maximizing the profits. It is worth noting that multiple conflicting objectives cannot be achieved at the same time in the order allocation model, but the computing methods proposed in this paper can calculate the results of different weight allocation among multiple objectives, so as to provide more feasible solutions for the actual situation. In addition, the objective function of maximum purchasing value is established based on the SSS results, which effectively connects the SSS and OA problems, and makes the two independent functions in supply chain management become a coherent series of decision-making activities. The coherent activities reduce the probability that the organization will lose profits due to lack of coordination. (4) Provides universality and scalability. The criteria system based on economy, environment and society factors in SSS, and the order allocation model with maximum purchasing value, minimum total cost and carbon dioxide emissions can be implemented in other practical applications. Therefore, the SSS-OA framework (Fig 1) is suitable for different types of manufacturing enterprises committed to sustainable development, such as household appliances, furniture production, electronic equipment, automobile industry, aviation industry, etc.

6.3 Limitations and future work

However, there are some limitations in this paper. Although we integrated the subjective method and objective method in the criteria weighting stage, the degree of differentiation in the subjective and objective dimensions is poor. In order to make up for this deficiency, we will strive to propose a comprehensive weight method that can fully show the discrimination degree between subjective and objective weights. Secondly, TODIM method fully considers the psychological factors of DMs to avoid risks, but the degree of risk avoidance of each individual is different in actual decision. We will consider applying the neural network method to the simulation of DMs’ psychological behavior. Artificial Intelligence is one of the main methods to solve decision-making problems. In the future, the integration of Artificial Intelligence technique and MCGDM method is a novel and perspective research direction.