APPLICATION OF THE ELECTRE TRI METHOD FOR SUPPLIER CLASSIFICATION IN SUPPLY CHAINS

The present research aims to propose a Supplier Selection Model that creates collaborative relationships in Supply Chains, so that suppliers can be previously categorized into the cooperation, the coordination and the collaboration levels. Applied, quantitative, exploratory and descriptive research methods were used. A bibliographical research, a questionnaire, and the quantitative modeling were adopted as methodological procedures. The managers responsible for the Supplier Selection process in the Brazilian Wind Energy companies participated in this research. First, a criteria framework for the Selection of Supplier of goods and services was developed. Second, a Multicriteria Decision Aiding Model was created and validated, enabling the classification of the suppliers according to the relationship levels in a systematic way in terms of their performance from a set of criteria by implementing the ELECTRE TRI method. Firms can use the Model periodically in order to revise the supplier assessment and, if needed, direct them to either an inferior or superior class.


INTRODUCTION
In today's world firms operate in an environment where they need to do business differently if they intend to stay competitive. In the past few decades, they have faced this challenge by implementing Collaborative Supply Chains that allows them to maintain and gain competitive advantages through collaborative efforts with their suppliers and customers (Cao et al. For that reason the Multicriteria Decision Aiding Approach was used in order to carry out this research, because it seeks to systematize and organize the supplier selection by supporting the Managers and reducing the decisions made solely based on experience. Its Methods have been widely implemented in various research fields, mostly for structuring decision-making problems with multiple criteria, or conflicting and poorly structured objectives (   According to Harrison, Hoek & Skipworth (2014), coordination is an essential step to achieve integration in the Supply Chain; whereas collaboration goes beyond that, because it encompasses long-term commitments, sharing of technology and control and planning integrated systems, and interdependence among the firms.
Zacharia, Nix & Lusch (2009) consider cooperation and collaboration to be different approaches to coordination. Cooperation involves communication and teamwork in order to defuse the tension between individual and common goals. On the other hand, collaboration requires an array of skills and a higher level of joint decisions, information sharing, and establishment of joint goals that can improve shared and individual goals.
In terms of cooperation, firms exchange basic information and build long-term relationships with a limited number of critical suppliers and customers. In terms of coordination, there is a continuous flow of critical and essential information through the use of information technology. In terms of collaboration, there is a high level of commitment, trust and information sharing. Relationships that are strategically important and complex to manage should be treated collaboratively  A common ground is found in the literature in which the highest level of integration in Supply Chains is collaboration, which is based on a high level of trust; commitment; joint planning and decision; resource, process, information and risk sharing; establishment of mutual objectives; and teamwork, so that the goals are achieved and great solutions are found (

Supplier Selection
For many years the traditional approach for Supplier Selection was based on price as the only decision criterion. Nonetheless, once firms understood that it is not efficient, they started to adopt criteria that involve both quantitative and qualitative factors that are hard to measure due to their subjectivity (

Ranking
Criterion Evaluation 1 Quality Extreme Importance 2 Delivery Considerable Importance 3 Performance History 4 Warranties policies 5 Production capacity 6 Price 7 Technical capability 8 Financial position 9 Procedural compliance Average Importance 10 Communication system 11 Reputation and position in industry 12 Desire for business 13 Management and organization 14 Operating control 15 Maintenance and repair service 16 Attitude 17 Impression 18 Packaging ability 19 Labor relations record 20 Geographical location 21 Amount of past business 22 Training aids 23 Reciprocal arrangements Slight Importance specific and less commonly used criteria that involve qualitative factors with a significant level of subjectivity that are hard to measure, such as: buyer commitment, co-design, conflict solution; cooperation; achievement of goals; incentives; innovation; prediction of interactive demand; interorganizational communication; JIT capability; joint action; close relationship; supplier initiative and commitment; supplier development programs; and top management support.
The qualitative criteria that are fundamental for Supplier Selection in partnerships, strategic alliances, and cooperative and collaborative relationships are referenced in the literature as soft factors. These criteria had been primarily discussed by Ellram (1990) who categorized them into four groups, as shown in Chart 3. reason special attention has been drawn to the implementation of methods that enable the combination of these criteria for evaluating suppliers. Therefore the Multicriteria Decision Aiding Approach was selected in order to create the proposed Model.

Multicriteria Decision Aiding Approach
The Multicriteria Decision Aiding Approach seeks to assist the solution of problems that demand a complex decision-making process by providing tools to Managers; that is, multicriteria decision-making problems (Almeida, 2011; Gomes & Gomes, 2019).
The multicriteria decision-making problem consists of a situation where there are two alternatives to choose from. Such choice is made by the desire of achieving multiple goals, which are oftentimes conflicting themselves and related to variables that represent them and allow the evaluation of each alternative. Thus, these variables are called criteria (Almeida, 2013).
The type of solution targeted for a particular multicriteria decision-making problem, that is to say, the way the decision maker opts to compare the alternatives is called a problematic (Almeida, 2013), which can be classified as: • Choice problem (P.α): it seeks to select a subset of alternatives; • Classification problem (P.β ): it seeks to assign each alternative to a class; • Ranking problem (P.γ): it seeks to rank the alternatives; • Description problem (P.δ ): it seeks to describe the alternatives and their consequences; • Portfolio problem: it seeks to choose a subset of alternatives that meets the goals according to particular restrictions. On one hand, Chai, Liu & Ngai (2013) consider the Multicriteria Methods methodological frameworks whose goal is to provide the decision makers with recommendations based on a set of alternatives, actions, objects, solutions or candidates, which are evaluated from various viewpoints; that is, criteria, attributes, features or objectives. On the other hand, Almeida (2013) states that the Multicriteria Method consists of a methodological formulation or a theory with a well-defined axiomatic structure, which can be used in order to create a Model that seeks to resolve a specific problem. According to the author, the Multicriteria Method differs from the Multicriteria Model, because it is more general and can be applied to a broader class of decision-making problems.
The Multicriteria Methods can be classified in many different ways. The classifications mentioned by Almeida (2013) are synthesized in Chart 5.
As stated by Gomes & Gomes (2019), the discrete Methods are suitable for resolving problems that have a finite number of alternatives. The continuous Methods, or Multicriteria or Interactive optimization, are suitable for resolving problems with an infinite number of alternatives. According to Almeida (2013), the majority of managerial problems have a discrete set of alternatives.
The Multicriteria Methods are classified as compensatory and noncompensatory due to the compensation that there can or can not be in the decision criteria (Almeida, 2013). In the former, a low performance of an alternative from a given criterion is compensated for by a better performance in another criterion, that is, trade-offs among criteria are considered. In the latter, there are no trade-offs, thus, the alternatives must have a satisfactory performance in most criteria.

11
Chart 5 -Classification of the Multicriteria Methods.

Classification Typologies
In terms of the nature of the set of alternatives

Multiattribute Utility or Single Criterion Synthesis Theory
It is derived from the american current of thought, in which the preferences of the decision maker for a particular alternative, evaluated by means of a set of criteria, are aggregated with a single utility value, which is measured in an additive manner (with trade-offs). A score is generated for each alternative based on the performance according to each criterion, thus, the best evaluated alternatives are the ones that have the best score.
Outranking It is derived from the french current of thought whose main purpose is to create binary relations that represent the preferences of the decision maker based on the available information (without trade-offs). Therefore, through a pairwise comparison the superior alternative in each criterion is verified, creating an outranking relation from the confrontation between two alternatives. Therefore, the alternative that shows superiority in most criteria is best evaluated. This approach is based on the Condorcet voting system.

Interactive Methods
They are developed in the Multi-objective Linear Programming MOLP context. They have computational steps and are interactive, that is, they allow trade-offs. They seek an alternative that is overtly superior according to all objectives, that is, a dominant one. Therefore, they aggregate the preferences of the decision makers and mathematical, interactive and successive calculations. Guarnieri (2015) emphasizes that the experts of the Multicriteria Approach divide the methods into three Approaches, which are presented and described in Chart 6. A fourth Approach is included by Gomes & Gomes (2019), the one of Hybrid Methods, in which concepts of two or more Approaches are used.
The Outranking Approach Methods differ from those of the Single Criterion Synthesis Approach as they allow for a more flexible modeling of the problem, because they do not accept the comparability of all the alternatives and do not impose a hierarchical structure of the criteria (Gomes & Gomes, 2019). However, as the Single Criterion Synthesis Methods achieve the analytical aggregation in order to achieve a score for each alternative, they facilitate the comparison of the alternatives (Almeida, 2013).

Ensslin, Montibeller Neto & Noronha (2001) explain that both the Single Criterion Synthesis
Approach and the Outranking Approach convey the idea of determining a general performance for each one of the alternatives. However, the latter determines the performance through the pairwise comparison of the performances of the alternatives for each criterion. Almeida (2013) concludes that the Outranking Methods show many characteristics that differ them from those of the Single Criterion Synthesis Methods. Nonetheless, the most important characteristic of the former is that it presents noncompensatory evaluations of the alternatives, whereas the latter presents compensatory evaluations.
Guarnieri (2015) states that Approach selection precedes the Method selection, which will greatly depend on the rationality of the decision maker once he shows his preferences (compensatory or noncompensatory). The author lists the main Methods for each one of the Approaches, as shown in Chart 7. There is no common ground in the adoption of the best Method. It is imperative to apply critical thinking, so that one's choice is adequate for the characteristics of the decision-making problem at hand (Gomes & Gomes, 2019;Guarnieri, 2015).
According to Guarnieri (2015), in order to select the Method the following must be considered: • The decision-making situation that encompasses the objectives; • The problematic at hand (choice, ordination or classification); • The types of criteria (quantitative and/or qualitative); • The rationality of the decision maker (compensatory or noncompensatory).

13
Chart 7 -Main Methods of the Multicriteria Decision Aiding Approach.

Approach Method Description
Multiattribute Utility or Single Criterion Synthesis Theory MAUT It is based on the modeling concepts of the traditional preference. It allows trade-offs among criteria and two situations: strict preference and indifference. An aggregate utility function is created, which aggregates the criteria in a single criterion synthesis, showing the preferences of the decision maker. It is an ordination Method. SMART It is a simplified version of MAUT, which analyzes the evaluation of the alternatives, taking into consideration the worst and the best stimulus. It uses strategies of the heroic approach in order to justify the linear approaches of the multidimensional utility functions. The aggregation of the preferences of the decision maker based on the alternatives and the criteria is additive, thus, it considers trade-offs. It is an ordination Method. TOPSIS It evaluates the performance of the alternatives through the similarity to the ideal solutions, in which the best alternative would be the one that is closest to the positive ideal solution and farthest from the negative one. The positive solution maximizes the benefit criteria and minimizes the cost criteria, whereas the negative ideal solution maximizes the cost criteria and minimizes the benefit criteria. It is an ordination Method that considers trade-offs. AHP It divides the problem into a variety of interrelated factors by means of the creation of a hierarchy, which enables the decision maker to define the priorities and judge the preferences according to the alternatives, comparing them in pairs for each criterion through matrices and based on the Saaty's numerical scale. It is an ordination Method that considers trade-offs.
Outranking ELECTRE It is composed of two main procedures: creation of one or many outranking relations, and the exploitation of these relations. The creation of one outranking relation, or more, aims to compare each pair of alternatives. It does not allow trade-offs. PROMETHEE It consists of making a pairwise comparison of the alternatives and building a relation of outranking values, which stands out because it involves concepts and parameters that have a physical or economic interpretation. It does not allow trade-offs.

Interactive Methods
Step Method -STEM For each calculation phase, the solution that minimizes a Chebyshev heavy distance to the ideal solution is sought, which is put to the appreciation of the decision maker in the dialog phase. If all the values of the objective functions are seen as satisfactory, the process comes to an end. Otherwise, the decision maker establishes that the objective functions accept to relax and the value of such a relaxation, so that the objectives that have not yet attained satisfactory values are enhanced. It is a choice and an ordination Method that considers trade-offs. Interval Criterion Weights -ICW The decision maker chooses a solution according to a sample of non-dominated solutions that is presented in each dialog phase. In each calculation phase various weighted sums of the objective functions are optimized through regularly dispersed combination weights in the parametric diagram, which requires that the decision maker explicitly indicates the weights. It is a choice Method that considers trade-offs. Pareto Race It carries out a free directional research on the non-dominated region. The information of the preferences consists of the indication of the objective functions that need improving, which leads to the alteration of the research direction. The solutions are calculated by establishing a direction that provides a variation on the values of the objective functions in line with the preferences of the decision maker, which is subsequently projected onto the non-dominated region. It is a choice and an ordination Method that considers trade-offs. TRIMAP It conducts a free research in terms of a progressive and selective learning of the set of non-dominated solutions, combining the reduction of the admissible region with the reduction of the parametric diagram. In each calculation phase a weighted sum of the objective functions is optimized. The decision maker can specify inferior limitations for objective functions, which are translated into the parametric diagram, and impose restrictions directly on the weights. It is a choice and an ordination Method that considers trade-offs.  Almeida & Costa (2003) emphasize other aspects such as: • The problem that is being analyzed; • The context and the available time for the process; • The available information and the degree of precision; • The structure of preferences of the decision maker in tandem with the required rationality.
According to Almeida (2013), selecting the Method is key to the process of creating the Multicriteria Models. Lima Júnior, Osiro & Carpinetti (2013) explain that the resolution of the Model must obey a set of rules established by the selected Method, and such a choice must be made in the development phase of the Model.
As stated by Almeida (2013), the Multicriteria Model can use any one of the Methods, even if it is not totally adequate for the problem, in order to avoid the one that is more complex and difficult for the Model. Nevertheless, an evaluation based on the confrontation between the simplicity and the precision of the Model is always necessary. For the author, the simplification will lead to errors, however, the Model can still be useful.

ELECTRE TRI Method
In this research the ELECTRE TRI Method was selected to compose the Model that was created and validated, because it proposes to resolve the classification problem (P. The selection of this Method also took into consideration the tendency found in the literature regarding the Supplier Selection subproblem in order to build collaborative relationships, in which the decision makers have shown a noncompensatory rationality; that is, without trade-off among the criteria, which is a characteristic of the Methods that compose the Outranking Approach (Guarnieri, 2012).
The ELECTRE Methods are implemented in two phases: i) creation of the outranking relation, which makes a pairwise comparison of the alternatives; and ii) exploitation of the outranking relation, in which a procedure or an algorithm is applied in order to resolve a problem according to the problematic at hand (Almeida, 2013).
These methods differ in terms of the type of problematic to be resolved, the intra-criteria and the inter-criteria information used, and the amount of outranking relations created and investigated (Gomes & Gomes, 2019 Figure 4 illustrates the profiles according to the criteria, in which b p+1 corresponds to the ideal alternative. The preferences for each criterion are defined according to a pseudo-criteria, whose preference and indifference thresholds, , respectively, constitute the intra-criteria information. Therefore, q j [g(b h )] specifies the largest difference g j (a) -g j (b h ), which preserves the indifference between "a" and b h on criterion g j ; and p j [g(b h )] represents the smallest difference g j (a) -g j (b h ), compatible with a preference of "a" in relation to b h on criterion g j ( • Set of weight coefficients (w 1 , w 2 , ..., w m ), used in the concordance test; , respectively), which represents the degree of credibility of the assertion aSb h (b h Sa, respectively), ∀a ∈ A, ∀b h ∈ B. In order to define the credibility of the index, the partial concordance indices c j (a,b h ), the concordance indices c(a,b h ) and the partial discordance indices d j (a,b h ). Accordingly, the following steps should be followed ( 4. Compute the credibility index σ (a,b h ) of the outranking relation: It is necessary to implement an exploitation phase, since the assignment of the alternatives to the classes does not directly result from the outranking relation S. Thus, the assertion aSb h (b h Sa, respectively) is considered to be valid if σ (a,b h ) ≥ λ (σ (b h ,a) ≥ λ , respectively), in which λ is a cutting level such that λ ∈ [0.5 • σ (a,b h ) ≥ λ e σ (b h ,a) < λ , which results in aSb h and not b h Sa, being a preferred to b h (aSb h ); Finally, two assignment procedures are used; that is, two classification procedures of the alternatives whose role is to analyze how an alternative is compared to the profiles, so that the class to which that alternative should be assigned is defined ( • Pessimistic procedure: it compares "a" successively to b i , for i = p, p-1, ..., 0; being b h the first profile, such that aSb h ,"a" must be assigned to class C h+1 (a C h+1 ); • Optimistic procedure: it compares "a" successively to b i , for i = 1, 2, ..., p+1; being b h the first profile, such that "b h is preferred to a","a" must be assigned to class C h (a C h ).
When the pessimistic procedure is used with λ = 1, an alternative can only be assigned to the class C h if g j (a) is equal to or higher than g j (b h ) for each criterion. On the other hand, when the optimistic procedure is used with λ = 1, an alternative can only be assigned to the class C h when g j (b h ) exceeds In the pessimistic procedure the assignment of alternatives to the classes is achieved in a more conservative way. The alternatives are assigned to an inferior class other than the one determined by the profile; whereas in the optimistic procedure the less conservative, the higher the class that the alternatives are assigned to. On one hand, when there was a convergence between the two procedures, the system created to evaluate and rank the alternatives was capable of comparing them to the profiles. On the other hand, when there was a divergence

METHODOLOGY
Applied, quantitative, exploratory and descriptive research methods were used. Twelve Managers who are responsible for the Supplier Selection in ten Brazilian Wind Energy companies participated in it. This research was divided into three steps, as shown in Figure 5. In step 1, an online questionnaire made available in April and May 2019 on SurveyMonkey® was used. It consisted of 04 questions whose purpose was to identify the currently used criteria for Supplier Selection, the criteria that could be included, currently used criteria from a list of 45 criteria validated by the literature, and the degree of importance given to each criterion in a Likert scale, varying from 1 -unimportant to 5 -very important.
Chart 10 shows the list of 45 criteria in which there are 30 general criteria of the Supplier Selection problem, and 15 specific criteria of the collaborative relationship subproblem, which are also called soft factors. These criteria were extracted from the bibliographical research about Supplier Selection discussed in Section 2.2. The criteria referenced by at least two authors were selected. The behaviors and the characteristics required on each relationship level in the Supply Chain according to Marqui, Moura & Alcântara (2013) were also included.
In order to create a criteria framework, the most recurrent criteria in the open-ended questions (1 and 2), the referenced ones and the best evaluated ones in the close-ended questions (3 and 4) were taken into consideration. Thus, an amount of 15 criteria were selected in order to create the Model. In Step 2 the Multicriteria Decision Aiding Model for Supplier Selection was created. For that purpose the procedure for resolving multicriteria decision-making problems proposed by Almeida (2013) was used, as shown in Figure 6.
In Step 3 the Model was validated, thus, one the Managers that participated in Step 1 was asked to evaluate the alternatives; that is, the suppliers of his firm. In terms of the Model validation, the ELECTRE TRI software was used, as ELECTRE TRI was the Multicriteria Method selected. This software was developed at the Laboratoire d'Analyse et de Modélisation des Systèmes pour l'Aideà la Décision, at Paris Dauphine University, whose implementation followed the methodological guide and manual for users developed by Mousseau, Slowinski & Zielniewicz (1999).

Once this
Step was finished, the suppliers were classified according to their performance from a set of criteria on one of the three relationship levels of the Supply Chain. Subsequently the suppliers who meet the requirements that allow the creation of collaborative relationships in the Supply Chain of the Wind Energy Sector were identified.

Criteria adopted by the firms
After conducting the questionnaire, there was evidence that 07 criteria, which were not found among the 45 criteria validated by the literature, emerged from the open-ended questions. However, each one was mentioned by only one of the Managers; thus, it was not sufficient to include them in the framework. For that reason the criteria were selected based on the answers to the close-ended questions.
From the 45 criteria validated by the literature, the ones referenced by more than 50% of the Managers were chosen among price, technical capability, cost, experience, quality, financial stability, performance history, production capacity, technical knowledge, transparency, delivery, commit-ment and trust. In addition, from that same list of criteria the specific criteria of the collaborative relationship subproblem were extracted according to their degree of importance: information sharing, joint actions and interdependence. It is important to highlight that the price and the cost were merged together, because they have interrelated definitions.
Accordingly the criteria framework for Supplier Selection consisted of 15 criteria. There were 09 general criteria for the Supplier Selection problem, and 06 specific criteria for the collaborative relationship subproblem, also called soft factors, as shown in Chart 11. Therefore, the selection of the 15 criteria was a personal choice of the authors, who took into account the answers of the Managers that participated in the research.
The criteria that compose the framework were used in order to create and validate the Multicriteria Model subsequently.

Model Development
The Multicriteria Decision Aiding Model for Supplier Selection was created according to the standard procedure for resolving multicriteria decision-making problems proposed by Almeida (2013).
In step 1 the decision maker is described, his engagement is classified as being either direct or indirect, and the problem is identified, so that an individual or a joint decision is made. In this Model the problem consists of an individual decision, and the decision maker is the Manager responsible for the Selection of Suppliers of goods and services in the Wind Energy company.
In step 2 the strategic objectives are identified; the fundamental or end objectives that are essential to guide the effort when decision and evaluation of alternatives must be made; and the goal objectives that are useful for developing a Model in order to analyze decision-making problems and provide alternatives. Thus, the strategic objective aims to build collaborative relationships in the Supply Chain of the Wind Energy Sector. The end objective seeks to identify the suppliers who meet the requirements in order to create those kinds of relationships. The goal objective seeks to categorize the suppliers into the three relationship levels (cooperation, coordination and collaboration) according to his performance from the set of criteria.
In step 3 the criteria that represent the objectives shown in the previous step are established. For this Model 15 criteria were selected. There were 09 general criteria for the Supplier Selection problem, and 06 specific criteria for the collaborative relationship subproblem, also called soft factors, as shown in detail in Section 4.1.
The selected criteria meet the three necessary properties or requirements to adopt a coherent group of criteria proposed by Roy (1996), such as: nonredundancy, completeness and consistency. They also meet the properties proposed by Keeney (1992) and referenced by Almeida (2013), such as: measurability, operability and comprehensibility.
Some of the selected criteria that compose the Model are categorized as developed criteria, which are only adequate to a particular decision context and measured qualitatively (i.e.: joint actions Chart 11 -Criteria framework.

Criteria Definition
General C 1 Production capacity Coherence between the quantity of products and services required by the buyer and the quantity of products and services produced by the supplier. C 2 Technical capability Planning capacity and production scheduling in terms of inspections and tests, equipment, maintenance plans and workforce organization. C 3 Technical knowledge Technical skill to meet the buyer's requirement of products and services. C 4 Delivery Keeping to the agreements reached in terms of deadlines, quantity and transportation. C 5 Financial stability Current financial position of the buyer.

C 6 Experience
Business experience and knowledge in understanding the specification of the requirements of products and services. C 7 Performance history Information about events, occurrences and past supply performance. C 8 Price/Cost Value of a good or a service according to the customer's expectation, or consistent with the marketing environment of the sector. C 9 Quality Meeting of the specificities of products and services agreed by the parties.

Soft factors
C 10 Joint actions Joint commitment and engagement to resolve problems and conflicts; development of activities and production, quality, logistics, commercial and distribution processes; and arrangement of meetings, technical visits and training. C 11 Information sharing Willingness of the supplier to provide the necessary information in order to assist the decision-making process. C 12 Commitment Willingness to put in efforts and provide resources to support the relationship and achieve the goals of the Supply Chain. C 13 Trust Certainty that the partner is not to take advantage of the buying firm, and that he will honor his commitments for the relationship sake, including ethnic aspects and information secrecy. C 14 Interdependence Dependence of a member regarding a particular knowledge that a partner of the Supply Chain has, and vice versa. C 15 Transparency Communication with the parties when any problem or possible interruption in the Supply Chain activities is immediately detected.
Source: Research Data (2020). and information sharing); hence, they require the development of qualitative evaluation scales, or subjective indices or scales. Others are classified as natural criteria, which can be used in various decision contexts and are measured quantitatively (i.e.: price, cost and delivery). However, in terms of collaborative relationships the interpretation of natural criteria is considered to have a qualitative connotation (Guarnieri, 2012).
Therefore, a verbal evaluation scale with Likert quantitative characteristics composed of five levels, varying from 1 -very low to 5 -very high, which is compatible with the human cognitive capacity to distinguish between different evaluation levels, was adopted for all the 15 criteria. It is a bipolar scale that measures the superior or inferior performance in terms of a value of central or intermediate position (Almeida, 2013).
Chart 12 exemplifies the evaluation scales developed for each one of the 15 criteria used in the Model.
Chart 12 -Evaluation scale for the quality criterion.
Scale C9 -Quality 1 The supplier shows a poor performance when it comes to keeping to the agreements of the specificities of a good or service. 2 The supplier shows an unsatisfactory performance when it comes to keeping to the agreements of the specificities of a good or service. 3 The supplier shows a moderately satisfactory performance when it comes to keeping to the agreements of the specificities of a good or service. 4 The supplier shows a satisfactory performance when it comes to keeping to the agreements of the specificities of a good or service. 5 The supplier shows a highly satisfactory performance when it comes to keeping to the agreements of the specificities of a good or service.
In step 4 the structure of the set of actions is established (a discrete set or a set of continuous variables), the problem is identified (choice, classification or ranking), and the alternatives are provided. In this Model the set of actions corresponds to a discrete set, that is, a limited number of alternatives, which is described as A = {a 1, a 2 , a 3, ..., a n }, where the number of alternatives is equal to n. There is a classification problem (P.β ), since the alternatives are expected to be assigned to the classes, that is, to one of the relationships of the Supply Chain.
The alternatives are the suppliers S 1 , S 2 , S 3 , ..., S n of the Wind Energy companies who are evaluated according to their performance from the criteria C 1 , C 2 , C 3 , ..., C n , and subsequently categorized into one of the three classes. Figure 7 illustrates the classification structure of the alternatives of this Model. In step 6 the most appropriate preference structure to indicate the preferences of the decision maker is evaluated; the most appropriate rationality for the decision maker regarding the problem at hand is established, which can involve the use of the compensatory or the noncompensatory approach; and finally, a preliminary selection of the Multicriteria Method was made. This step must be developed in tandem with steps 7 and 8.
Because there is a classification problem (P.β ), the ELECTRE TRI Method was selected to compose this Model, which aims to assign alternatives to predefined classes ( In step 7 the intra-criteria evaluation was carried out, which depends on the type of Method to be adopted. Nonetheless, the results that affect the review of this choice made in the previous step can be obtained. In the Outranking Methods the evaluation of the indifference and the preference thresholds is considered (Almeida, 2013 • Preference threshold p j [g(b h )]: it represents the smallest different g j (a) -gj(b h ), compatible with the preference of "a" in relation to b h on criterion g j ; • Indifference threshold q j [g(b h )]: it specifies the largest difference g j (a) -g j (b h ), which preserves the indifference between "a" and b h on criterion g j .
The veto and the discordance thresholds, especially adopted in ELECTRE Methods, are also related to the intra-criteria preference of the decision maker. In the ELECTRE TRI Method a set of veto thresholds (v 1 (b h ), v 2 (b h ), ..., v m (b h )) is used in the discordance test, in which v j (b h ) represents the smallest difference g j (b h ) -g j (a), incompatible with the assertion aSb h ( In situations where there is the classification problem (P.β ), the description of the profiles to the classes is provided in this step, establishing the boundaries or limits between them (Almeida, 2013). As previously mentioned, the classes correspond to the three relationship levels of the Supply Chain: cooperation (Class III), coordination (Class II) and collaboration (Class I). The profiles that determine each one of these classes were defined according to the discussion presented in section 2.1, in which the limits were established taking into account the variation of the evaluation scale of the criteria (from 1 to 5).
The only restriction regarding the definition of these limits is that they must enable their comparison to the evaluation of the alternatives according to the criteria. Furthermore, it was evidenced that the best class (Class I) does not have a superior limit, and that the worst class (Class III)   Consequently, the suppliers who have the best performance according to all the criteria are sorted into Class I, because they meet the requirements that allow the creation of collaborative relationships in the Supply Chain. Whereas the suppliers who have an average or a weak performance are sorted into Classes II and III, for they still do not meet the requirements that allow the creation of collaborative relationships and need to make strategic changes, so that those relationships are built.
In step 8 the parameterization of the proposed Method is carried out, gathering the intra-criteria information from the decision maker, which enables the quantitative combination of the criteria in order to evaluate the alternatives. In the Outranking Approach Methods the inter-criteria evaluation can be represented by the weights of the criteria, which indicate the degree of importance (Almeida, 2013).
In this Model the weights of the criteria (w 1 , w 2 , w 3 , ..., w n ) must be determined by the decision maker. In order to indicate the relative importance of the criteria, direct rating as a weighing technique was used; accordingly, the decision maker should value each weight using the previously selected measurement scale, so that the values obtained are normalized (Gomes & Gomes, 2019). Thus, the selected scale varies from 0 to 100.
In step 9 the alternatives are evaluated broadly. Once the Model is consolidated, the selected Multicriteria Method is effectively implemented through the preferences of the decision maker according to the problem at hand.
In step 10 the sensitivity analysis is carried out, which consists of the study and analysis of the impact on the Model output caused by variations in the input; that is, the impact caused by the variations introduced in the input data, or in the Model parameters, were evaluated according to the results obtained from it (Almeida, 2013;Gomes & Gomes, 2019). In this step one can conclude that the final result of step 9 is not appropriate; thus, it requires a review of the previous steps.
The procedure implemented in this step depends on the problem at hand. The classification problem (P.β ) enables to evaluate to what extent the variations in the input data and in the parameters cause alterations that lead to, for example, the assignment of an alternative to a different class instead of the one that was initially determined by the Model (Almeida, 2013).
Almeida (2013) considers two types of procedures for the sensitivity analysis: a) isolated evaluation of parameters or of one type of input data, and b) joint evaluation of all parameters and input data, or a subset of those.
In step 11 after the two previous steps were concluded and there was no need to go back to the previous steps, the final analysis of the results is presented and the recommendations for the decision maker are made. In this step the decision maker is advised to what extent he can rely on the Model and the risks of the decision-making process are informed.
Finally, in step 12 the action is carried out, or the procedures of the set of referenced actions are adopted, which is not applied in the present research. Figure 8 summarizes the proposed Multicriteria Decision Aiding Model. The proposed Models allows to categorize the suppliers into three relationship levels in a systematic and structured way according to their performance from a set of criteria in order to identify the ones that meet the requirements that allow the creation of collaborative relationships in the Supply Chain. Moreover, it aims to support Managers by minimizing decisions made solely based on their experiences, which can lead to risks and uncertainty. Furthermore, it aims to have them identify with greater reliability and ease the suppliers that meet the requirements that allow the creation of collaborative relationships in the Supply Chain.
It is paramount to emphasize that the Model is broad; therefore, not only is it suitable for the Wind Energy Sector, but it can also be adapted to other production sectors. In doing so, the selected criteria and the parameters should be changed.

Model Validation
Initially the Manager chose to evaluate the 10 main suppliers with whom his company keeps the highest amount of purchasing negotiations, as shown in Chart 14.

S 1
Wind turbines S 2 Wind turbine blades S 3 Subcomponents and supplies for wind turbines S 4 Subcomponents and supplies for wind turbine blades S 5 Subcomponents of the hub S 6 Subcomponents of nacelle S 7 Wind turbine assembly S 8 Project development services S 9 Preconstruction and construction services S 10 O&M services of wind farms Source: Research Data (2020).
For the inter-criteria evaluation direct rating was the weighing method chosen. The Manager weighed the 15 criteria of the Model in a scale that varies from 0 to 100. Table 1 shows the weights of the criteria. Table 1 -Weights of the criteria.
Criteria C 1 C 2 C 3 C 4 C 5 C 6 C 7 C 8 C 9 C 10 C 11 C 12 C 13 C 14 C 15 Weights (w) 50 100 90 70  After that the weights were normalized by calculating their sum and dividing each one by the total sum. Table 2 shows the normalized weights.  For the intra-criteria evaluation the Manager evaluated the 10 suppliers according to the 15 criteria in a Likert scale, varying from 1 -very low to 5 -very high. Table 3 shows the consequence matrix, or decision matrix, which consists of the basic form of input data for most of the Multicriteria Methods. Each line of the matrix represents the consequence measures (or evaluations) of the alternative i in relation to the criteria m; and each column expresses the consequence measures of the alternative n in relation to the criteria j. An alternative i for a criterion j, results in the value function v j (a i ), which is based on the evaluation of consequences to be obtained in each criterion (Almeida, 2013;Gomes & Gomes, 2019). Table 3 -Consequence matrix.

Criteria General
Soft factors C 1 C 2 C 3 C 4 C 5 C 6 C 7 C 8 C 9 C 10 C 11 C 12 C 13 C 14

Application of the ELECTRE TRI Method
In the ELECTRE TRI application it is imperative to define the preference p j [g(b h )] and the indifference q j [g(b h )] thresholds for each criterion, which are the necessary information for the intra-criteria evaluation, enabling one to consider hesitation and uncertainty as related to human judgment ( For the application of the ELECTRE TRI Method the software ELECTRE TRI 2.0a requires input data in order to calculate: the criteria list (name, code, weight and preference direction); the classes (name) and the profiles (name, code and limit); list of alternatives (name, code and performance in each of the criteria); preference p j (b h ), indifference q j (b h ) and veto v j (b h ) thresholds; and the cutting level λ .
An intermediate value of λ = 0.7 was primarily chosen, considering that the cutting level λ can vary from 0.5 to 1, and that the higher, the more rigorous the assignment process of the alternatives.
The alternatives were assigned according to two assignment procedures: the pessimistic procedure, in which the assignment is carried out in a more conservative way, that is, the alternatives are assigned to an inferior class other than the one determined by the profile; and the optimistic procedure, which is less conservative and the alternatives are assigned to a superior class (Costa, Santafé Júnior & Haddad, 2007; Guarnieri, 2012).
On one hand, when there were convergences between the two procedures, the system created to evaluate and classify the alternatives was capable of comparing them to the profiles of the classes. On the other hand, when there were divergences, the system was incapable to do so, which could be caused either by incoherence of the evaluator, and of the Model (including the set of criteria), or by the data collection system. The divergence of these classifications is common in situations where there are conflicting criteria, which is inherent to the problem; thus, it can not be seen as a modeling error ( According to Chart 15, for the cutting level λ = 0.7 the assignments of the alternatives S 7 , S 8 , S 9 and S 10 converged in both assignment procedures, whereas the alternatives S 1 , S 2 , S 3 , S 4 , S 5 and S 6 diverged. When there is a divergence, it is likely that the decision maker chooses the pessimist assignment procedure in case he is more strict; however, he is likely to choose the optimistic assignment procedure in case he is more flexible (  Coordination Collaboration S 4 Coordination Collaboration S 5 Coordination Collaboration S 6 Coordination Collaboration S 7 Collaboration Collaboration S 8 Collaboration Collaboration S 9 Coordination Coordination S 10 Collaboration Collaboration Source: Research Data (2020). Table 4 -Statistical overview of the assignments for a cutting level λ = 0.7.

Sensitivity Analysis
In the sensitivity analysis the isolated evaluation of parameters was used, that is, of a single type of input data, the cutting level λ , since the preference p j (b h ) and the indifference q j (b h ) thresholds equal to zero were established for this Model, and the veto thresholds v j (b h ) were disabled.
Chart 17 shows the assignment of the alternatives to the classes for λ = 0.6.

Alternatives
Pessimistic Optimistic S 1 Collaboration Collaboration S 2 Collaboration Collaboration S 3 Collaboration Collaboration S 4 Collaboration Collaboration S 5 Collaboration Collaboration S 6 Collaboration Collaboration S 7 Collaboration Collaboration S 8 Collaboration Collaboration S 9 Collaboration Collaboration S 10 Collaboration Collaboration Source: Research Data (2020).
As shown in Chart 17, for the cutting level λ = 0.6 the assignments of all alternatives converged in the two assignment procedures. Table 5 presents a statistical overview of the assignments of alternatives. Table 5 -Statistical overview of assignments for a cutting level λ = 0.6.
Chart 18 shows the comparisons of the alternatives and the profiles of the classes.
As shown in Chart 18, the alternatives do not present incomparabilities when compared to the profiles of the classes, because the value of the cutting level is smaller and less rigorous than the previous one (λ = 0.7) (Guarnieri, 2012).
Chart 19 shows the assignment of the alternatives to the classes for λ = 0.8.  Chart 19 -Assignment of the alternatives for λ = 0.8.

Alternatives
Pessimistic Optimistic S 1 Coordination Collaboration S 2 Coordination Collaboration S 3 Coordination Collaboration S 4 Coordination Collaboration S 5 Coordination Collaboration S 6 Coordination Collaboration S 7 Collaboration Collaboration S 8 Collaboration Collaboration S 9 Coordination Collaboration S 10 Collaboration Collaboration Source: Research Data (2020).
As shown in Chart 19, for the cutting level λ = 0.8, the assignment of the alternatives S 7 , S 8 and S 10 converged in the two assignment procedures; whereas the alternatives S 1 , S 2 , S 3 , S 4 , S 5 , S 6 and S 9 diverged. Therefore, one verifies that the assignment of alternatives for this cutting level shows a higher divergence, which indicates the incapacity of the system to compare the alternatives to the profiles in higher cutting levels. Table 6 presents a statistical overview of the assignments of alternatives.
Chart 20 shows the comparisons of the alternatives and the profiles of the classes.
As shown in Chart 20, 07 alternatives present incomparabilities when compared to the profile of Class II; thus, it shows that the decision maker is not able to make all the comparisons, or that he does not wish to make them (Almeida, 2013). Table 6 -Statistical overview of the assignments for a cutting level λ = 0.8.
Alternatives Class I Profiles Class II Profiles S 1 Strict Preference Incomparability S 2 Strict Preference Incomparability S 3 Strict Preference Incomparability S 4 Strict Preference Incomparability S 5 Strict Preference Incomparability S 6 Strict Preference Incomparability S 7 Strict Preference Strict Preference S 8 Strict Preference Strict Preference S 9 Strict Preference Incomparability S 10 Strict Preference Strict Preference Source: Research Data (2020).
Chart 21 compares the assignments of the alternatives for the three cutting levels.
Chart 21 -Comparison of the assignments for different cutting levels.
It is observed that the assignments of all alternatives converged in the two assignment procedures only for the cutting level λ = 0.6. Guarnieri (2012) states that the lowest cutting levels offer more convergent assignments; however, they also reduce the level of credibility of the Model in providing more rigorous solutions. In contrast, higher levels increase the reliability of the Model and enable more accurate decisions to be made.
It is worth saying that in the sensitivity analysis other input data, or parameters, could be altered in order to evaluate the impact of the Model, such as the preference, the indifference and the veto thresholds, and the weights of the criteria.

Final analysis of the results and recommendations for the decision maker
The best results were obtained for the cutting level λ = 0.6, in which not only did the assignments of all alternatives converge in both assignment procedures, but also that there were no incomparabilities. Thus, at this level the 10 evaluated suppliers showed a high performance in the set of criteria and were classified in Class I -Collaboration. That is, they meet the requirements that allow the creation of collaborative relationships in the Supply Chain in the Wind Energy Sector.
For that reason it is feasible to adopt collaborative strategies through partnerships and long-term contracts between the Wind Energy company and these 10 suppliers, which tend to contribute significantly to the minimization of disharmony in the supply of goods and services, the reduction of costs, the increase in competitivity and the warranty of supply, as well as other advantages.
The result obtained in this application does not seek to provide the decision maker with the solution to his problem, or establish a single truth, but to support the decision-making process by recommending actions or courses of actions.

CONCLUSION
The present paper proposed a Supplier Selection Model in order to build collaborative relationships in the Supply Chain, so that the Suppliers are previously categorized into the cooperation, coordination and collaboration levels. The ELECTRE TRI Method was used as a tool to achieve the main goal of the research.
The proposed Model aims to aid the decision-making process of managers whose firms are involved in the Supply Chains, and not only the ones from the Wind Energy Sector. Thus, that allows them to identify with greater reliability and ease the suppliers who meet the requirements that allow the creation of collaborative relationships.
Firms can use the Model periodically in order to revise the supplier assessment and, if needed, direct them to either an inferior or superior class. For example, a supplier who has been primarily categorized into the collaboration level (Class I) can, due to a new evaluation, show a poor performance according to the set of criteria; hence, he can be reassigned to an inferior class, either to coordination (Class II) or to cooperation (Class III).
The increase in the quantity of firms and Managers is recommended, so that there can be a more reliable identification of the criteria selected for the Supplier Selection, and that the proposed Model is implemented in other productive sectors where there is an incentive to the adoption of strategies for Collaborative Supply Chains.
Moreover, it is recommended the use of other softwares for future applications of the ELECTRE TRI Methods, such as the J-ELECTRE-v2.0. The latter is an executable .jar file that does not need to be installed and can be executed in any operating system, as it only requires a recent JAVA SE program. This software was developed by Pereira, Costa and Nepomuceno (2019), Professors at the Federal University of Fluminense -UFF, and it is available on the following website: https://sourceforge.net/projects/j-electre/files/.
In conclusion, it is also recommended that in order to compare the results other suitable Multicriteria Methods to resolve the classification problem are selected, such as PROMSORT, which is derived from PROMETHEE. As ELECTRE TRI, the latter belongs to the Outranking Approach or the French School in Multicriteria Decision Aid. In addition, it is suggested the use of the Fuzzy Logic, which is commonly integrated to the Multi-criteria Methods due to its capacity to easily capture the subjective evaluations of the decision makers; thus, it models the subjective processes of human evaluation, converting the evaluation scales into Fuzzy numbers.