ترکیب شبکه های بیزی و روش هزینه کل مالکیت برای تجزیه و تحلیل انتخاب تامین کننده

In this study, we analyze the supplier selection process by combining Bayesian Networks (BN) and Total Cost of Ownership (TCO) methods. The proposed approach aims to efficiently incorporate and exploit the buyer’s domain-specific information when the buyer has both limited and uncertain information regarding the supplier. This study examines uncertainty from a total cost perspective, with regards to causes of supplier performance and capability on buyer’s organization. The proposed approach is assessed and tested in automotive industry for tier-1 supplier for selecting its own suppliers. To efficiently facilitate expert opinions, we form factors to represent and explain various supplier selection criteria and to reduce complexity. The case study in automotive industry shows several advantages of the proposed method. A BN approach facilitates a more insightful evaluation and selection of alternatives given its semantics for decision making. The buyer can also make an accurate cost estimation that are specifically linked with suppliers’ performance. Both buyer and supplier have clear vision to reduce costs and to improve the relations.

Effective operations for companies are vital for success in the marketplace. This can only be achieved by integrating suppliers who provide high quality products, flexible operations, and systems; who maintain close relations, and who contribute to the product design operations (Stevenson, 2009). Therefore, selecting the right suppliers has become one of the most important purchasing functions in supply chain management (Boer et al., 2001 and Chen et al., 2006). With the increasingly important role of suppliers in supply chain management, the selection process strategy has changed; other than scanning a series of pricelists, a wide range of qualitative, quantitative and environmental criteria has now been folded into the process (Ho et al., 2010 and Humphreys et al., 2003). Researchers have proposed a number of methods to measure the suppliers’ performance and select them according to the determined criteria (Degraeve and Roodhooft, 1999 and Roodhooft and Konings, 1997). Although each approach has advantages in terms of selecting and evaluating the supplier, ultimately they also have some limitations. First, none explicitly consider the uncertain nature of the problem context. Uncertainty in supplier selection primarily arises in two different ways: uncertainty of supplier performance on a specific criterion such as uncertainty in delivery reliability of the supplier, and uncertainty of the resulting poor performance effects of a supplier on the purchasing company, such as uncertainty in costs at the buyer due to the delivery performance of the supplier. Second, the buyer sometimes needs to make a decision about a supplier with only limited experience or information regarding the supplier. However the buyer might have some domain-specific knowledge that makes a difference and needs to be accounted for in the selection process. Current selection models do not explicitly account for this type of variation in the process. Furthermore, supplier selection criteria have specific causal relations and consequences with relationship to the buyer, and many models have shortcomings in terms of formalizing these relations. For example, if the supplier shows poor performance on a delivery capability, this drawback can easily increase multiple cost items: downtime costs, operation costs, logistics costs, etc. Modeling and exploring the interdependencies among variables, the supplier and buyer recognize accurate effects of supplier performance. This results in improved operations and relations between supplier and buyer. In this study, we propose an integrated approach combining Bayesian Networks (BN) and Total Cost of Ownership (TCO) to overcome the aforementioned limitations of current approaches. Our goal is to more clearly identify uncertainty issues, and integrate and utilize the buyer’s domain-specific knowledge. Even if the buyer has incomplete information, he can still evaluate and select alternatives with the proposed approach. The model will also explore the interdependent relations between supplier selection and different cost items in the selection process. The application process is presented for a tier-1 supplier in automotive industry. Bayesian Networks (BN) are very powerful for making inferences and drawing conclusions based on available information (Jensen, 1996). They are effective for modeling uncertainty by accepting probability distributions. BN can combine expert and domain knowledge that allows flexible inference even with partial and limited information (Lauritzen, 1995). The domain knowledge of a buyer normally encodes in the form of conditional statements. BN allow modeling of probabilistic causal relations among variables (Bishop, 2006). Therefore, BN can facilitate a more insightful evaluation and selection of alternatives given the semantics used for decision making. On the other hand, TCO provides a better inspection opportunity for determining the total cost caused by supplier activities on a buyer’s organization. The TCO approach is a structured methodology for determining the true cost of acquisition of a product, considering all the costs related to purchasing and using the product. TCO considers the buyer’s entire value chain and mainly evaluates the supplier performance by taking into account all the costs caused by a supplier (Degraeve, Labro, & Roodhooft, 2000). These costs are not limited to the purchasing price but also include cost elements such as: quality, transportation, maintenance, and administration (Degraeve et al., 2000 and Ellram, 1995). As opposed to an initial-price perspective that mainly accepts short term approach, TCO allows for a long-term perspective selecting different buying situations (Ferrin & Plank, 2002). The remainder of this paper is structured as follows. In Section 2, we present the relevant research concerning the supplier selection. In Section 3, we provide a brief overview of Bayesian Networks, including inference. In Section 4, we explain the details of our model. In Section 5, we test our proposed framework using an illustrative example, and present the results and detailed sensitivity analyses to identify the critical factors in the supplier selection process. The value of information is discussed for both mean and variance points of views. The last section is allocated for a summary and conclusion of the proposed method.

The proposed method combining Total Cost of Operations (TCO) and Bayesian Networks (BN) provides several advantages in analyzing the supplier selection problem. The proposed framework accounts for and handles uncertainty. The probabilistic nature of the supplier selection problem due to supplier performance and its result on buyer organization makes BN a powerful candidate comparing to the other methods. We explicitly model the uncertainty of the total cost and compared supplier performance by accepting probability distributions over different cost items and selection criteria. The proposed model with BN allows a system of reasoning under the absence of complete information. It is very common for a buyer to select suppliers with limited information due to the lack of past experience. The proposed approach encodes domain and limited information of the buyer in a structured framework so that flexible inferences are possible, and drawing conclusions from these queries is feasible. This approach also facilitates to combining domain knowledge and historical data. Another advantage of the proposed model is to guide data collection and strategic thinking. Decomposing the supplier selection problem into a set of casual relationships allows modelers the complete understanding of the problem domain. Decision makers, subject matter experts, engineers from different departments and disciplines need to follow a structured framework while identifying criteria and their states and casual mechanisms within and between criteria and different cost items. It is recommended to organize brainstorming sessions to guide thinking and to encode the perceptions of decision makers. Expert opinions and judgments are on the center of the proposed model. The better decisions situations occur only if the knowledge of stakeholders is directed in a well-organized way. In our study, we made a clear distinction between supplier selection criteria that are mostly observable, and the factors that explain these criteria. This hierarchical approach resulted in an increase in efficiency while eliciting the required information from subject matters expert. Professionals now need to identify fewer parameters and need to follow a structured approach. This study not only compares the suppliers based on mean total cost, but also compares variability of these cost items based on expert perceptions. The buyer will have insights as to the relative importance of cost items as well as an estimation of various cost items, and total cost caused by the performance of the supplier. Furthermore, the model provides a framework that allows for testing different continua of information settings. Our complete and intense sensitivity analyses reveal that flexibility, delivery, and price are among the most critical factors that make the highest difference in supplier selection. The presented case study already shows the practicality and flexibility of the approach for selecting and evaluating of the potential suppliers and determining a healthy supplier base. This approach is applicable to real scenario especially when the buyer does not have history of relations with suppliers. When there is limited information regarding the suppliers, the belief of the decision makers plays a critical role in decision making to reduce the vagueness. It simultaneously considers quantitative and qualitative criteria in supplier selection. Furthermore, the randomness of the possible outcomes results in probabilistic variability that entails systematic errors. The proposed method is well suited to deal with both vagueness and probabilistic variability. Although the selection criteria, relations among criteria and cost elements as well as their affects on costs are unique to automotive industry in this study, provided framework can easily be generalized to any other process. The unique requirements can be easily extended according to the needs of the company with proposed framework.