انتخاب تامین کننده در زنجیره تامین چابک: مدل پردازش اطلاعات و یک تصویر

Agile supply chains need to be highly flexible in order to reconfigure quickly in response to changes in their environment. An effective supplier selection process is essential for this. This paper develops a model that helps overcome the information-processing difficulties inherent in screening a large number of potential suppliers in the early stages of the selection process. Based on radial basis function artificial neural network (RBF-ANN), the model enables potential suppliers to be assessed against multiple criteria using both quantitative and qualitative measures. Its efficacy is illustrated using empirical data from the Chinese electrical appliance and equipment manufacturing industries.

The task of supplier selection has always been considered a key one within purchasing and supply management (Dickson, 1966; Kraljic, 1983; Weber et al., 1991; Choi and Hartley, 1996; De Boer et al., 2001; Sarkar and Mohapatra, 2006). Indeed, some consider it to be the most important of all the responsibilities of the function, as the choice of supplier has a significant impact on the optimisation of the quality, quantity, timeliness and price of purchased goods and services (Dulmin and Mininno, 2003; Sarkis et al., 2007). Furthermore, suppliers have a direct and significant impact on the quality, cost and lead time of new products and technologies needed to meet new market demands (Vokurka and Fliedner, 1998; Meade and Sarkis, 1999; Humphreys et al., 2007). Three recent trends in purchasing and supply practices have further served to emphasize the importance of selecting the right supplier. Firstly, the increased use of outsourcing has led to more firms spending a greater proportion of their total revenue on externally sourced goods and services, thereby increasing the impact of suppliers’ performance on purchasers (Weber and Ellram, 1993). Secondly, the trend towards supply base reduction increases the impact that any given supplier is likely to have on a purchaser's performance (Power et al., 2001). Thirdly, and perhaps most importantly, the trend towards closer relationships between vendors and purchasers based on collaboration and co-operation increases the role and contribution of suppliers in the operations of the purchaser (Heide and John, 1990). This is especially the case where purchasers adopt a ‘partnership sourcing’ (Macbeth and Ferguson, 1994) approach, increasing the purchasers’ dependence upon their suppliers to the extent that suppliers can become integral to their core competences (Gadde and Snehota, 2000). This can have a significant impact on competitiveness because it facilitates the mobilisation of resources to track evolving changes in markets, technologies and material development as well as customer demand. Interdependent partners can focus and rapidly replicate narrow aspects of the value creation process where competitive advantage is greatest (Quinn, 1992). The task of selecting the right suppliers is also an extremely demanding one for three key reasons. Firstly, today's business environments are typically seen as becoming inherently more unstable due to fast-changing market conditions, customer demands, actions of competitors, and so on (Hakansson and Snehota, 2006). In more dynamic markets, the response of many firms has been to adopt the concept of the agile supply chain (ASC) (Christopher, 2000). An agile supply chain is a dynamic alliance of member companies, the formation of which is likely to need to change frequently in response to fast-changing markets. In ASCs, the task of supplier selection is thus not a one-off infrequent activity. Rather, changing market requirements and customer preferences require a broader and faster supplier selection process (Sarkis, 2001; Arteta and Giachetti, 2004), which requires the use of a wide set of selection criteria (Yusuf et al., 1999; Cagliano et al., 2004). The adoption of ASCs may increase the frequency with which a purchaser needs to seek out a new strategic supplier. Although the partnership sourcing approach envisages a long-term relationship between purchaser and supplier, the requirements of more dynamic markets may even lead a purchaser to re-evaluate the basis of these relationships more frequently. Secondly, the increasing globalisation of world trade and Internet-facilitated communications provides purchasers with increased opportunities for sourcing goods in foreign countries. Indeed, many businesses have seized such opportunities. In particular, China has become the ‘world's factory’ (Nassimbeni and Sartor, 2007), the source of some one-third of the West's manufactured goods. Global sourcing increases the number of potential suppliers considerably when compared to restricting the choice to domestic suppliers. However, it can also increase the difficulties of obtaining meaningful information about potential suppliers. Purchasers may have to rely solely on publicly available information, especially in the early stages of a search for new suppliers. Thirdly, potential suppliers are likely to need to be assessed against multiple and often conflicting criteria, between which trade-offs are typically required (Chen et al., 2006). Furthermore, selection is complicated by the fact that different potential suppliers may have different performance characteristics for different attributes (Xia and Wu, 2007). For example, the potential suppliers who can provide a raw material at the lowest price may not have the best quality or after-sales service among the competing suppliers. Therefore, supplier selection is an inherently multi-objective decision, which seeks to minimise some evaluation criteria whilst and, at the same time, maximising others (Dickson, 1966). These factors combine to make supplier selection in ASCs a very complex problem. It might be expected that purchasing and supply managers would look to use quantitative methods to help inform their decision-making. However, until recently quantitative methods have typically been found to be inadequate and limited. Many decision-support models emphasize the final stages of the supplier selection process (e.g. Weber et al., 1991; Weber and Ellram, 1993; Dulmin and Mininno, 2003). However, De Boer and Van der Wegen (2003) have argued for the need to shift the emphasis to earlier stages, such as pre-classification and the formulation of criteria, as the quality of decision-making at the final choice stage is largely dependent on decisions that have already occurred in the previous stages. Secondly, relatively little attention had been paid to the task of selecting new suppliers (De Boer et al., 2001), which is particular important in ASCs. Thus, more and more managers have fallen back on the use of purely qualitative methods, often based on subjective judgements, to address the problem of supplier selection (Gencer and Gurpinar, 2007). Recent advances in computer programming have provided opportunities to develop new quantitative approaches that can help decision-makers. In particular, artificial neural networks (ANN), especially radial basis function artificial neural networks (RBF-ANNs) appear to have the potential to help in supplier selection. RBF-ANN is a system with strong adoption ability, which can consider and adopt new restrictions over time. This characteristic is particularly important in ASC supplier selection, which typically takes place under the condition of information imperfection and distortion. Despite its apparent suitability for use in fast-changing economic and competitive environments, few researchers have tried to apply RBF-ANN to the supplier selection problem in ASC. This paper seeks to contribute to the advancement of methods available to improve the supplier selection process in two ways. Firstly, it aims to build on existing literature, (notably the work of De Boer et al., 2001; Lin and Chen, 2004) to develop a practical model of the supplier selection process. Secondly, it aims to develop a quantitative method of classifying potential suppliers into one of four different types (using the classification of Kraljic, 1983), based on RBF-ANN, which is able to overcome the information-processing difficulty inherent in assessing large numbers of potential suppliers against multiple criteria. The development of such a model would help purchasers to select suppliers from amongst those most appropriate to their supply strategy. The paper is structured as follows. Section 2 reviews the existent literature on supplier selection, pointing to the deficiencies in current models and methods used to tackle the problem. Using this literature, Section 3 develops a conceptual model of the supplier selection process. Section 4 draws on the computational literature to develop and present an information-processing model that uses RBF-ANN for classifying potential suppliers in an ASC. By way of illustration, the model is developed for application within the Chinese electrical appliance and equipment manufacture industry. Section 5, then shows how the model can be applied in practice by applying it to firms within that industry, using publically available data. Section 6 closes the paper with a short discussion of the issues raised and pointing the way to future research requirements.

The uncertainty and ambiguity involved in the decision-making process, makes supplier selection in ASC highly complex. This is particularly the case when there are a very large number of potential suppliers to be considered. As such, the process would benefit from an increased formalization aimed at reducing the subjectivity of those involved. Decision-making would undoubtedly be improved if decision-makers could spend more time interpreting evaluations of potential suppliers rather than on considering how the evaluation is conducted. A key feature of the conceptual model presented in this paper is the use of the classification matrix, which increase the visibility of the assessment of each potential supplier's strength and weakness. This provides an opportunity for decision-makers to make more rational judgements. The proposed model has significant advantages in comparison with other alternatives. Firstly, in comparison to mathematic programming and linear weighting, the RBF-ANN model can easily consider both quantitative and qualitative measures (Lee and Ou-Yang, 2009). Secondly, the RBF-ANN model has a self-learning ability that other models do not (Moody and Darken, 1989). Thirdly, compared to cluster analysis, the RBF-ANN model can tolerate inexact and/or missing data (Bianchini et al., 1995; Hassoun, 1996). Fourthly, because of its black box characteristic, the RBF-ANN model is easy to use by decision-makers compared to the fuzzy set approach (Jang and Sun, 1993). More concretely, as demonstrated by the empirical illustration in Section 5, the use of the RBF-ANN information-processing model has the potential to improve the efficiency of the second phase in ASC supplier selection processes. In our view it offers advantages over other supplier selection models. For example, the use of cluster models requires that every single potential supplier is classified against the required criteria. However, the use of an RBF-ANN method such as this, requires that only a relatively small number of potential suppliers be so classified. Using these as the training samples within RBF-ANN enables all other potential suppliers to be classified automatically and efficiently, generally within very few seconds. This enables large numbers of potential suppliers to be classified into different types on the basis of potentially huge datasets. This enables the final selection of suppliers to be made from amongst similar types. This can not only help to improve the efficiency of supplier selection process but also the quality of decision-making. This should in turn help the purchasing organization establish a competitive advantage within its supply chain compared with its competitor supply chains. The RBF-ANN model for supplier selection is useful and will help to make the final selection more efficient. The model is especially suited to purchasing situations where there are large numbers of potential suppliers to consider. It has both theoretical and practical strengths. Being based on well-established management science, RBF-ANN offers a reliable and knowledge-based model. The model reduces the solution space for the supplier selection problem by categorizing the potential suppliers accurately and speedily. Use of Kraljic (1983)'s matrix alongside RBF-ANN enables the otherwise very daunting task of evaluating all potential suppliers against all required criteria to become a practical proposition. It is worth emphasizing that use of this method enables decision-makers to build their own customised criteria according to their own industry's characteristics. The criteria proposed in this paper are just one example of what might be done in practice. This enables the original criteria specified by Kraljic, which may no longer be suitable for all supplier selection situations today, to be improved and modified according to the particular decision-making environment. Therefore, even though the model requires an inescapable and relatively long data preparation time like others previous models (e.g. Bevilacqua et al., 2006; Sha and Che, 2006), it still can be seen as a very useful and necessary phase in the supplier selection processes (as shown as Fig. 1). In real-business decision-making situations, the RBF-ANN model simplifies what might otherwise be a complicated pre-classification process whilst keeping the model output reliable and helpful. The application of the RBF-ANN information-processing model proposed in this paper, demonstrates the relative user-friendliness of the approach when compared with other methods, such as mathematic programming and AHP, especially during both the training and application phases. In addition to the use of publicly available data, the model requires decision-makers to make only simple judgements about the potential suppliers’ characteristics. One advantage of the proposed method is that organizational decision-makers do not need to understand the details of how RBF-ANN works. They do not need to know what happens within the RBF-ANN calculations. They need only to give the inputs that the RBF-ANN requires (i.e. their evaluation and judgement on certain criteria). The RBF-ANN will then produce the outputs automatically and efficiently (Jang and Sun, 1993). Despite these advantages, it must be recognised that an RBF-ANN information-processing model also has some disadvantages. Firstly, the assessment it provides depends entirely on the criteria applied. These remain within the control of those involved in the decision-making process. It should also be noted that the training phase can be affected by the complexity of the decision-making environment and any incoherence or inconsistencies stemming from the experts’ judgements about qualitative criteria or within the quantitative data used. If such instances do occur, it would be necessary to extend the training phase by introducing further samples or additional data to improve the knowledge representation. Secondly, compared with other technologies, RBF-ANN needs a complicated and potentially time-consuming data preparation phase. Building the RBF-ANN requires that data is available from a sufficiently large representative sample of suppliers in order to undertake network training. Importantly, the model also assumes the availability of an adequate supply of data on all potential suppliers under consideration. Finally, it needs to be recognised that the training phase of an ANN takes place within a specific decision-making context. If this changes it is necessary to re-train the network. As the use of RBF-ANN in supplier selection is a relatively new application, there is a considerable need for further research. In particular there is a need to apply and test the model in practice to establish its efficacy in real organizational supplier selection decision-making situations. One interesting and potentially useful exercise would be to compare the performance of this model with that of other supplier selection models that provide similar type of output (e.g. cluster models, DEA, screening procedures). Also, further work is required to overcome the flaws in the model discussed above. Also, the question of how to mix an RBF-ANN information-processing model with other technologies as a part of complex system of ASC supplier selection needs further investigation. A further interesting research direction is that of testing Hinkle et al.'s (1969) model vis-a-vis the RBF-ANN model to compare their efficiency and effectiveness. Finally, the issue of how to apply the next two stages of the 4PCM, namely final selection and application feedback, needs to be addressed in detail.