An agent-based modeling framework for the design of a dynamic closed-loop supply chain network

Globalizing world conditions provide a great market opportunity to businesses. In 2020, the global supply chain management market was valued at 15.85 billion U.S. dollars and is expected to reach almost 31 billion U.S. dollars by 2026 [118]. To gain a competitive advantage, businesses give more importance to the design and management of their supply chain networks day by day. The supply chain network is a complex structure that consists of actors, such as suppliers, manufacturers, distribution centers, retailers, customers, and the flows between them (Fig. 1).

Interest of traditional forward supply chains starts with the process of acquisition of raw materials from suppliers and moves on to the production of products at manufacturers, and the delivery of products through distributors and retailers to customers. Examining, designing, and managing of this complex structure are crucial for the efficiency of businesses. In a 2020 survey, demand-side challenges, such as faster response time were cited among the most difficult hurdles supply chain companies face [118]. On the other hand, there are some catastrophic impacts of supply chain activities in terms of environment. Loss of biodiversity, deforestation, running out of raw materials, toxic waste, damage to ecosystems, hazardous emissions, and depleting landfill capacities are some of these impacts. According to the report by the World Wide Fund for Nature (WWF) until 1970 humanity was demanding resources and emitting carbon dioxide at a rate that the ecosystem within their borders could keep up with, however the situation has dramatically reversed in the last fifty years because of explosion in global trade, consumption and human population growth, as well as an enormous move towards urbanization. Humanity enterprise currently demands 1.56 times more than the amount that Earth can generate and an average 68% decline occurred in population sizes of mammals, birds, amphibians, reptiles, and fish between 1970 and 2016 [102].

Nowadays, there is a growing need to deal with the challenges made by supply chain operations on the global environment and control their negative impact [67]. Thus, with an increasing importance of sustainability, reverse logistics, ecological footprint concepts, businesses have started to incorporate reverse flows into their supply chain networks for several reasons, such as preventing environmental deterioration by reducing resource and energy consumption during industrial activities, generating less waste, meeting the expectations of conscious customers or stakeholders, and fulfilling the requirements of environmental protection and waste management laws obliged by governments. In addition to forward flow extending from supplier to customer, the systems dealing with reverse flow simultaneously for recovering or disposing properly of used products are called closed-loop supply chain (CLSC) networks. Reverse flow-specific facilities, such as collection centers, disassembly centers, repair centers, recycling centers, refurbishing centers, remanufacturing centers, and landfills may be located in CLSC networks for the efficient management of recovery and disposal of returns along with located centers in traditional supply chains. Efficient design and modeling of CLSC networks is crucial for the management of these centers and for both the forward and reverse flows between them, simultaneously.

In recent years, humanity has faced many threats such as the depletion of natural resources and increasing natural disasters due to the deterioration of the ecological balance. Because, environmental concerns of governments and customers have increased, the interest in CLSCs are getting attention in both academy and practice. For a CLSC that can compete in the global market successfully, effective supply chain network design is inevitable. There are many studies addressing CLSCs from different perspectives, such as network design and modeling, performance measurement, inventory management, and scheduling in the literature. Here, only CLSC network design and modeling studies are mentioned due to widespread literature. The first comprehensive studies on CLSC modeling in the literature are studies by Jayaraman et al. [47] and Fleischmann et al. [35]. Jayaraman et al. [47] presented a binary mixed integer programming model for obtaining the optimum quantities of transported and produced quantities and locations of remanufacturing/distribution centers. Fleischmann et al. [35] proposed a general location problem for the product recovery network. They compared the reverse logistics networks with traditional forward ones.

In addition to these studies, some of the comprehensive studies about CLSC network design and modeling are mentioned below. Pishvaee et al. [76] proposed a robust optimization model for handling the inherent uncertainty of input data in a CLSC network design problem. Amin and Zhang [8] examined a CLSC network containing multiple products, facilities, collection centers, and demand markets and proposed a mixed integer linear programming (MILP) model that minimizes total cost. The proposed model was solved by two ways including weighted sums and e-constraint methods. Demirel et al. [29] proposed a mixed integer programming model for a CLSC network with multi-periods and multi-parts considering two main policies as secondary market pricing and incremental incentive policies. In the study, using a realistic network example, several scenarios were generated and explored. The authors investigated the effects of various parameters such as demand, capacity, purchasing costs and size of the network on the performance of the problem. Rezapour et al. [82] proposed a bi-level model for strategic reverse network design and operational planning of a closed-loop, single-period supply chain operating in a competitive environment with price-dependent market demand. An existing supply chain comprising the production and distribution of new products, and a new competing supply chain with both new and remanufactured products was considered in the study. Özceylan et al. [71] presented CLSC network design and modeling for end-of-life vehicles treatment in Turkey. They developed a linear programming model to handle the reverse material flows with the aim of reintegrating them into forward supply chains. Several CLSC scenarios were discussed to show the performance of the proposed model and its applicability in the automotive industry. Paydar et al. [74] proposed a MILP model for the CLSC consisting of the collection and distribution of used engine oil. Robust optimization approach was used in the study to deal with the uncertainty in the amount of oil collected. Hajiaghaei-Keshteli and Fard [51] developed a new mixed integer non-linear programming model to formulate a multi-objective CLSC network design problem that considers the transportation cost reduction supposition. In the solution of the problem, in addition to traditional and current metaheuristic methods, the algorithms were used by hybridizing according to their strengths. Sahebjamnia et al. [85] developed a model for closed-loop tire supply chain network design considering economic, environmental, and social dimensions. In the study, the main advantages and disadvantages of individual algorithms were considered and four new hybrid metaheuristic algorithms were developed to solve large-scale problems. Yıldızbaşı et al. [107] developed a mixed integer programming model for automotive CLSC network design problem. Four different interactive fuzzy programming approaches were used in the study to tackle the trade-offs among the objectives. Fakhrzad and Goodarzian [33] presented a new fuzzy multi-objective programming approach for a production–distribution model for a multi-product, multi-period, and multi-level CLSC problem. The problem was formulated as multi-objective MILP model.

Goodarzian et al. [41] proposed a new multi-objective, multi-stage, multi-product, multi-period pharmaceutical supply chain network with the problem of production, distribution, purchasing, ordering, inventory, keeping, allocation, and routing under uncertainty. In the study, a new robust fuzzy programming method was developed. Isaloo and Paydar [45] aimed to improve performance in conditions of uncertainty associated with the plastic injection industry. Accordingly, the authors proposed a bi-objective mathematical programming model aimed at minimizing environmental emissions and total cost of the system including transportation costs, operating costs, and plant establishment fixed costs. Nayeri et al. [69] proposed a multi-objective mathematical model to configure a CLSC network for a water tank, considering sustainable measures. In the study, fuzzy robust optimization was applied to deal with the uncertainties of the CLSC network problem caused by changes in parameters such as transportation costs and demands. The proposed model was solved using goal programming approach. Pourmehdi et al. [78] developed a multi-objective linear mathematical model under uncertainty to optimize a sustainable CLSC for steel. The uncertainty was modeled in the stochastic environment and the proposed model was developed through a fuzzy goal programming approach. Abdi et al. [1] developed a new stochastic optimization model for the CLSC network design problem. Whale optimization algorithm which is a new nature-inspired algorithm and particle swarm optimization were used to address the model developed using a two-stage stochastic programming. Babaeinesami et al. [16], proposed a multi-objective model in the area of CLSC problem integrated with lot sizing by considering lean, agility and sustainability factors simultaneously. In the study, responsiveness, environmental, social and economic aspects were regarded. Strategic and operational backup decisions were developed to increase the resiliency of the system against disruption of the facilities and routes simultaneously. Biçe and Batun [19], considered the problem of designing a CLSC network including the uncertainty in demand quantities, return rates, and quality of returned products. They designed the problem as a two-stage stochastic mixed-integer program to maximize the profit of the system. Boronoos et al. [20], developed a multi-objective mixed integer nonlinear programming model for a closed-loop green supply chain network design problem. In the study, the objectives were minimization of the total costs, total CO₂ emissions, and robustness costs in both forward and reverse directions, simultaneously. Chouan et al. [22], designed a multi-stage sugarcane supply chain network to process the by-products produced. In the study, three hybrid metaheuristic algorithms were proposed to handle the complexity of the problem and the performance of the algorithms was investigated using Taguchi experiments. Diabat and Jebali [30] studied the CLSC network design problem for durable products that can be disassembled into different components when they reach their end-of-life. Authors proposed models for designing the CLSC network based on various environmental legislation assumptions. The models were applied on a case study for washing machines and tumble dryers arising in Germany. Lotfi et al. [57] designed a CLSC by considering sustainability, flexibility, robustness, and risk aversion. In the study, robust counterpart model was used to handle uncertainties and a two-stage MILP model was proposed for the problem. Mosallanezhad et al. [65] presented a mathematical model for the shrimp supply chain network that optimizes the total cost of the entire network. In the study, three meta-heuristics were used to solve the shrimp supply chain network design problem composed of distribution centers, wholesalers, shrimp processing factories, markets, shrimp waste powder factory, and shrimp waste powder market. Pazhani et al. [75] developed strategic decision-making models for two different CLSCs with multiple products over multiple periods. Supply chain networks designed to assist decision making in terms of inventory, location and shipping were modeled using integer linear programming. Salehi-Amiri et al. [86] designed a CLSC network for the walnut industry by reviewing past studies. For the designed network, a MILP was developed that minimizes the overall costs. Yolmeh and Saif [109], proposed a mixed integer non-linear programming model for a CLSC network design problem integrated with assembly and disassembly line balancing under demand and return uncertainty. In the study, an enhanced decomposition approach was developed to solve the proposed model. Zahedi et al. [111] proposed a new CLSC network with sales agency and customers. The aim of the model, which had four echelons in the forward direction and five echelons in the backwards direction, was to maximize the total profit. The structure of the model was based on MILP and the proposed model was investigated through a case study regarding the manufacturing industry.

Akbari-Kasgari et al. [6], designed a copper network to reduce the effects of earthquakes on mining operations and used backup suppliers as a resilience strategy. In the study, multi-purpose models consisting of economic, environmental and social purposes, as with backup and without backup, were presented. Arabi and Gholamian [10] aimed to optimize the design of a closed-loop stone supply chain network. In the study, a multi-period multi-product mixed integer quadratic programming problem was considered. In the proposed model, two-stage stochastic programming was applied to address quality uncertainty in the mining supply chain taking into account the quality characteristics of the final stones. Kazancoglu et al. [50] aimed to present a multi-objective optimization model for a green dual-channel supply chain network that addresses economic and environmental issues. A MILP was proposed in a green dual-channel and CLSC network design. Kim and Chung [52] formulated a supply chain model to determine whether reverse logistics facilities, manufacturing centers, suppliers should relocate to their home country from the host country to maximize the total profit based on the level of reshoring drivers. The effects of productivity-adjusted costs and reverse logistics on the reshoring decisions were analyzed. Salehi- Amiri et al. [87] formulated a CLSC network for the avocado industry by developing a bi-objective model. In the study, the costs of the avocado industry and the social factor of job employment opportunities were discussed within the framework of total cost minimization and employment maximization objectives. Tavana et al. [94] designed an integrated multi-objective MILP model to design sustainable CLSC networks. In the model, cross-docking, location-inventory-routing, time window, supplier selection, order allocation, transportation modes with simultaneous pickup, and delivery under uncertainty were considered. Tirkolaee et al. [96] developed a multi-objective mathematical model to design a multi-period, multi-echelon, multi-product, sustainable mask CLSC network during the COVID-19 outbreak. In the study, a MILP model was proposed to address the locational, supply, production, distribution, collection, quarantine, recycling, reuse, and disposal decisions. Apart from the above studies, there have been many studies in the field of CLSC network design and modeling in the literature [9, 11, 28, 42, 44, 53, 56, 68, 77, 79, 88, 99, 103, 104, 106, 115, 116]. In some of the studies, exact solution algorithms were used for solving the problems [13, 28, 42, 44, 77, 79], as well as heuristics and meta-heuristics solution methodologies [11, 14, 34, 68, 96, 107], and simulation techniques [37, 72, 92, 95]. To reach comprehensive studies on CLSC network design modeling, readers may refer to the works of [7, 32, 43, 46, 91, 100].

Supply chains have a complex structure composed of many components. The most important of these components is customers, the only source of revenue for the supply chain. Since customer behavior directly influences the profitability and cost of the network, it needs to be addressed in detail in the supply chain network design. In the literature, studies of [15, 24, 66, 70, 84] considered the customer behavior in supply chain networks. Only in the study of [69], customer behavior had been considered in the reverse logistics network design problem. In the study, the authors proposed a goal-programming model to design a green supply chain network by dividing the customers into three segments according to their green expectations.

Although there are several studies dealing with the issues of ABM and supply chain together in the literature [2‐5, 12, 17, 18, 21, 25, 27, 36, 38‐40, 48, 49, 54, 55, 60, 61, 63, 64, 73, 81, 83, 89, 90, 98, 101, 105, 110, 112‐114], the use of ABM methodology in supply chain network design and modeling is much less when compared to other areas, such as scheduling, capacity planning, and production planning [3]. Furthermore, very few of these studies have handled reverse flows with ABM technique. Golinska et al. [39] discussed the fundamental problems arising from the simultaneous material flow planning in the CLSC. The authors proposed a model for the integration of CLSC through an agent-based system. Yang and Wang [104] proposed an ABM approach to overcome the difficulty of predicting reverse logistics activities due to the constraint of information transparency. In a tutorial titled “How to build a combined agent-based system dynamics model in AnyLogic”, an example was given, in which customers were defined as agents. In the example, customers' demand behaviors for products were considered agent-based [117]. Golinska [40] investigated specific tools that allow efficient management of the information accompanying the material flows in a CLSC and presented a theoretical background. The author discussed the potential for implementing agent-based solutions to improve reverse logistics management. Mishra et al. [64] proposed a multi-agent model to overcome the complexity of recycling used products and related logistics management issues. The authors handled waste classification, recycling, logistics, and reuse of products in the proposed agent architecture to assist manufacturing industries in efficient management of their green supply chains. Sauvageau and Frayret [88] proposed an agent-based simulation model to analyze the performance of various procurement and production policies in the cyclic paper industry. The authors introduced a procurement behavior model considering both the marketing price and the inventory requirements in collaboration for a large pulp maker that was also performing recycling operations in North America. Authors also presented a waste paper marketing model, imitating the market price and controlling the validity of price forecasting. Pandia et al. [73] used ABM to measure the performance of a reverse logistics company. The authors identified each member of the reverse logistics network as an agent and ensured each member measures their own performance. Since the behavior of customers has a direct impact on the profitability of the supply chain, it is necessary to consider customer behavior in network design. There are few studies that consider customer behavior in supply chain network design. Although there have been few existing studies analyzing the effect of customers’ behavior on supply chains or performing the ABM approach for managing forward or reverse logistics network design, to the best of the authors’ knowledge, there is no study that focuses on the impact of customers’ behavior on the design and modeling of CLSCs and uses the ABM approach for managing forward and reverse flows, simultaneously. In this study, to fill this gap in the literature, a generic design of dynamic CLSC network that deals with forward and reverse flows was proposed. The ABM method was used with the aims of integrating customer behaviors into the network, coping with the dynamic customer arrivals, and obtaining a more realistic structure by eliminating the assumptions of the model. Demands were classified for new and refurbished products in the proposed model. The effects of parameters, such as advertising, other users’ experiences, and market incentive threshold values on customer purchasing behavior were taken into consideration. On the other hand, it is obvious that lead time grows into a more critical factor in customers’ decision making. For example, in the UK, 65% of automobile customers pointed out the waiting time from order to delivery has a significant effect on their decisions to buy a new car; moreover, 61% of car buyers like to receive their car within 14 days or less [31, 93]. Therefore, in the case of customers not finding the product that they preferred (new product/refurbished product) in the distribution centers, the possibility of changing the preference and purchasing another type of product and resulting penalty costs were also taken into consideration.

The applicability of the proposed model was demonstrated in an example. The effects of different parameters to the network were analyzed via several scenarios. The main contributions of this research are to deal with the computational complexity associated with solving the large-scale supply chain network design problems using analytical modeling techniques as well as taking real-life conditions into account while designing CLSC networks. The novel aspects of the research can be listed as follows: (i) to develop a generic CLSC network design model focusing on customer behaviors (ii) to consider dynamic customer arrivals (iii) to eliminate required assumptions for solving such a problem via analytic methods using ABM (iv) to take into account the production, separation, and disassembly times, and economic life of the new and refurbished products (v) to involve quantity, time, and quality uncertainties of returns, and (vi) to present different scenarios according to different parameter settings to observe the behavior of the supply chain network for different conditions.

The advantages of applying green supply chain activities in terms of conservation of natural resources, energy savings, and reduction of environmental damage have increased the importance given to CLSC networks and effective management of the forward and reverse flows in these networks.

One of the most difficult issues in supply chain management is to manage the changes in customer demands. The proposed model handles the dynamic customer arrivals so as the demands in the system. In the proposed model, customer demands are obtained according to the defined customer behaviors via agent technology. Uncertainty uncertainty in the time, quality, and quantity of returns increases the complexity of the supply chains incorporating reverse flows compared to traditional supply chains and makes it more difficult to manage them effectively. The demands depend on several factors and are shaped by the customer features, behaviors, and interactions among themselves. By the proposed ABM architecture, it was aimed to design and model a generic CLSC network, considering dynamics inherent in supply chains and all actors involved in the supply chain have been identified as agents. The applicability of the proposed model is shown via scenario analyses. By the analyses, the effects of the changes in the rate of advertisement, capacities of the suppliers, market incentive threshold values, and customer behaviors are examined. In the proposed approach we also considered various parameter values as stochastic, such as returning rates of end-of-life products, qualities of returned products, and life cycles for different types of products to obtain more realistic model. The demonstration of the proposed model on an example and scenario analyses for different parameter settings showed that the model can effectively cope with the dynamic and stochastic character of the data, especially customer behaviors. In this context, the developed model presents an effective decision support system for supply chain managers in today's dynamic work environment.

The success of the ABM technique in handling the dynamic structure, individual goals and behaviors into the model is very suitable for the nature of the supply chain. The proposed model is especially helpful in managing the difficulties in modeling supply chain networks that include reverse flow, which are more complex than forward flow networks. The proposed approach provides to the companies a network management opportunity that is highly understandable and highly visual, successful in reflecting reality, and able to follow the changes instantly. The developed ABM architecture can be improved by considering transportation times, vehicle occupancy rates and storage costs as future research. The ABM technique can be used for multiple products or for competitor manufacturers. Furthermore, the preference of the customer agent in choosing the closest collection and distribution center can be modeled with the proposal mechanism and results can be compared. Today, besides the environmental and economical aspects, social issues are also getting attention as one of the main trivets of sustainable supply chain management. Therefore, proposed model can be improved for supply chain network design in which social factors are taken into account. Lastly, the proposed approach can be applied to a real-world problem.