Multi Agent Approach for Environmental Customer Collaboration: Study Case in Automotive Spare Parts Sector

GSCM can be defined as “integrating environmental thinking into Supply Chain Management including product design, material sourcing and selection, manufacturing process, delivery of the final product to the customers as well as end of life management of the product after its useful life [9]. GSCM provides many environmental practices along the product life cycle [10]. The environmental customer collaboration is a key GSCM practice, which can improve the Supply Chain performance, as customers are one of the most important stakeholders that firms depend on for their survival [11]. Furthermore, the customer collaboration influence organizations to adopt some strategies such as GSCM [1]. Indeed, many researchers identified collaboration with customers for eco-design, cleaner production, and using less energy during product transportation as an important variable [12]. Therefore, Organizations need to make interactions with customers to improve the environmental Supply Chain performance [13].

The environmental regulation and institutional pressure is another important factor, which encourages the GSCM adoption by the companies. For example, Chinese Government implemented some environmental regulations on manufacturers to promote GSCM and therefore to increase export and attract foreign investment [14]. Coskun et al. [15] combined different tax systems and analyzed their impact on the GSCM. Yu et al. [16], constructed three green manufacturing decision making models which are penalty and reward model, the non-government intervention model and the tax subsidy model, and discovered that the penalty and reward model is the best model to encourage Supply Chain partners to adopt GSCM practices. Ameknassi et al. [17] highlighted that the product recovery and the government subsidy have a positive impact on reverse Supply Chain performance.

Due to the complexity of Supply Chain network, it becomes difficult for managers and decision makers to anticipate the effects of their management policies on the Supply Chain performance. Hence, the development of new modelling approaches can be very helpful and provide great benefits for organizations [18]. Modelling approaches like analytic models and classical operational research methods can’t solve completely the inherent complexity of the Supply Chain network, such as the high number of enterprises and all the interactions that take place between them, or the uncertainty present in most of their processes [19].

In the literature conducted by Jain, Wadhwa, and Deshmukh [20], it was indicated that the use of advanced communication technology could be useful for Supply Chain collaboration and integration. Wei et al. [21] concluded that information technology based exchange relationship and information integration can improve the performance in Supply Chain.

In the past years, there has been a big interest in using simulation approach to address Supply Chain networks topics. In particular, the usage of MAS in the Supply Chain modelling [22], because of the existing similarities between the actors in the Supply Chain network and the agents in the Multi Agents Systems. In fact, MAS can take into account the different interactions between Supply chain partners who act independently based on their own policies which gives Supply Chain decision makers and managers a better understanding of the whole system [19]. Therefore, MAS becomes one of the powerful modelling tool for Supply Chain networks [23].

A multi agent system is composed of many agents taking specific roles and interacting with each other in order to solve some problems beyond their individual capacities [24].

The agents interaction has different levels and it can go from information exchange to cooperation, coordination and negotiation to manage the different activities [25]. The "agent-based modeling is the most appropriate tool for systems with high degree of location and distribution and dominated by discrete decision such as Supply Chain [26]. In fact, Multi-agent systems aim to deal Supply Chain complexity such as bullwhip effects, poor communication results, and poor coordination between members of the supply chain [27].

The approaches based on the agent modelling have been often seen as a promising way to address complex problems and heterogeneous issues in the Supply Chain such as resources allocation problems. However, MAS systems have a distributed structure, which may have some limitations in terms of the solution quality. In other words, the agents will not have a complete overview of the state of the system and also will not realize easily the impact of their individual activities [28]. Therefore, the characteristics of the agent based models can be complemented by the other optimization models such Multi Objective Linear Programming (MOLP) [29].

Optimization methods such MOLP are used in the Green Supply Optimization problems. In fact, Hugo and Pistikopoulos [30] used MOLP, which integrated Life Cycle assessment into the design and planning decisions. Many environmental factors were considered together with financial criteria to formulate the planning task. Abdallah et al. [31] presented a different MOLP in which the green procurement concept was studied. The aim was to select a supplier with minimum impact on overall carbon footprint of the supply chain in order to minimize the carbon emission costs together with traditional supply chain costs. Tognetti et al. [32], developed a multi-objective optimizations model for strategic production networks planning which integrates both emissions and Supply Chain cost taking into account the production volume allocation and the energy mix. In this research, the proposed model is formulated as MOLP.

In the GSCM, the MAS modelling has gained more attentions among researchers. Ghadimi et al. [33] provides structured information exchange process to make better sustainable decisions by maintaining a long-term relationship between a manufacturer and its suppliers. Uygun and Dede [34] highlighted the benefits distribution of GSCM partners according to their sensitivity to green products and income sharing contract. Mishra et al. [35] propose a MAS structure to solve the complexity of including waste management and recycling in the GSCM. Giret et al. [36] propose a reverse manufacturing process following a service oriented manufacturing paradigm through a virtual market supported by intelligent agents software.

To our knowledge, none of the above works considered both the customer collaboration and environmental regulation in their models. Besides, there is no research study that investigates the applicability of MAS systems in enhancing communication and information exchange specifically for the environmental customer collaboration. For this reason, many researchers highlights that the issues and requirements in the other dyadic relationship such as manufacturer-retailer in the Green Supply Chain should be further investigated [33].

The main contribution of this paper is to develop a MAS framework to improve the information exchange and collaboration process with customers to improve the financial and environmental performance taking into account the environmental regulation as well. Furthermore, the proposed framework encompasses a sub model using the Multi Objective Linear Programming MOLP to complete the agent technology, because the properties of agent based approaches and optimization techniques can complement each other.

4.1 Delivery sub model

In this section, the delivery allocation model of the developed MAS approach is explained. This mathematical model is used as the internal behavior activity of the DAA2 incorporating both financial and environmental indicators and it is based on MOLP.

In fact, total cost measure to account for all financial expenses related to the delivery of products and environmental cost is evaluated based on the environmental taxation as well as the total emission of CO₂ for a given transportation mode. In this study, we consider that these cost indicators have the same proportion of the total cost in the implementation process. Table 3 illustrates these costs elements:

Table 3. Costs elements

Since two Supply Chain performance indicators are used in the modeling approach, a multi objective optimization is implemented to construct this model as there is a few number of approaches combining financial and environmental issues in multi objective frameworks to address the supply chain design problem [37]. The workflow of this model starts with the objective function and constraints formulation, then it retrieves the input data from the data base. The outcome of this model will allow a company to determine the optimal delivery quantities to the customers, select the most appropriate transportation mode to move the products and measure CO₂ emissions related to the delivery process. Therefore, useful information will be provided to the decision maker, allowing for better analysis and creating more sustainable decisions. Figure 2 illustrates the workflow of the proposed model:

2.png

Figure 2. MAS structure

The description of the proposed model is explained in two sub-sections namely problem definition, and model formulation.

4.1.1 Problem definition

The proposed model is composed of two main echelons namely manufacturer warehouses and customer locations. Many products are considered and there are several transportation modes (i.e. road, rail, etc). below are the assumptions stated to establish this model:

• Demand of all customers is known and deterministic;

• Every customer demand is always satisfied by any manufacturer warehouse.

• Several transportation alternatives are available;

• Products are defect free;

• Distance between network nodes (used for CO₂ emissions evaluation) are assumed to be given by the straight path between facilities;

• The unit transportation costs is predefined

4.1.2 Model formulation

The formulation of the proposed model is divided into two parts, objective function and constraints. The model has two main objectives, the first one is to minimize total delivery cost which is the summation of transportation cost and storage cost, and the second is to minimize the environmental cost which is measured based on the environmental taxation and the CO₂ emissions related to the delivery process. it is assumed that the total emission of CO₂ is due to transportation. The mathematical formulation of the objective function is described in Eq. (1).

$\operatorname{Min} \sum_{i} \sum_{j} \sum_{l} \sum_{m}(d e i j l m . T C i j l m+S C i m) . Qijml+\operatorname{Min} \sum_{j} \sum_{k} \sum_{l} \sum_{m}(de jklm.TC jklm+S C j m) Q j k l m$                     

$+\operatorname{Min} \sum_{i} \sum_{j} \sum_{l} \sum_{m}(X ijl. de ijlm. Y l) \cdot Q ijlm+\operatorname{Min} \sum_{j} \sum_{k} \sum_{l} \sum_{m}(X j k l . de jklm. Y l) . Q j k l m$                     (1)

The total demand from the customer k should equal or less then delivered quantity from distributer j using transportation mode l for product m.

$\sum Q j k l m=d k m$        (2)

Total of quantity delivered by manufacturer I should equal or more than quantity delivered by distributer j to customer k using transportation mode l for product m.

$\sum Q$ ijlm $\geq \sum$ Q jklm          (3)

Total of quantity delivered by manufacturer I to distributor j should be equal or less than the transportation mode l capacity for product m.

$\sum Q$ ijlm $\leq$ C ijml                (4)

Total of quantity delivered by distributor j to customer k should be equal or less than the transportation mode l capacity for product m.

$\sum Q j k m l \leq C j k m l$                (5)

Total of quantity delivered by manufacturer I to distributor j should be equal or less than the warehouse capacity for product m.

$\sum Q i j m l \leq C A$ ijml               (6)

Total of quantity delivered by distributor j to customer k should be equal or less than the warehouse capacity for product m.

$\sum Q j k m l \leq C A j k m l$            (7)

In this current research, MATLAB has been utilized to optimize the developed delivery allocation bi-objectives mathematical model to identify the optimal delivery quantities which will be then used for the negotiation process.

4.2 Negotiation process

The negotiation is one to one bilateral negotiation. Both distributor and customer agents should negotiate the total cost provided by the delivery sub model and reach an agreement.

The distributor agent has the objective of minimizing the total cost including environmental and financial costs. It has also a maximum total costs Max TCDA that cannot be exceeded as well as a storage capacity CA that cannot be exceeded. Its dynamic knowledge consists of the negotiation of the total costs with the customer agent. This one has a fixed demand D that should be satisfied by the distributor agent. It has also the objective of minimizing the total cost TCCA. It has also to negotiate with the distributor agent the total costs.

Figure 3 describes the negotiation process between the distributor and the customer agents using exchanged messages (FIPA ACL Messages).

3.png

Figure 3. MAS structure

The parameters used in this negotiation process can be initiated differently according to the business case. In our model, we consider Begin_TCDA as the output costs of the sub delivery model including storage, transportation and environmental costs for a fixed delivered quantity.

In addition to the above-mentioned variables, we use the following function to sed a message from a source agent to destination agent: (Msg, Source, Destination).

The negotiation process is initiated by CA. it sends a call for proposal message (Msg1.CA.CFP) to DA. Therefore DA generates its first offer (Msg2) with cost equal either to Begin_TCDA or Min_TCDA

Msg1: CA. CFP

Msg2: DA.Propose (DA.first_offer)

1. If (Begin_TCDA <= Min_TCDA)

2. Then DA_offer(0) = Begin_TCDA

3. Else DA_offer(0) = Min_TCDA

4. EndIf

5. Send (DA_ offer(0) , DA , CA)

Once Msg2 is received, CA generates its first offer as well (Msg3) with a target cost equal to Max_TCCA.

Msg3 CA.Propose (CA.first_offer)

1. CA.offer(0) = Max_TCCA

2. Send (CA_offer(0) , CA , DA)

DA starts to evaluate the received first offer from CA. the first step is to check if CA_offer(k) doesn’t exceed the Begin_TCDA to make sure delivery sub model output is taken into account. On the other hand, CA_offer(k) should be higher than Min_TCDA. If these conditions are met, then the new DA_offer(k) will be equal to the proposed CA_offer(k), otherwise DA appliers a new reduction function R which is a difference between its last made offer and the new CA_offer(k). If the new DA-offer (k) including the R value is still higher than the Min_TCDA, then the R value is deducted from the last made DA_offer(k-1), otherwise DA_offer(k) will be the same as the previous one DA_offer(k-1). The result is sent to CA for evaluation (Msg4).

Msg4: DA.Propose (DA.offer)

1. If ((CA.offer(k) <= Begin_TCDA ) OR (Min_TCDA <= CA.offer(k)) )

2. Then DA_offer(k)=CA_offer(k)

3. Send (DA_offer(k), DA, CA)

4. Else R = DA_Offer(k-1) – CA_offer(k)

5. If ((DA_Offer(k) - R) <= Min_TCDA

6. Then DA.Offer(k) = DA.Offer(k-1) – R

7. Else DA.Offer(k) = DA.Offer(k-1)

8. End If

9. Send (DA_offer(k), DA, CA)

10. End If

CA evaluates the DA_offer(k) by comparing it with the last made offer. If the DA_offer(k) is less than CA_offer(k-1), CA sends an accept_proposal message (Msg5) to DA. Otherwise if the received offer DA_offer(k) is similar to previous one, then CA sends a refuse_proposal message (Msg6).

Msg5: CA.Accept_proposal ()

1. If DA.Offer(k) ≤ (CA_Offer (k-1))

2. Then Send (Accept_proposal, DA, CA)

3. Endf

Msg6: CA.Refuse_proposal (CA.offer)

1. If (DA.Offer(k) = DA.Offer (k-1) )

2. Then Send (Refuse_proposal, DA, CA)

3. EndIf

In this paper, a hybrid approach based on multi agents systems and multi objective linear programming and was presented. The objective of this approach is to make information exchange and collaboration between the Supply Chain partners more easier for the GSCM implementation. The proposed framework encompasses of the MAS structure, the agents definition and interaction, mathematical delivery allocation model based on MOLP and the negotiation process between the different agents.

The applicability of the proposed framework was demonstrated through an industrial case study in the automotive spare parts sector. First, it was concluded that the different agents of this framework managed to reach an agreement about the total delivery cost. Second, the output of the delivery sub model was considered as a final result for the negotiation process 6 times. Finally, the proposed framework helps to decrease slightly the total cost in comparison with the initial scenario without negotiation. We also concluded that an additional quantity efficiency negotiation protocol is necessary to enhance the applicability of our delivery model output which will be developed in future works.

Enhancing the proposed negotiation process by adding other constraints related to the lead time can be proposed as a perspective for future works.