The Heineken Company, known as Heineken, is a global and family-owned brewing company of Dutch heritage. With more than 300 different high quality beers and ciders and 167 breweries around the world, consumers are enjoying their products in more than 190 countries. Heineken Netherlands Supply (HNS) is a regional Operating Company (OPCO) and with three breweries in the Netherlands, HNS supplies Heineken products to their customers, including Heineken Germany (HGER). The supply chain between HNS and HGER is used as a case study for this thesis.

This supply chain is a Closed-Loop Supply Chain (CLSC) in which Returnable Packaging Materials (RPM) such as crates, are circulating through the supply chain. Heineken, including HNS and HGER, aspires to continuously eliminate inefficiencies and charge future growth through strategic investments and initiatives.

Nowadays, production volumes for the German market are growing increasingly and supply chains are under pressure since they are required to be faster, more flexible, more efficient and consumers have high expectations regarding product availability. Hence, strategic decisions regarding supply chains have become more important and featuring reliable data to measure Key Performance Indicators is essential. Therefore, Heineken is planning on introducing Radio Frequency Identification (RFID) gateways to measure actual crate cycle times, since they are currently based on assumptions.

Scope
The HKR Cluster crate will be the first RPM Stock-Keeping-Unit (SKU) which is going to be tracked through the supply chain, since the largest beer volume of the total volume for the German market (33%) are kept in this SKU. The initial idea is to install the RFID gateways and at 3 main locations:

• Brewery in Den Bosch (Netherlands)
• LCDB: Logistic center in Den Bosch (Netherlands)
• Warehouse in Werne (Germany)

For the purpose of this master thesis a scope and system boundaries have to be defined. A figure is presents the CLSC between HNS and HGER. The scope is limited to the three main locations in the Reverse Logistics (RL) flow of the CLSC. Since the HKR Cluster crate is considered to be the first RPM SKU to be tracked, this research will focus on this type of RPM.

Objective and Method
From the initiative to implement RFID gateways, it can be concluded that there is not enough RPM visibility and information transparency in the supply chain, which results in lack of integral RPM flow control. Improvement in this area can make the supply chain intelligent and more efficient. In literature, digital twins are found to be explored as a means of improving performance of physical entities. A digital twin is a virtual representation of a real entity and the concept has gained much interest over the years. The world of supply chain and logistics is lagging behind when it comes to adapting digital possibilities to current conditions. Therefore, the objective of this master thesis is to enable supply chain control of RPM flows using data provided by the RFID gateways, from a Digital Twin design perspective. The research is driven by the ambition and visions for digital transformation in supply chains.

To obtain the research objective the following main research question is defined: How can real-time control in the reversed supply chain be enabled, with use of RFID data?. To answer the main research question multiple sub research questions are defined and are used as guidance through the thesis. A current state analysis will help to understand how the supply chain currently operates and performs. Based on this analysis a Digital Supply Chain Twin (D-SC-T) framework for the current supply chain is proposed. Then, a mathematical model for control is proposed and simulations are done in MATLAB. The impact of control will be assessed and evaluated by comparison of financial Key Performance Indicators (KPI's) in the current state (no control) with the future state financial KPI's (control).

Current state analysis
Before systems are modelled or designed, a current system states analysis is performed to determine how the current supply chain operates. The analysis confirms the earlier found inefficiencies in the current state. The cycle times are based on assumptions and approximate to be 25 weeks, which is considered to be high. This is because the average time spent at the locations is high due to large inventories. There are large safety stocks to avoid out of stock situations, while there is limited storage capacity. It is very common that at the inventory locations, LCDB and warehouse, the storage space is at full capacity and an external storage location has to be rented. Furthermore, the planning for production at the brewery is made a relatively long time in advance and therefore lacks flexibility. It can be concluded that there is little RPM visibility throughout the supply chain and data availability for planning departments. Change in demand, weather and events can cause inaccurate forecasting. In conclusion, there is no centralized control of inventory levels in the current supply chain.

Heineken’s reversed supply chain, driven by returnable packaging, is defined to be a push-based supply chain. Crates are pushed through the channel from the location where it is returned by the customer up to the brewery. Every supply chain agent has its own priorities and inventory management preferences. This can lead to unnecessary inventory costs.

Design
Digital twins are found to be explored by means of improving performance of physical entities by using models combined with various data to interpret and to predict the behavior of a real system. Therefore, digital twins have the potential to increase the intelligence of a specific environment. This leads to the motivation of digital twins of supply chains. First steps towards D-SC-T creation are done by proposing a framework according to its functions and requirements.

The prediction function includes analysis of the behaviour of the supply chain before actual run-time. Planned crate flow processes are simulated prior to the actual transportation decisions in the supply chain system (pro-active planning). Consequently, supply chain parameters can be tested, while potential impacts on the supply chain performance can be evaluated. The monitoring function enables optimization when models are enriched with real-time data from physical sources, such as RFID. Therefore, the RFID gateways allow tracking and supervision of the current states of crate flow and inventory at the main locations. The RFID data, including live positional data from the crates, can be fed into the digital twin. If the current state measurements deviate from the preferred state, transportation decisions in the supply chain system can be calculated (re-active planning). Supply chain performance and behaviour diagnosis is usually enabled after an event and is done by data analysis.

Model Predictive Control (MPC) is a control strategy, by means of controlling a process based on some form of model. Literature shows that the digital twin and MPC have similarities in the way they capture and interpret the current state of the physical system and being able to use that current state to change the future state. Therefore, MPC seems a very suitable option for control of the inventory levels of the supply chain within the digital twin framework. A centralized MPC control model for the control of inventory levels and crate flows within Heineken's reversed supply chain is proposed.

The described MPC control model has the objective to optimize the supply chain performance by reducing Operating Expenditures (OPEX) and Capital Expenditures (CAPEX). The controller will accurately keep track of where crates are located in the supply chain and calculate the related OPEX and CAPEX, while meeting the requirements.

Results
Simulation experiments are done to be able to quantify the impact of control on the supply chain in OPEX and CAPEX. In the experiments, the controller reacts to disturbances and unforeseen events, while optimizing inventory levels and meeting demand. This is demonstrated by 3 scenarios.

• Scenario 1: Current supply chain with actual events LCDB: Logistic center in Den Bosch (Netherlands)
• Scenario 2: Current supply chain with disturbance: peak in demand
• Scenario 3: Supply chain with additional RFID gateway location with disturbance: capacity limitation

A Table presents the simulation results for all scenarios and are presented including the results for the same scenarios with no control. When comparing the results of current supply chain with actual events, the most remarkable result is the difference in the location where inventory is allocated. In the base case scenario, the crate inventory levels are much higher at the LCDB, while with MPC control, the results show higher inventory levels at the warehouse. These more detailed results are shown in Chapter \ref{chap:Simulation and Results}. In the second scenario the simulated event is an unforeseen high beer demand due to weather changes. The controller reacts to the occurring event and meets the demand at the brewery in time. In the third scenario the effect of having more supply chain information by including one additional RFID gateway location is determined. More detailed results and explanations on how the controller reacts to various events are provided in Chapter \ref{chap:Simulation and Results}.


Conclusion, Discussion and Recommendations
This thesis has created insights on what a digital supply chain is and what the effect of control can be on the supply chain performance. The controllability of the crate flows in Heineken’s reversed supply chain driven by returnable packaging can be improved, using a centralized MPC control model within the proposed digital twin framework, combined with RFID data from the proposed gateways. These gateways measure crate positions and quantities per time. The controller uses this data to interpret and predict the supply chain behaviour. The RFID data of the current supply chain states are fed into the model. The controller interprets the states and calculates which actions lead to less CAPEX and OPEX, while meeting the modeling requirements. Due to the RFID measurements, data is visible and transparent for planning departments and other stakeholders and better coordination along supply chain agents can be made possible.

For this research MPC is the chosen control method for the control part within the digital twin. Therefore it only covers a small part of the wide variety of different control methods which could have been investigated and tested. Other control methods are still to be investigated.

This thesis offers a theoretical digital twin solution for the problems they have at Heineken’s CLSC. But only a solution for the measure and control part has been brought forward. A digital twin also carries out big data analytics and machine learning possibilities. How these growing technologies fit into the digital twin concept could be interesting for further research.