Exploring Real-Time Transportation Planning in a Dynamic and Multi-Objective Supply Chain: an Intelligent Transportation System

(1)

Exploring Real-Time Transportation Planning in a

Dynamic and Multi-Objective Supply Chain: an

Intelligent Transportation System

by

Lars Kuperus (S2973626)

l.kuperus@student.rug.nl

University of Groningen

Faculty of Economics and Business (FEB)

MSc Technology & Operations Management (TOM)

MSc Supply Chain Management (SCM)

Supervisors:

1st: dr. J.A.C. Bokhorst

2nd: dr. ir. D.J. van der Zee

(2)

1

Abstract

With the current advances in Information and Communication Technologies, it is possible to plan and control the operation of a fleet of vehicles in real-time. However, theory on the application of this real-time control mechanism is deficient. This includes both the highly dynamic environments, where multiple real-life characteristics are included, and the multi-objective environments, where sustainability is a growing concern. This study intends to fill this gap, by developing and evaluating a real-time planning system in a dynamic and multi-objective internal supply chain, based on the current developments in Intelligent Transportation Systems. By means of a Design Science approach, the system is developed through literature reviews and a case study at a starch manufacturer in the Netherlands, after which different configurations are evaluated in a simulation study. The new system constitutes of real-time assignment and redeployment, with the number of trucks, queue discipline, assignment rule, repositioning policy and diversion as the core building blocks. The simulation study indicates that the empty travel distances (environmental measure) is lowered with less trucks in the system, configuring the queue priority to the production rate discipline, using the assignment rule of shortest travelling distance and making use of the wait policy after unloading. The number of stock outs (financial measure) is lower when there are more trucks available in the system and the queue discipline is based on the real-time silo inventory levels. Assignment rules and repositioning are only of minor importance for lowering the total number of stock outs. However, some of these main effects found should be interpreted in the light of interaction effects with other elements in the system. Diversion was found to have no significant effect on the empty travel distances and number of stock outs in the network. So, both the design and the evaluation provide a guidance for practitioners to create and get insights in the building blocks of the future-proof real-time planning systems, by taking both the financial and environmental impact into consideration.

Keywords: vehicle routing problem; real-time; multi-objective; dynamic; intelligent

(3)

2

List of Tables

Table 2.1: Overview of Design Science stages and research methods ... 16

Table 3.1: Overview of native starch flows during the inter-campaign period ... 20

Table 3.2: Current amount of trucks per route ... 21

Table 4.1: Relevant literature on Dynamic Pick-up and Delivery Problems ... 30

Table 6.1: Unloading silos’ characteristics ... 41

Table 6.2: Descriptive statistics of experimental factors ... 45

Table 6.3: Factorial ANOVA results (main effects) ... 46

Table 6.4: Factorial ANOVA results (interaction effects) ... 48

Table B.1: Model scope ... 81

Table B.2: Model detail ... 83

Table E.1: Loading silo frequencies ... 87

Table F.1: Distribution parameters and Goodness-of-Fit Test (loading rates at Hollandia) .... 88

Table F.2: Distribution parameters and Goodness-of-Fit Test (loading rates at EURO) ... 89

Table F.3: Distribution parameters and Goodness-of-Fit Test (loading rates at AAA) ... 90

Table F.4: Distribution parameters and Goodness-of-Fit Test (loading rates at OKO) ... 91

Table F.5: Distribution parameters and Goodness-of-Fit Test (loading rates at WTM) ... 92

Table F.6: Distribution parameters and Goodness-of-Fit Test (loading rates at P&O) ... 93

Table F.7: Distribution parameters and Goodness-of-Fit Test (unloading rates at GNV) ... 94

Table F.8: Distribution parameters and Goodness-of-Fit Test (unloading rates at FOX63) .... 95

Table F.9: Distribution parameters and Goodness-of-Fit Test (unloading rates at FOX7) ... 96

Table F.10: Distribution parameters and Goodness-of-Fit Test (unloading rates at FNEXT) . 97 Table F.11: Distribution parameters and Goodness-of-Fit Test (unloading rates at VMF) ... 98

Table F.12: Distribution parameters and Goodness-of-Fit Test (unloading rates at DexTAK)99 Table G.1: Statistics of production rates and production stops ... 100

(7)

6 Table I.1: Overview of experiments and results ... 105

(8)

7

List of Figures

Figure 3.1: Waiting times and occurrence during loading and unloading ... 24

Figure 3.2: Breakdown of truck lead times ... 24

Figure 5.1: Planning procedure of the Intelligent Real-Time Transportation System ... 36

Figure 6.1: Interaction between number of trucks and queue discipline (empty distance) ... 50

Figure 6.2: Interaction between number of trucks and assignment rule (empty distance) ... 51

Figure 6.3: Interaction between number of trucks and repositioning policy (empty distance) 52 Figure 6.4: Interaction between queue discipline and assignment rule (empty distance) ... 52

Figure 6.5: Interaction between queue discipline and repositioning policy (empty distance) 53 Figure 6.6: Interaction between assignment rule and repositioning policy (empty distance) . 54 Figure 6.7: Interaction between number of trucks and queue discipline (stock outs) ... 56

Figure 6.8: Interaction between number of trucks and assignment rule (stock outs) ... 57

Figure 6.9: Interaction between queue discipline and repositioning policy (stock outs) ... 57

Figure 6.10: Interaction between assignment rule and repositioning policy (stock outs) ... 58

Figure A.1: Complete Planning Pathway ... 75

Figure B.1: Sequence of steps proposed for simulation ... 77

Figure C.1: Geographical locations of the internal supply chain ... 85

Figure D.1: Overview of native starch flows during the inter-campaign period ... 86

Figure F.1: Comparison of observed and expected loading rates at Hollandia ... 88

Figure F.2: Comparison of observed and expected loading rates at EURO... 89

Figure F.3: Comparison of observed and expected loading rates at AAA ... 90

Figure F.4: Comparison of observed and expected loading rates at OKO ... 91

Figure F.5: Comparison of observed and expected loading rates at WTM ... 92

Figure F.6: Comparison of observed and expected loading rates at P&O ... 93

(9)

8

Figure F.8: Comparison of observed and expected unloading rates at FOX63 ... 95

Figure F.9: Comparison of observed and expected unloading rates at FOX7 ... 96

Figure F.10: Comparison of observed and expected unloading rates at FNEXT... 97

Figure F.11: Comparison of observed and expected unloading rates at VMF ... 98

Figure F.12: Comparison of observed and expected unloading rates at DexTAK ... 99

Figure H.1: Utilisation Time-Series ... 101

Figure K.1: Simulation model inputs and methods ... 109

Figure K.2: Flow within the simulation model (core) ... 109

(10)

9

List of abbreviations

GPS Global Positioning System

GIS Geographic Information System

ICT Information and Communication Technology

ITS Intelligent Transportation System

VRP Vehicle Routing Problem

DVRP Dynamic Vehicle Routing Problem

CVO Commercial Vehicle Operations

AFMS Advanced Fleet Management Systems

PDP Pick-up and Delivery Problems

D-VRPPD Dynamic Vehicle Routing Problem with Pick-up and Deliveries

D-SCP Dynamic Stacker Crane Problem

D-DARP Dynamic Dial-a-Ride Problem

LTL Less-Than-Full Truckloads

SLPDP Single Load Pick-up and Delivery Problem

TPDP Truckload Pick-up and Delivery Problem

DPDFL Dynamic Pick-up and Delivery with Full Truckload

LP Linear Programming

IP Integer Programming

MIP Mixed Integer Programming

NLF Nearest Load First

STDF Shortest Travel Distance First

FCFS First Come First Serve

LSF Lowest Stock First

ALI Advanced Load Information

KW Knowledge Window

MPS Modified Potato Starch

NPS Native Potato Starch

AGV Automated Guided Vehicle

MTB Mean Time Between

MTT Mean Time To

ANOVA Analysis of Variance

HSD Honest Significant Difference

SD Standard Deviation

M Mean

SS Sum of Squares

MS Mean Square

(11)

(12)

11

1. Introduction

In the last two decades, the role of technological advances and sustainability has been one of the major interests in the supply chain management field (Stevens & Johnson, 2016). Logistics and transportation, as part of the broader supply chain management concept, has become an essential business function and is also subject to the increasing importance of technology and sustainability integration (Speranza, 2018). Improved schedules are important because of their positive effect on the supply chain efficiency and the total cost encountered by the company (Tamannaei & Rasti-Barzoki, 2019). There is a growing interest in the academic literature towards the vehicle scheduling and routing problems, because of difficulties in managing and scheduling transportation vehicles in the current complex and fast changing environment (Regnier-Coudert, Mccall, Ayodele, & Anderson, 2016). The rapid and fast changing information flows in the supply chain, with varying information about demands, inventories and deliveries make transportation and logistical problems more dynamic in nature (Speranza, 2018). Together with the increasing pressure from different stakeholders to evaluate and improve sustainability throughout the supply chain (Papageorgiou, 2009), it is expected that the transportation and logistics sector will change significantly (Speranza, 2018).

(13)

12 system in a simulation model, the following objective is emulated: the development and evaluation of a real-time planning system, based on the developments in Intelligent Transportation Systems.

This purpose is pursued through a Design Science approach, where design alternatives (different parameter configurations) are constructed based on a literature review and a diagnosis of the current situation, which are then evaluated in a simulation model to discover the potential contributing effect of real-time planning on multiple objectives in a dynamic supply chain. The problem and improvement case covers the internal supply chain of a starch manufacturer in the north of the Netherlands. This internal supply chain comprises multiple silos and production facilities, where bulk transport is performed by a fleet of homogeneous trucks hired from a third-party logistics provider. This supply chain is subject to dynamic demand requests (production rates), service times and disruptions at production facilities.

(14)

13

2. Research objectives and method

This chapter sheds light on the objectives of this research, together with the methods used to attain those objectives and answer the corresponding research questions. First, a background is provided with the first introduction of the problem and the corresponding literature on real-time planning in dynamic environments and the presence of multi-objective models. From that starting point, the research objectives are formulated and translated into the research questions of this study. Last, the research design is described, displaying the contents of the design science research method used in this paper.

2.1 Background

Operating a fleet of vehicles is one of the key activities in many contexts, like the simple pickup and delivery problems, and the relocation of trucks for third-party logistics providers (Pillac, Guéret, & Medaglia, 2012). This fleet management has the aim to move loads within a network of nodes as efficient as possible, regarding the operational constraints (Jaoua, Riopel, & Gamache, 2012). With the growing need for more dynamic and realistic models and controls in transportation (Jaoua, Gamache, & Riopel, 2012), the complexities of these systems are becoming substantial (Braekers, Ramaekers, & Van Nieuwenhuyse, 2016). These real-life characteristics include the dynamic arrival of demand information, travel and service times, and the availability of resources in the network (Braekers et al., 2016). Comparing these real-life attributes with the current models on fleet management revealed gaps in applicability of these academic models: there is a need for more dynamic and uncertainty incorporating models that are effective and applicable to the real-life situation of vehicle routing and scheduling (Braekers et al., 2016; Crainic, Gendreau, & Potvin, 2009; Lahyani, Khemakhem, & Semet, 2015).

(15)

14 A main technology and concept regarding this trend towards the use of information technologies in transportation is the development of Intelligent Transportation Systems (ITS), which will be the key focus of this particular research. The core of an Intelligent Transport System consists of “obtaining, processing, and distributing information for better use of the transportation system, infrastructure and services” (Crainic et al., 2009, p. 543). This class of Intelligent Transport Systems creates new challenges and opportunities for operational researchers, especially the development of new decision and optimisation systems for fleet management, relying on the hardware introduced in recent years (Pillac et al., 2012).

Next to the trends mentioned above, there is growing interest and pressure to increase the sustainability in the supply chain (Hassini, Surti, & Searcy, 2012). Transportation is responsible for 14% of the total CO2 emissions and there is a maturing concern from the society towards

this carbon footprint (Dekker, Bloemhof, & Mallidis, 2012), making sustainability one of the major concerns in logistics management (Zhu, Sarkis, & Lai, 2008). The intended shift towards the green supply chain is not in line with the models on transportation planning and scheduling, where the performance of the solution is measured solely on costs or profits (Dekker et al., 2012; Liu & Papageorgiou, 2013). Therefore, there is a need for the development of multi-objective models, that take a broader look at the supply chain planning performance than the narrow cost perspective (Papageorgiou, 2009).

2.2 Research objectives

A real-time planning system, based on an Intelligent Transportation System, could have a positive influence on both the economic and sustainability perspective within the supply chain. Developing a real-time intelligent transport system for the dynamic and multi-objective supply chains is therefore the core element of this study, leveraging the information exchange potential in the supply chain through the use of an Intelligent Transportation System that enables real-time planning in a highly dynamic environment.

The research objectives are therefore two folded, first, alternative real-time planning models (different parameter settings) are developed for the transport planning in a dynamic and multi-objective environment, relying on an Intelligent Transportation System and second, these alternative planning models are tested and analysed through a simulation study. To realise the objectives identified above, the following research questions are formulated:

(16)

15 2. What are the relevant elements in designing a real-time planning system for a dynamic

problem?

3. How should the parameters be set in the design of alternative real-time planning models based on Intelligent Transportation Systems?

4. What is the difference in performance between the different planning configurations? 5. How can the differences between the alternatives be explained?

2.3 Research method

The research method used in this paper concerns the Design Science approach, originating from the engineering and artificial sciences (Simon, 1996). This Design Science approach is in the core a problem-solving method, focused on the creation and evaluation of artefacts, like the development and validation of information systems (Hevner, March, Park, & Ram, 2004). This makes Design Science research different from the theory-building and –testing methods, which are not focused on the exploration of the problem and solution through a new design, but on the building and testing of theories by empirical research methods (Holmström, Ketokivi, & Hameri, 2009). The Design Science methodology adopted in this paper, covering the steps of a Design Science research, are derived from the methodology described by Wieringa (2014). This Design Science cycle comprises the stages that match and apply to Design Science research in information systems and software engineering (Wieringa, 2014), which corresponds with the purpose of this study: developing a real-time intelligent transport system based on information exchange.

(17)

16 the artefact is transferred to the real-world and evaluated in the actual real-world setting (Wieringa, 2014), is not part of this research and therefore not within the scope of the goals, methods and results of this study. The research questions, as established in section 2.2, are linked to those steps mentioned above in a way that each question is covered in one of the following sections. This link will be further described in the next subsections (2.3.1-2.3.4), together with the specific research method that is used in the upcoming chapters. An overview of the stages of Wieringa (2014), linked to the chapters and research methods used in this paper, is presented in Table 2.1.

The research problem at hand is investigated at a starch manufacturer in the north of the Netherlands, characterised by dynamic aspects in demand, waiting times, (un)loading times and disruptions. The research focuses on the internal supply chain, covering six silos and three factories, all located in the northern part of the Netherlands. The transportation of starch is performed by bulk trucks, hired from a third-party logistics provider that is located close to the starch manufacturer. The planning and the required number of trucks is, respectively, performed and determined by the logistic service provider, making the company an important partner in the supply chain and internal processes of the company.

Design Science stage Chapter(s) Research method

Problem investigation 3 Interviews and data

observations (case)

Treatment design 4, 5 Literature review and

interviews (case)

Treatment validation 6, 7 Simulation

Table 2.1: Overview of Design Science stages and research methods

2.3.1 Analysing current planning model

(18)

17 together with the ability to cover the contextual conditions (Yin, 2014). Therefore, the analysis of the current planning model is carried out by interviews (field experts) and observations (data) during a case study, to be able to answer the question of how the current planning model is organized and how it is performing. Next to that, the context is an important element to consider regarding the current system, as the new artefact (planning system) is expected to be applicable and perform in that same context. As mentioned, by means of this chapter (3), the first research question ‘How is the transportation planning currently organized and performing?’ will be answered.

2.3.2 Real time and dynamic planning models

Chapter 4 provides an overview of dynamic, real-time and Intelligent Transport Systems, based on available literature. Therefore, the method used in this chapter is a literature study on the relevant concepts and systems. This chapter starts with an introduction of the dynamic supply chains and systems, which is the context of the study. Next, an overview of real-time planning developments is presented, to provide a state-of-the-art picture of the knowledge on real-time planning in Pick-up and Delivery Problems. The last part covers the Intelligent Transport Systems and the corresponding information exchange in the supply chain. Next to the interest for the core of real time and intelligent transport planning models, there is also a section that covers the current state of multi-objective models in transport planning. Based on the literature provided in this chapter, the second research question ‘What are the relevant elements in designing a real-time planning system for a dynamic problem?’ will be answered.

2.3.3 Design of alternative planning models

(19)

18 parameters be set in the design of alternative real-time planning models based on Intelligent Transportation Systems?’.

2.3.4 Evaluation of alternative planning models

(20)

19

3. Current system details

This chapter describes the current planning system of the starch manufacturer in the Netherlands, covering a description of the current planning characteristics and process, together with an insight in the performance of the current system. First, the characteristics of the company are described, entailing the production process and layout of the internal supply chain. The next section highlights the current planning process in the internal supply chain, encompassing a description of the current steps taken to plan the pool of trucks. Next, the coordination aspect is highlighted, together with a comparison between day and night planning. The last section covers the performance of the current planning and truck allocation process.

3.1 Production process and internal network

The company of interest for this research is a large starch manufacturer in the Netherlands, concentrated in the northern part of the Netherlands and a couple of factories in Germany and Sweden. This research focuses on the three factories in the Netherlands, that rely on the bulk transport performed by a third-party logistics provider. Two of those factories (GNV and TAK) convert the potato into starch and protein, which only takes place during the campaign period (September - February). The third factory (FHL) only produces the modified potato starch (MPS), which is a production process that is in need of the native potato starch (NPS). All three factories are characterised by a continuous process, where the factory runs for 24 hours per day. The starch is stored in internal and external silos, respectively located at the production plants or at different locations in the northern parts of the Netherlands. During the inter-campaign period, the starch stored in the silos is used for the production lines, to secure the input of starch for the derivate lines (all three production locations) and direct customer delivery (NPS). For the campaign period, the pool of trucks for internal bulk transport has to transport starch from GNV and TAK to both the silos and the derivate lines. Within the inter-campaign period, the transport only comprises the transport from silos to derivate lines.

(21)

20 (unloading silos) is depicted in Table 3.1, a schematic overview of the locations and flows is provided in respectively Appendix C (Figure C.1) and Appendix D (Figure D.1). Except for the silo at TAK (WTM), all other loading silos are located external to the production locations and are therefore classified as external silos. The strict combination of silos and production locations exists because of the fact that there are three main types of native potato starch: food, semi-food and non-food. These types are stored in different silos and used for the different production lines, which highlights the structure of flows between the silos and the production locations. This classification is made because of the characteristics needed in the different products (industry or food focused), which are the input materials for the final customer.

Loading silo Loading silo location Unloading silo Unloading silo

location

Hollandia Nieuw Buinen GNV silo GNV

Euro Veendam GNV silo GNV

AAA Alteveer FOX63 / FOX7 FHL

OKO De Krim FOX63 / FOX7 FHL

WTM Ter Apelkanaal FNEXT / FOX63 FHL

WTM Ter Apelkanaal DexTAK TAK

OKO De Krim DexTAK / VMF TAK

P & O Nieuwe Pekela VMF TAK

Hollandia Nieuw Buinen VMF TAK

(22)

21

3.2 Planning process

This part of chapter 3 describes the current planning process, along with the current planning difficulties and the coordination that takes place in the internal bulk transport network. As previously mentioned, the trucks are hired from a third-party logistics provider, who is also responsible for the planning of the trucks. This pool of trucks, which is a constant amount of resources, is dedicated to the bulk transport of the client. The planners of the third-party logistics provider schedule a certain, fixed, amount of trucks on specific routes/flows. The amount of trucks on each route or a combination of routes is determined by the experience and beliefs of the planners. The amount of trucks on each route, or a combination of routes, is provided in Table 3.2 (for a total of nine trucks). In this current situation, the trucks keep driving between the pick-up and delivery location of their predefined routes. Concurrently, the planners keep an eye on the production rates and buffer silo levels, to make a choice between the predefined routes for trucks. This is only the case when the trucks are assigned to more than one route, otherwise the trucks become unutilised. When a truck is not needed on the route where it is assigned to, the third-party logistics provider tries to find another purpose or short-dated client for this truck, outside the internal network for this manufacturer. The entire planning and assignment process is therefore based on experience and fixed routes, with minor adjustments by the real-time observations of the starch levels in buffer silos and production rates. So, diversion, smart repositioning, transport prioritisation and assignment rules (different than fixed route assignment) are not in place, leaving a large potential for efficiency improvements untouched.

Route(s) Number of trucks

EURO / HOL - GNV 4

AAA - FHL 1

WTM / P & O – TAK + WTM - FHL 1

OKO – TAK / FHL 3

(23)

22 An important element to mention, in light of the truck planning and coordination, is the difference between day and night operations. During daytime, the planners are in contact with the truck drivers, to be able to allocate the truck to another client or make small adjustments in the routes assigned to the truck. During night time, the planners are not available to be contacted by the truck drivers, which creates the situation where all truck drivers stick to their route(s). One truck driver within the pool of trucks is assigned to be a coordinator, answering demand requests from the factory and assignment questions from the other drivers in the pool. Again, these decisions are based on the intuition and beliefs of the coordinator, without a clear foundation in the data or predefined rules. This approach results in difficulties and performance issues, which are described in the next section of this chapter.

3.3 Performance and difficulties

This last section of chapter 3 describes the difficulties in managing the internal supply network, under the current scheduling approach. Next to that, the performance of the current situation is depicted, based on the available data and the interviews conducted at both the manufacturer and the third-party logistics provider. These performance issues are linked to the difficulties and drawbacks of the current system.

(24)

23 The second drawback in the current system relates to real-time information sharing and decision making. Within the current system, the planner is responsible for the monitoring of the silo levels at the production locations, resulting in manual and intuition based decisions. This monitoring is difficult because of the fact that for some of the production locations, like the GNV, real-time silo monitoring is not available. Here, transport orders are provided by the production locations, through the use of personal messages for the planners at the third party logistics provider. This makes it difficult to get a clear and comprehensive picture of the entire network, enhancing the difficulty to make accurate real-time decisions. Based on that information and communication difficulty in the current situation, more trucks are needed to deal with the variability within the internal network and trucks are assigned to specific routes. Therefore, this lack of information exchange requires a larger number of trucks within the pool and the policy that fixes trucks to certain routes.

The drawback introduced above is even more highlighted during night time, where the planner is not available for real-time decision making. As mentioned before, a truck driver is assigned to be the coordinator, keeping track of real-time requests and the allocation of trucks. Again, this is based on the intuition and beliefs of the coordinator, which lacks the data driven and comprehensive foundation of an overview of the entire pool of trucks. Because of the absence of a planner with accurate and comprehensive information, trucks are mainly fixed to their less demanding route. Next, the lack of a comprehensive overview and central communication leads to increased waiting times, because of an inefficient alignment between trucks and demand for starch.

(25)

24

Figure 3.2: Breakdown of truck lead times

0 20 40 60 80 100 120 140 160 180 GNV TAK FHL Tim e (m in u te s) Production location Loading Unloading 0% 10% 20% 30% 40% 50% 60% GNV TAK FHL O cc u rr en ce o f w a iti n g (p er ce n ta g e) Production location Loading Unloading

(26)

25

4. Organizing real time transport planning

This chapter deals with the context of an Intelligent Transport System, that is capable of planning a fleet of vehicles in real-time in a multi-objective and dynamic supply chain. First, the shift from static transportation to dynamic transportation problems is described, to shed a light on the characteristics and difficulties of a dynamic environment. This is followed by an overview of the available literature on real-time planning and the relevant literature on a special class of routing problems, the Dynamic Pick-up and Delivery Problems. The last subchapter covers the background of Intelligent Transportation Systems, creating an introduction to the Intelligent Real-Time Planning System.

4.1 Rich transportation problems

One of the most challenging, complex and central components of modern fleet management is the Vehicle Routing Problem (VRP) (Golden, Raghavan, & Wasil, 2008). The original VRP formulation, which is a generalisation of the Travelling Salesman Problem, was initially presented by Dantzig & Ramser (1959). This VRP has the aim to construct an optimal route plan, by determining a set of vehicle routes to fulfil customer demand, while the costs and distance travelled are minimised (Drexl, 2012a). The problem class studied in this paper is the Pick-up and Delivery Problems, which is discussed and linked to intelligent real-time planning systems in sections 4.2 and 4.3. Where most of the VRP problems are initiated and motivated through theoretical problems, there is an ongoing interest towards new variants that include complex constraints and goals (Lahyani et al., 2015). This interest originates from the aroused concern about real-life complexities in vehicle routing (Lahyani et al., 2015) and the continuous input from the sector of transportation and logistics (Drexl, 2012b). This class of extended, real-life Vehicle Routing Problems, is often referred to as Rich Vehicle Routing Problems (Lahyani et al., 2015).

(27)

26 (Psaraftis, 1980). Where the first one considers information changes during the planning process (e.g. new order arrivals), the latter one is concerned with the uncertainty in the availability of the information (e.g. distribution of order arrival times) (Pillac et al., 2013). The most prevalent dynamic events are linked to these uncertainties in information evolution and availability, namely: customer demands, service times and waiting times (Hasle & Kloster, 2007; Pillac et al., 2013; Ritzinger et al., 2016).

This Dynamic Vehicle Routing Problem (DVRP), where information is managed when it enters the system, increased interest in last decades (Fleischmann, Gnutzmann, & Sandvoß, 2004; Pillac et al., 2013; Ritzinger et al., 2016). The DVRP was first mentioned by Wilson & Colvin (1977), stating a Dial-A-Ride Problem that is controlled by a computer. This report is followed by many other papers, with an impressive evolution in the last twenty years, stimulated by the advances in Information and Communication Technologies (ICT) (Psaraftis, Wen, & Kontovas, 2016). For the definition of dynamic and stochastic problems, the taxonomy of Pillac et al. (2013) is used. This classification is based on the information dimensions of Psaraftis (1980), namely the formerly introduced information evolution and information quality. This means that dynamic and stochastic problems are characterised by (partly) unknown input values, which are revealed over time, according to stochastic knowledge that is available in the system (Pillac et al., 2013). This requires an ongoing adaption of vehicle routes and allocations, together with the necessary support of certain technologies, to have real-time communication between the vehicles and the dispatcher (Pillac et al., 2013). Involving this dynamic information aspect in the problem increases the complexity, introducing new challenges to plan the fleet of vehicles in an efficient manner (Pillac et al., 2013; Ritzinger et al., 2016).

(28)

27 stochastic inputs (Braekers et al., 2016; Ritzinger et al., 2016). These models should be extended with more stochastic driven events and inputs, together with efficient solution methods for these enriched problems (Braekers et al., 2016; Ritzinger et al., 2016).

The original VRP, together with some simple extensions, is a classical optimisation problem, with efficient algorithms available to solve the problem (Baita, Pesenti, Ukovich, & Favaretto, 2000). However, more dynamic and real-life models are in need of other solution approaches because of the difficulties and complexities that are not included in the more static approaches (Baita et al., 2000; Crainic et al., 2009; Lahyani et al., 2015). As a result, more practical focused test platforms should be designed and developed, together with the execution of simulation experiments, to test the solution approaches in Dynamic Vehicle Routing Problems (Regnier-Coudert et al., 2016).

4.2 Dynamic Pick-up and Delivery Problems

(29)

28 Similar to the general and traditional VRP, the Pick-up and Delivery Problems are also researched from the perspective of dynamic and real-life characteristics (Berbeglia et al., 2010). Despite the fact that most of the studies on PDP focused on static models, it is possible to divide the available literature on dynamic one-to-one problems in three categories: the Dynamic Vehicle Routing Problem with Pick-up and Deliveries (D-VRPPD), Dynamic Stacker Crane Problem (D-SCP) and the Dynamic Dial-a-Ride Problem (D-DARP) (Berbeglia et al., 2010). The former (D-VRPPD), is the general model, where the vehicle is capable of handling more than one request at the time (Regnier-Coudert et al., 2016). Most of the available literature focuses on this problem with Less-Than-Full Truckloads (LTL) (Crama & Pironet, 2019). When the transportation entities are humans, the D-VRPDP is classified as a Dynamic Dial-a-Ride Problem (D-DARP) (Regnier-Coudert et al., 2016). The problems in which a vehicle can only carry one load and therefore perform one request at the time, are called the Stacker Crane Problems (SCP) (Berbeglia et al., 2010). This name refers to the practical problem of loading and unloading entities by crane operations (Berbeglia et al., 2007). In the literature also referred to as the Single Load Pick-up and Delivery Problem (SLPDP), the Truckload Pick-up and Delivery Problem (TPDP) (Fleischmann et al., 2004), and the Dynamic Pick-up and Delivery with Full Truckload (DPDFL) (Zolfagharinia & Haughton, 2017). The main application for this type of problem is the operation of a fleet of trucks, that transports full truckloads between pick-up and delivery locations (Berbeglia et al., 2010). This represents the problem that is researched in this study. Therefore, the next sections of the theoretical background focus on this specific class of Pick-up and Delivery Problems.

(30)

29 (2014). Both studies focused on the planning and operation of a fleet of vehicles, where Tjokroamidjojo et al. (2006) minimised the costs through Integer Programming (IP) and Zolfagharinia & Haughton (2014) maximised profits by developing and solving a Mixed Integer Programming (MIP). A review on the studies stated above and on the literature summaries in Zolfagharinia & Haughton (2016, 2017), reveals that most of all relevant literature on the topic of D-SCP considers only one objective and only one stochastic and dynamic aspect. Unless some articles that only take into account the responsiveness or total travel time, all other studies are focused on the optimisation of costs or profits (Zolfagharinia & Haughton, 2016, 2017).

This is in line with the need for multi-objective models in transportations and logistics planning, including the demand for the integration of sustainability in planning models (Dekker et al., 2012; Montoya-Torres et al., 2015; Psaraftis et al., 2016). Next to that, most of the relevant literature on Dynamic Pick-up and Delivery Problems only incorporates one stochastic and dynamic input variable, mainly the customer or transportation request (Berbeglia et al., 2010; Zolfagharinia & Haughton, 2016, 2017). This is consistent with the observations from Braekers et al. (2016) and Ritzinger et al. (2016), highlighting the need for transportation planning models and systems that incorporate more than one real-life and stochastic element. Table 4.1 highlights the relevant literature on Dynamic Pick-up and Delivery Problems, indicating the two observations mentioned before: most of the previous research focused on one output measure (financial) and only includes one or two dynamic elements. This study is intended to fill this gap, by incorporating multiple objectives in the evaluation of the intelligent real-time planning system and including multiple real-life (stochastic) elements in the foundation of the system. The next section introduces this system, by mentioning the evolution and basics of Intelligent Transportation Systems.

Author(s) (year) Objective Dynamism Interval

Powell et al. (1988) Profit maximisation Demand arrival Discrete (6h)

(Regan et al., 1998) Profit maximisation Pick-up deadline Continuous

(Godfrey & Powell,

2002) Profit maximisation Demand arrival

(31)

30 (Yang, Jaillet, &

Mahmassani, 2004) Cost minimisation Demand arrival Continuous

(Tjokroamidjojo et

al., 2006) Cost minimisation

Demand arrival,

Pick-up location Discrete (24h)

(Zolfagharinia &

Haughton, 2016) Profit maximisation Demand arrival Discrete (12h)

(Zolfagharinia &

Haughton, 2017) Cost minimisation Demand arrival Discrete (6h and 24h)

This study Multi-objective

Production rates (demand arrival), (un)loading times,

Pick-up location

Continuous

Table 4.1: Relevant literature on Dynamic Pick-up and Delivery Problems

4.3 Intelligent Real-Time Transportation Systems

(32)

31 One of the most critical aspects concerning these systems is the availability of real-time information (Ichoua, Gendreau, & Potvin, 2007). However, with the recent tremendous advances in sensor, networking, geographic information and communication technologies, it is possible to gather data in real-time and include higher levels of dynamicity in operational planning (Billhardt et al., 2014). This is important because of the fact that only modelling the dynamic events is not enough, these should be integrated with information about uncertainties and current states, to fulfil the potential of real-time planning systems (Speranza, 2018). In contrast to the potential of real-time planning in logistics and transportation, the literature on this application is relatively immature (Nguyen, Zhou, Spiegler, Ieromonachou, & Lin, 2018). Contrarily, the application of real-time planning in the manufacturing domain is more elaborate, covering the development of real-time manufacturing control systems through modelling and simulation (Nguyen et al., 2018). Following the same approach, the transportation and logistics domain can benefit from the real-time control of the routing optimisation problem (Wang, Li, Zhou, Wang, & Nedjah, 2016).

From the perspective of real-time planning, the development of Information and Communication Technologies (ICT) is a crucial enabler and necessity for the potential benefit of this application (Psaraftis et al., 2016). These technologies can be considered as a tool that enhances the transparency, responsiveness and efficiency in supply chains (Coronado Mondragon, Lalwani, Coronado Mondragon, Coronado Mondragon, & Pawar, 2012). For transportation and logistics, the introduction of new services and technologies by academics and companies has led to the development of Smart and Intelligent Transportation Systems (ITS) (Mirzabeiki, 2013). An ITS is in general referred to as “tomorrow’s technology, infrastructure, and services, as well as the planning, operation, and control methods to be used for the transportation of persons and freight” (Crainic et al., 2009, p. 541). In particular, logistics operations can be enhanced by the exchange of information and status updates through an ITS (Schumacher, Rieder, Gschweidl, & Masser, 2011). This matches the increasing demand for information exchanges in logistic networks, stimulated by the growing complexity that comes with the implementation of just-in-time practices (Crainic et al., 2009) and the ever-increasing number of parties involved in transport operations (Stefansson & Lumsden, 2009).

(33)

32 simplification and automation of fleet management operations, with a particular focus on the institutional level (e.g. regional, national, continental) (Crainic et al., 2009). AFMS are, on the other hand, designed to manage the fleet of vehicles on a lower (corporate) level (Pillac et al., 2013). Therefore, AFMS are defined as “Advanced systems aimed at simplifying and automating freight and fleet management operations at the carrier or business-to-business level” (Crainic et al., 2009, p. 545). Considering the scope of this research, the AFMS is the category of interest in this section. In general, these AFMS are linked to problems of Pick-up and Delivery with uncertain demands and service times (Pillac et al., 2013). These freight ITS use and aid different transportation information types, including vehicle and freight location information, inventory information, cargo information and vehicle identity information (Mirzabeiki, 2013). The information is gathered and processed by different sub systems of an ITS, encompassing vehicle location monitoring systems, route planning systems and freight location monitoring systems (Mirzabeiki, 2013). The AFMS aim, together with the different subsystems, to integrate all the available information in the current transportation practices to enhance the effectiveness of the real-time operation (Crainic et al., 2009).

(34)

33

5. System design

This chapter considers the design of the new system, together with the design alternatives that are linked to this core system. Therefore, the first section of this chapter describes the core of the system, containing the process flow within this core system and the stakeholders that are relevant for this system design. Second, alternatives within this core system are described, focusing on the number of trucks, assignment rules (queue discipline), repositioning and diversion. These elements create the alternatives within the core model, by changing the rules or numbers compared to the starting situation (number of trucks, assignment rules and repositioning policy) or by adding a new element to the core of the Intelligent Real-Time Transportation Planning System (diversion). The dynamic nature of the transportation network is linked to these configuration elements, forming buffers or rules against variability in demand.

5.1 System core

The core of the system is based on the concepts related to Intelligent Transportation Systems (ITS), which is a relatively new trend in transportation and logistics and can be defined as a system that consists of “obtaining, processing, and distributing information for better use of the transportation system, infrastructure and services” (Crainic et al., 2009, p. 543). This system relies on the newly introduced technologies, like GPS and mobile phone connectivity possibilities (Pillac et al., 2013), to connect the different actors in the transportation network and exchange real-time information on various aspects (Regan et al., 1998). Therefore, the philosophy of this Intelligent Transportation System is to connect the shipper and the carrier, by establishing a connective flow of information within and between both stakeholders in this internal transportation network. The content and the flow within the system is focused on the software, operating at the hardware that is already available, related to GIS, GPS and mobile communication technologies. This is in line with the need for further developments in software and decision support systems in the light of Intelligent Transportation Systems (Crainic et al., 2009).

(35)

34 order. This assignment and linking activity is triggered by two events in the system: a new request arrives (load-initiated dispatching) or a vehicle becomes available (vehicle-initiated dispatching) (De Koster, Le-Anh, & Van Der Meer, 2004). Both dispatching strategies are included in the system, to have a comprehensive decision mechanism that links transportation orders and trucks from different events in the system. This is a continuous real-time process, driven by the events and status of all actors in the internal network, which is different than discrete decision intervals used in most prior studies, where decisions are made during planned moments in time (Table 4.1) (Zolfagharinia & Haughton, 2017).

Both the assignment and repositioning of trucks is performed by the backbone of the Intelligent Real-Time Transportation System, the dispatch centre. This is a fully automated decision support system, making assignment and repositioning decisions based on the algorithms within the system. Until recently, most organisations relied on human dispatchers to manage the movements of the fleet of vehicles (Crainic et al., 2009; Pillac et al., 2013), as in the current situation of the transport network. These human dispatchers have cognitive limitations, that hinder the real-time monitoring of a complex and large transportation network (Crainic et al., 2009). However, with the newly introduced technologies and powerful computing possibilities, it is feasible to automate this monitoring and decision making (Crainic et al., 2009; Pillac et al., 2013). This dispatch centre receives the transportation requests from the factories and controls the movements of vehicles within the fleet. These vehicles are the entities that transport the product from the pick-up point to the delivery point, characterised by an individual status at any point in time: moving loaded, moving empty and idle (Regan et al., 1998).

(36)

35 The planning procedure of the Intelligent Real-Time Transportation Planning System is, with a distinction between order-initiated dispatching and vehicle-initiated dispatching, based on the procedure and processes described by Fleischmann et al. (2004) and Le-Anh & De Koster (2005). The planning procedure is visualised in Figure 5.1. Both procedures constitute the entire process and decision making within the core of the system, without the diversion capability and the activities performed by the vehicles (driving, loading, unloading). When a transport order from the factory arrives, the system checks the entire pool of vehicles for trucks that are idle (status). When one or more vehicles are idle, the order is assigned to the vehicle that scores the highest on the rating that is active (assignment rule). A message is send by the dispatch centre towards that vehicle, with the information about the order (pick-up and delivery point). When there is no idle truck in the pool of vehicles, the order is placed in the waiting list for orders. The priority of that order in the queue is based on the queue discipline that is active, the priority of all orders in the queue is updated when a new order arrives.

(37)

36

Figure 5.1: Planning procedure of the Intelligent Real-Time Transportation System

5.2 Number of trucks

(38)

37 of trucks will be changed during the evaluation (simulation experiments), to investigate the effect of the number of trucks on the performance of the different configurations within the Intelligent Real-Time Transportation Planning System, under the dynamic characteristics of the transportation network.

5.3 Assignment Rules

The second variable, which is part of the system and responsible for a set of different configurations, is related to the policy that links the vehicles and transport orders: the assignment rule. This dispatching or assigning policy is defined as “waiting until the last minute to fix a load’s assignment decision” (Tjokroamidjojo et al., 2006, p. 344). Within these policies, the assignment rules are simple to use and relatively well performing solution strategies (Le-Anh & De Koster, 2005). An assignment rule creates an assignment between the vehicle and the order, which is immediately executed (Fleischmann et al., 2004). The orders are placed in a queue that is ranked by priorities (queue disciplines). As stated before, this assignment is triggered by a vehicle becoming idle or the arrival of a new order, where both are capable of operating under the same assignment rules (Fleischmann et al., 2004). The use of these assignment rules in fleet management is, in most cases, applied to the problem of Automated Guided Vehicle (AGV) dispatching (Jaoua, Riopel, et al., 2012). However, other vehicles that are in direct contact with the dispatching agent, could be dispatched in the same way as an AGV and make therefore use of the same assignment rules (De Koster et al., 2004).

(39)

38 other assignment rule included in this study matches a random vehicle to the first order in the queue (matching unit). Both policies (queue disciplines and assignment rules) are used to create different configurations of the Intelligent Real-Time Transportation Planning System.

5.4 Repositioning

The third variable and an important element in the operations of a fleet of vehicles, constitutes the repositioning of vehicles. This repositioning of empty vehicles is a key activity to be able to continue the operations of the fleet, by balancing the supply and demand for future occasions (Arinos & Morabito, 2016). This asset repositioning can be defined as “the matter of where the truck will be positioned for serving future (unknown) loads” (Zolfagharinia & Haughton, 2016, p. 104), forming an important mechanism that influences performance measures such as cost and travelling distance (Topaloglu & Powell, 2007). With the variability in demand, influencing the amount and timing of transport orders, this is an important element to consider (coping with the unknown demand requests). This mechanism constitutes of a dispatching agent, making a decision on the repositioning location of the vehicle that becomes available (Crama & Pironet, 2019), to reposition the vehicles in a way that reduces empty movements (Tjokroamidjojo et al., 2006; Zolfagharinia & Haughton, 2017).

As shown in the core of the system, the first step of a vehicle that becomes available is to check for orders that are waiting. When there are no orders to process, the vehicle is repositioned. In general, there are two main policies and rules to follow in these situations: move the truck to a location or let the truck wait at the last delivery point (Crama & Pironet, 2019). This optimistic policy of moving the truck to another location, based on the knowledge window of the dispatching agent, might put the truck in a preferable position to access the next load (Zolfagharinia & Haughton, 2016). This allows reductions in fleet size with no devaluation in service quality (Dejax & Crainic, 1987). These effects will be evaluated by using two different repositioning policies: waiting at the last delivery location and moving back to the last pick-up point (loading silo).

5.5 Diversion

(40)

39 (Zolfagharinia & Haughton, 2017, p. 439). This redirection of empty moving vehicles requires real-time information on the position of vehicles and a fast communication between the dispatcher and the vehicle driver (Pillac et al., 2013; Zolfagharinia & Haughton, 2016). Therefore, it might serve as a relevant element in the context of this network, with unknown and variable transport requests. Within the context of vehicle routing, multiple studies indicated the positive effect of adding this diversion ability to the model (Ferrucci & Bock, 2015; Ichoua, Gendreau, & Potvin, 2006), with a potential positive effect of up to 4.3% on the overall performance (number of unserved customers, travel times and lateness) (Ichoua et al., 2006).

(41)

40

6. Simulation model and results

The previous chapter highlighted the different components of the Intelligent Real-Time Transportation Planning System (sections 5.2-5.5), which are the experimental variables that are part of the simulation part of this study. This chapter presents the design of this simulation study, together with the main results obtained through the evaluation of the different configurations of the Intelligent Real-Time Transportation Planning System. First, the design of the simulation is described, covering the experimental design and the model content and scope. Thereafter, the configuration is presented, describing the randomness, warm-up period, run length and number of runs of the simulation model. After the configuration, the model verification and validation are discussed. The last section covers the results of the simulation study, illustrating the outcomes of the simulation experiments.

6.1 Simulation model

6.1.1 Model description

The simulation model includes orders, trucks and starch. The transport orders are generated by the unloading silos (production lines), based on the amount of starch in the unloading silo. The order is linked to a loading silo before it enters the system. The orders are, when there is no truck available, placed in the queue for orders (based on queue discipline). The orders are assigned to the trucks according to the assignment rule, where after the truck moves to the loading silo that is linked to the assigned order. After loading, it moves to the unloading silo (production line) and unloads the starch. The amount of starch in the unloading silo is lowered by the demand from the production line (dynamic). After unloading, and there is no order in the queue, the truck is repositioned (based on repositioning policy). When the repositioning is set to the return policy, the truck might be pre-empted by an arriving order (based on the diversion capability). After repositioning, the truck is added to the list of available vehicles. A complete overview and description of the model is provided in Appendix K.

6.1.2 Model content

(42)

41 6.1.2.1 Input data

The fixed factors within the simulation model, which are therefore not part of the experimental design, relate to the loading and unloading rates, production rates, time between production stops (MTB stops), time between production stop and start (MTT start), reorder levels, initial starch levels and the loading silo choice (load silo for a new order). Except for the reorder levels and initial starch levels, all the other input is dynamic in nature (based on distributions), creating variability in the transportation network. The loading and unloading rates are, based on historical data, fitted with a statistical distribution. A complete overview of the statistical distribution (tonnes per minute) for each loading and unloading silo is provided in Appendix F (Figures F.1-F.12), where the distributions are evaluated through the use of the Chi-squared test and the Kolmogorov–Smirnov (KS) test. The Chi-squared test is a well-known technique for testing the goodness-of-fit for a statistical distribution, which shows the match between the statistical distribution and the empirical data (Robinson, 2014). However, the use of different cell ranges might lead to significant different results, highlighting the need for the second test: the Kolmogorov–Smirnov test (Robinson, 2014).

The production rates (tonnes per hour), MTB stops (hours) and the MTT start (hours) are modelled by empirical distributions, the best alternative when there is no appropriate statistical distribution (Robinson, 2014). Statistics on these empirical distributions are provided in Appendix G (Table G.1). The production rates (tonnes per hour) at FOX63 and FOX7 are both modelled as a triangular distribution, with the minimum, mode and maximum rates based on expert opinion. This is because of the fact that there is only a mean production rate available, no other historical data. The reorder levels are set to the company’s standards and the initial amount in the unloading silos is set between maximum capacity and reorder level. An overview of the unloading silo characteristics is provided in table 6.1. The load silo for an order (pick-up location) is determined based on the historical frequencies of the combinations between the order location (unloading silo) and the loading silos (Appendix E: Table E.1).

Unloading silo Capacity (ton) Initial level (ton) Reorder level (ton)

GNV 400 325 250 FOX63 60 55 50 FOX7 85 65 45 FNEXT 100 85 70 VMF 95 82 70 DexTAK 85 50 68

(43)

42 6.1.2.2 Experimental factors

The experimental factors are already introduced and described in the previous chapter, constituting of the number of trucks (5, 6, 7 and 8), queueing priority (FCFS, highest production rate and lowest current starch level), assignment rule (random and STD), repositioning policy (wait at unloading silo and return back to loading silo) and the diversion capability (true and false). This results in 72 different Intelligent Real-Time Transportation Planning Systems configurations, each with a different set of experimental factor values (diversion is only capable for the repositioning policy that sends the truck back to the loading silo). All configurations (experiments) are listed in Appendix I (Figure I.1).

6.1.2.3 Output measures

The output measures are included to evaluate the effectiveness of the different configurations, focused on the economical and sustainable implications. Therefore, the two main outputs measures are the number of unloading silo runouts per simulation run (economical) and the average empty driven distance (km) per order (sustainability). The number of stock outs is an important cost (financial) measure for the production company, resulting in relatively costly production disruptions (Dion, Hasey, Dorin, & Lundin, 1991). The empty distance travelled by vehicles is a widely used performance measure for environmental sustainability, creating an important initiative to improve sustainability in the transportation sector (Evangelista, 2014). Next to that, the utilisation of the pool of trucks, waiting times before assignment, lead time of trucks and the total orders completed are included (failure identification outputs), to be able to provide a background on changes in the two main measures, which are also included in the discussion section.

6.1.3 Model configuration

This section describes the configuration of the simulation model, depicting the randomness between runs, warm-up period, run length and number of runs. The model was coded in Java language, by using the Anylogic platform. An overview and explanation of the coded model are presented in Appendix K, including actual screenshots and representations of the model.

6.1.3.1 Common random numbers

(44)

43 the variance and the required number of runs (Robinson, 2014). However, this study has the aim to analyse the results through Analysis of Variance (ANOVA) tests, requiring independent samples. Therefore, a random seed is used to generate unique simulation runs, both within and between experiments.

6.1.3.2 Warm-up period

The second configuration element considered is the warm-up period, representing the time it takes for the model to be in a realistic condition (Robinson, 2014). This realistic condition is defined as the time “that the initial transient has passed and the model output is in steady-state” (Robinson, 2014, p. 175). A widely used method to determine the length of the warm-up period is the Time-Series Inspection, a graphical method involving “the visual inspection of time-series of the output” (Robinson, 2014, p. 175). For this study, the utilisation level of the pool of trucks is plotted for each hour of a simulation run. Running 10 replications and calculating the average for each hour results in the time-series plot presented in Appendix H (Figure H.1). The point at which the plot appears to be steady is assumed to be the length of the warm-up period (Robinson, 2014), which in this case is set to 120 hours (5 days).

6.1.3.3 Run length

The third configuration element, to be set before running the experiments, covers the run (replication) length. Following the number of replications and the situation wherein the simulation is non-terminating, the run length is an important element to configure (Robinson, 2014). There is, in contrast to the determination of the number of runs, no clear guidance on how to determine the run length (Robinson, 2014). A general rule of thumb indicates that the total run length should be at least ten times longer than the warm-up period (Banks, Carson, Nelson, & Nicol, 2010), together with the advice that a longer run length is preferred over the shorter equivalence (Robinson, 2014). With the warm-up period set to 5 days, the run length should be at least 50 days. To overestimate the run length and keep a balance with the time required to run the simulation, the run length is set to 6 months (184 days).

6.1.3.4 Number of runs

(45)

44 of replications can be determined by the Confidence Interval Method (Robinson, 2014). Within this approach, the number of replications is set to the number where the interval is below the desired deviation level (Robinson, 2014). For this model, both output measures (average empty travelling distance per order and total number of stock outs) and multiple experiments (configurations) are performed, to be able to select the configuration that requires the highest number of runs. This resulted in a number of 30 replications (runs), with the total stock outs as output measure against a 95% Confidence Interval (5% deviation/significance level). This threshold and the surrounding values are depicted in Appendix H (Table H.1).

6.1.4 Model verification and validation

The simulation model is verified by comparing the conceptual simulation model (Appendix B) with the final simulation model. This conceptual model is built in cooperation with the starch manufacturer, together with an evaluation of the final conceptual model by means of a walk-through with the company representatives (field experts in this transportation network). Next to this approach, the animations within the simulation software were investigated to verify the correct working of the model under different configurations. The model validation encompasses the input data, model behaviour and experiments. The input data was delivered by the starch manufacturer and was checked for and corrected at preciseness, by comparing the final input data with the calculations and opinions from the starch manufacturer. In a separate walk-through with the company representatives (field experts in this transportation network), the rightness of the model and behaviour were discussed. As described before, the configuration of the experiments is based on techniques that are well used in this setting: the Time-Series Inspection for the warm-up period, the general rule of thumb for the run length and the Confidence Interval Method for the determination of the number of runs.

Exploring Real-Time Transportation Planning in a Dynamic and Multi-Objective Supply Chain: an Intelligent Transportation System