Inland ports and information exchange: useful concepts to improve container transhipment terminals?

Size: px
Start display at page:

Download "Inland ports and information exchange: useful concepts to improve container transhipment terminals?"

Transcription

1 i

2 (Blank page) ii

3 Inland ports and information exchange: useful concepts to improve container transhipment terminals? Master thesis submitted to Delft University of Technology in partial fulfilment of the requirements for the degree of MASTER OF SCIENCE in Complex Systems Engineering and Management Faculty of Technology, Policy and Management by Melle Minderhoud Student number: To be defended in public on May 14th 2018 Graduation committee Chairperson First Supervisor Second Supervisor External Supervisor : Prof.dr.ir. A. Verbraeck, Section Policy Analysis : Dr. Y. Huang, Section Systems Engineering and Simulation : Dr. J.H.R. van Duin, Section Transport and Logistics : Dr.ir. C. Versteegt, Macomi iii

4 Acknowledgements This thesis is my final work as a student Complex Systems Engineering and Management at the TU Delft. It not only concludes my graduation, but also my life as a student in Delft. It have been some incredible years. I can honestly say that I have enjoyed both my education here at TPM as my life in Delft tremendously. I am very lucky that my research was supervised and guided by a very experienced and skillful team of people. This not only improved my research a lot, but also gave me the opportunity to keep learning and improving my personal skills. I would like to thank my supervisor form the TU Delft: Yilin, Ron and Alexander. Many thanks Yilin for helping me structuring my thoughts and writing. Ron, thanks for your very useful feedback which helped me placing this case study in its scientific context. Alexander, many thanks for your sharp remarks which pushed me to keep improving my work. Special thanks to my supervisors from Macomi: Pawel and Corné. Pawel, thanks for your extensive feedback on my simulation model and research. I truly believe that I have learned a lot from you while carrying out this project. Corné, thanks for the opportunity to graduate at Macomi and providing me this interesting topic. Also many thanks for sharing your experience in terminal operations and simulation. Thanks to the whole Macomi team for your support. I am very pleased that I was able to cooperate with Navis in this project. Oscar, many thanks for sharing your experiences and vision on the container transport chain. Also special thanks to Manuel, for sharing your tacit knowledge on container terminal operations. The hours we have spent together discussing various topics are vital to this research. I also like to thank the rest of the Navis Atom team for patiently answering my questions and the fun we have had at the office. My understanding of information exchange in the container transport chain is mainly deepened by the various experts I interviewed. I would like to thank all of them for the interesting discussions. Finally I would like to thank my friends, family and girlfriend Linda. Not only for their help during this research, but especially for their unconditional support during my life as a student in Delft. Enjoy Reading! Melle Minderhoud Delft, May iv

5 Executive summary A large increase in containerized maritime transport, bigger ships and bigger call sizes cause the need for container transhipment terminals to increase the volume they can handle without increasing their handling costs and handling time at the terminal. This research aims at deepening the understanding of how this goal can be achieved through integration of an inland port and improving information exchange in the container transport chain. First it is researched how information exchange and integration of an inland port can improve a container transhipment terminal given the current volume. Next it is estimated how much additional volume can be handled through integration of an inland port and improved information exchange. In this research the handling costs inside the terminal are measured by the gang productivity, as labor cost are 50% to 70% of total costs in a manual container terminal. The handling time is measured by the ship turnaround time at the terminal. Promising information exchange alternatives are identified. Improving future event information in the container transport chain improves the container terminal more than other information types. Information exchange can only be improved among partners that have established a certain level of trust. In the container transport chain this trust relationship is aligned with financial dependency. Transport partner are more inclined to improve information that is already exchanged than to share new information. Improving the accuracy of the estimated time of arrival (ETA) and real time sharing of the berth window are selected as the most promising alternatives to improve the container terminal. A more accurate ETA will improve the productivity of the gangs by reducing the overmanning and undermanning the terminal and by improving the allocation of the ship to shore (STS) cranes. Real time sharing of the berthing window reduces the waiting time of ships at the terminal as ships get notified that they can decrease speed when resources at the terminal will not be available upon arrival. Ships will save bunker costs when reducing their speed. When resources in the terminal are available earlier than the arrival of the ship, ships will get notified to increase speed, improving the utilization of the terminals equipment and workforce, resulting in less handling costs at the terminal. Using a simulation model in the context of a transshipment terminal in the port of Algeciras, the integration of an inland port is designed. Both the effects of this inland port as improved information exchange are estimated. The turnaround time and gang productivity of container terminals with congested yard can be improved through handling and storing containers at the inland port. This reduces the yard occupancy, improving the productivity of the yard equipment and the STS cranes. A transshipment terminal can be improved by storing long-stay transshipment container at the inland port in addition to handling import and export containers there. Customs facilities should only be implemented at the inland port when transhipment containers are stored there. No dedicated stacks in the yard of the deep-sea terminal for transport towards the inland port should be reserved as it will increase ship turnaround time at the terminal. An ETA that is exact the actual time of arrival (ATA), improves productivity of the gangs by around 2.5%. Real time sharing of the berth window increases the productivity by around 2%. A more accurate ETA decreases the effectivity of sharing of the berth window. Both alternatives do not improve the ship turnaround statistically speaking. Around 2.5% additional volume can be handled in the deep-sea terminal through real time sharing of the berth window. A similar results be achieved by integration of inland port where 2.5% of the transhipment containers and import and export containers are stored. In this case up to 5% additional volume could be handled through integration of the inland port. Combining the integration of an inland port with sharing of the berth window increases this number further. These results are case specific and dependent terminal specific properties like dwell times, productivity of the yard equipment and the container split in the terminal. v

6 Table of Contents Acknowledgements... iv Executive summary... v Table of figures... x Table of tables... xii 1 Introduction Research Problem... 1 Background Problem Statement Research questions Research flow diagram Research methodology and methods Deliverables The case: A transhipment terminal in the port of Algeciras Structure Literature Analysis Inland ports, Inland terminals, dry ports and extended gate Two main functions of inland ports... 7 Relationship between inland port and productivity deep-sea terminal... 8 Knowledge gap of close terminal controlled inland ports Information exchange in the container transport chain... 9 Information sharing vs Quality of information... 9 Research on information exchange in the container transport chain Measuring improvement of a container terminal Conclusion of literature analysis Problem identification Stakeholders involved in container terminal operations Overview of stakeholders and needs General factors and KPI s within an container terminal Financial dependencies in the transport chain Demarcation of the container terminal and transport chain Demarcation of the physical transport of a container in the transport chain Demarcation of terminal operations relationships Design options of an inland port Optimize for skipping of the main stack in the deep-sea terminal Container exchange strategy vi

7 Customs facilities at inland port Transport link between container terminal and inland port Buffer stacks to enable peel off operations Overview of design options inland port Design alternatives for information exchange Identification of information currently exchanged Description of selected information exchange alternatives Selection of information exchange alternatives Synthesis problem identification Model Specification Requirements of the model Main Assumptions Main objects and relationships Main objects and relationships Modelling effect yard occupancy on productive moves yard equipment Modelling of the inland port Modelling of information exchange alternatives Modelling of inaccurate ETA Modelling of the nomination of gangs based on the ETA Modelling of the allocation of cranes based on the ETA Modelling real time sharing of the berth window Modelling the arrival of ships Classes of ships and service level agreement Generation of the pro forma berthing schedule Increasing volume by increasing the number of ships calling the terminal Choice for discrete event simulation & Simio Implementation in Simio Case study: transhipment terminal in the port of Algeciras Transhipment terminal on the European Asian trade lane Specifications of the terminal Container Split Fleet mix Extending the container terminal with an inland port Quality of the gathered data Verification & Validation Validation of the conceptual model vii

8 Verification of the simulation model Validation of the simulation model Experimentation Warm up period, Runtime and replications Base case Experimental Design Improving accuracy ETA and share berthing window Customs at the inland port Exchange strategy Transport between inland port and deep-sea terminal Use of buffer stacks in deep-sea terminal Growth potential deep-sea terminal Sensitivity analysis Global sensitivity scan of uncertain variables Robustness of the growth potential towards assumptions and estimations Analysis of results Used methods for analysis The effect of information exchange Designing an inland port Customs facilities Exchange strategies Mode of transport The effect of buffer stacks Growth potential Growth potential of an inland port Growth potential of real time sharing of the berth window Growth potential of combining both alternatives Conclusion & discussion of growth potential experiments Sensitivity Analysis Global sensitivity model Robustness of the results Conclusion Answer to research questions Answer to sub question Answer to sub question Answer to sub question viii

9 Answer to sub question Answer to main research question Discussion Relevance of thesis Scientific Relevance Societal Relevance Limitations Limitations regarding the case Limitations regarding the conceptual and simulation model Future work Reflection ix

10 Table of figures Figure 1 - Research flow diagram... 4 Figure 2 - Research methodology & methods... 5 Figure 3 - Effect inland port on ship turnaround time in terminal... 8 Figure 4 - Effect real time sharing berthing window and more accurate ETA on ship turnaround time Figure 5 - Effect truck appointment system on ship turnaround time Figure 6 Financial dependencies in the container transport chain Figure 7 Relationship between factors and KPI s Figure 8 - Classification of information currently exchanged Figure 9 - Classification of information exchange alternatives Figure 10 - System diagram of a container terminal Figure 11 - Class diagram of most main objects and relationships Figure 12 - Relationship yard occupancy and handling time multiplier productive move in yard Figure 13 - Flow of containers of a deep-sea terminal coupled to an inland port Figure 14 - Visualization of uncleaned data Figure 15 - Relationship accuracy ETA and time till ATA (cleaned) Figure 16 - Delays and advances or arrival time over time Figure 17 - Allocation of cranes at arrival of ships algorithm Figure 18 - Algorithm sharing of the berthing window Figure 19 - Paradigms in Simulation Modelling (Adapted from Borshchev & Filippov (2004, p3, Figure 3)) Figure 20 - Simio model in interactive mode Figure 21 - Demarcation of supply chain (seaside) Figure 22 - Geographical representation of transport options Figure 23 - High level experimental design Figure 24 - Yard occupancy and handling time multiplier with low, medium and high point estimation Figure 25 - Effect information sharing alternatives on productivity gangs Figure 26 Statistical significance of information exchange alternatives on productivity gangs Figure 27 Statistical significance of information exchange alternatives on ship turnaround time Figure 28 - The relationship between customs and ship turnaround time Figure 29 - Effect storing empty containers on turnaround time Figure 30 - Effect exchange strategy on ship turnaround time Figure 31 - Effect exchange strategy on yard occupancy Figure 32 - Effect exchange strategy on utilization RTG's Figure 33 - Effect exchange strategy on productivity STS cranes Figure 34 - Effect exchange strategy on productivity gangs Figure 35 - Effect mode of transport and additional volume on turnaround time Figure 36 - Relationship buffer stacks with turnaround time Figure 37 - Effect implementation inland port and handled volume on turnaround time Figure 38 - Effect implementation inland port and handled volume on gang productivity Figure 39 - Growth potential created by real time sharing of the berth window given the turnaround time constraint Figure 40 - Growth potential created by real time sharing of the berth window given the gang productivity constraint Figure 41 - Growth Potential inland port and berth window sharing x

11 Figure 42 - Growth potential inland port and berth window sharing Figure 43 - Tornado chart sensitivity simulation model7.5.2 Robustness of the results Figure 44 - Robustness against a high number of used RTGs per STS crane Figure 45 - Robustness against a low number of RTGs per used STS crane Figure 46- Robustness against a high productivity of the RTGs Figure 47 - Robustness against a low productivity of the RTGs Figure 48 - Robustness against a high share of 40s Figure 49 - Robustness a low share of 40s Figure 50 - Robustness against a high dwell time of the containers Figure 51 - Robustness against a low dwell time of containers Figure 52 - Robustness against a low impact of the yard occupancy on the handling time multiplier 79 Figure 53 - Robustness against a high impact of the yard occupancy on the handling time multiplier Figure 54 - Overview container terminal (Kemme, 2012, p561, fig 1) Figure 55 - European and Asian yard lay-out (Carlo, Vis, & Roodbergen, 2014, p414) Figure 56 - Container terminal yard terminology (Zhao & Goodchild, p328, fig 2) Figure 57 - Yard equipment and their storage capacity (Stahlbock & Vo??, 2008, p4) Figure 58 - Dependecy BAP, QCAP and QCSP (Bierwirth & Meisel, 2015, p618, fig5) Figure 59 - Most important information exchange seaside Figure 60 - Most important information exchange on the landside Figure 61 - Classification of information currently exchanged Figure 62 - Accuracy of estimated time of arrival Figure 63 - Difference between ETA and ATA at gang nomination Figure 64 - Relationship improved accuracy ETA and real time sharing of the berth window on ship turnaround time Figure 65 - Effect of sharing container availability and truck appointment system on ship turnaround time Figure 66 - uncleaned dataset showing ETAs over time till ATA Figure 67 - R Code for cleaning dataset Figure 68 - Distribution of inaccuracy of ETA Figure 69 - Accuracy of ETA per ship (zoomed) Figure 70 - Generated inaccuracies ETAs Figure 71 - Poisson distribution vs Normal distribution Figure 72 - Effect distribution for deciding percentage loading moves Figure 73 - Overview of the simulation model in Simio Figure 74 - Structure ship generator Figure 75 - Sea route to terminal Figure 76 - Nominating gangs for inbound ships Figure 77 - Nominate gangs for ships arriving between 20:00 and 02: Figure 78 - Change ETA inbound ships Figure 79 - Implementation of the berth Figure 80 - Implementation of the yard Figure 81 - Implementation of Train and external rail terminal Figure 82 - Implementation of the inland port Figure 83- Effect warm-up period on variance and mean average turnaround time Figure 84 - Effect runtime on variance and mean average turnaround time Figure 85 - effect number of replications on mean and variance average turnaround time Figure 86 - Effect improved information exchange on waiting time xi

12 Figure 87 - effect improved information exchange on workforce nominated Figure 88- effect different exchange strategies and handled volume on average turnaround time. 138 Figure 89 - effect exchange strategy and volume on productivity gangs Table of tables Table 1 - Definitions and abbreviations...xiv Table 2 - Classification of inland ports... 8 Table 3 Needs regarding the transhipment container terminal from most important stakeholders 14 Table 4 - Factors & KPIs Table 5 - Measuring the factors and KPIs Table 6 - morphological chart design inland port Table 7 - Classes of ships Table 8 - Specifications studied terminal in Algeciras Table 9 - Container split landside Algeciras Table 10 - Container split seaside Algeciras Table 11 - Fleet mix of studied transhipment terminal Table 12 - General input variables experiments Table 13 - Experimental design information exchange experiments (first part) Table 14 - Experimental design customs experiment Table 15 - Experimental design exchange strategies Table 16 - Experimental design mode of transport Table 17 - Experimental design buffer stacks Table 18 - Experimental Design growth potential Table 19 - parameters global sensitivity scan Table 20 - Level of uncertainty in variables Table 21 - Functions based on point estimation handling time multiplier Table 22 - Most important P-values of information exchange experiment on productivity gangs Table 23 - Two sample T-test for customs experiment Table 24 - Most important p-values: effect exchange strategy on turnaround time Table 25 - P-values: impact exchange strategy on gang productivity Table 26 - P-values: experiment mode of transport Table 27 - T - tests for the use of buffer stacks Table 28 - P-values: mean of ship turnaround base case compared with scenarios handling additional volume Table 29 - P-values: mean of gang productivity base case compared with scenarios handling additional volume Table 30 - Tukey multiple comparisons of means turnaround with different volumes and BW sharing Table 31 - Tukey multiple comparisons of means gang productivity with different volumes and BW sharing Table 32 - P-values: Means ship turnaround of scenarios with additional volume compared with the base case Table 33 - P-values: Means gang productivity of scenarios with additional volume compared with the base case Table 34 - P-values: robustness against high number of used RTGs per STS crane Table 35- P-values: Robustness against a low number of RTGs per used STS crane xii

13 Table 36 - P-values: Robustness against a high productivity of the RTGs Table 37 - P-values: Robustness against a low productivity of the RTGs Table 38 - P-values: Robustness against a high share of 40s Table 39 - P-values: Robustness against a low share of 40s Table 40 - P-values: Robustness against a high dwell time of the containers Table 41 - P-values: Robustness against a low dwell time of the containers Table 42 - P-values: Robustness against a high impact of the yard occupancy on the handling time multiplier Table 43 - P-values: P-values: Robustness against a high impact of the yard occupancy on the handling time multiplier Table 44 - Interest of different actors Table 45 - Goals, KPI's and resources of different actors Table 46 - Additional state variables container class Table 47 - Additional states of ship class Table 48 - Extreme Condition testing (observations) Table 49 - Tracing of a ship (simplified) Table 50 - Tracing of a transhipment container stored at the inland port (simplified) Table 51 - Levene's test of equal variance for average assigned cranes Table 52 - Welch two sample T-test for average assigned cranes Table 53 - Levene's test for equal variance for turnaround time Table 54 - Welch's two sample T-Test for turnaround time Table 55-95% confidence interval average turnaround time for different warm-up periods Table 56-95% confidence interval average turnaround time for different runtimes Table 57 95% confidence interval for different number of replications Table 58 - Levene's test for homogeneity of variance customs experiment Table 59 - Two sample T-test for customs experiment Table 60 - Levene's test for homogeneity of variance of the use of bothers with current volume Table 61 - T-test for difference in mean using buffers with current volume Table 62 - levene's test for equal variance of using buffers with high volume Table 63 - T-test for difference in mean using buffers with high volume xiii

14 Table 1 - Definitions and abbreviations Abbreviations and definitions Term Abbrev. Description Actual time of arrival ATA The time a vessel arrives at the terminal. Actual time of completion ATC The time the terminals services to a ship is completed. Actual time of departure ATD The time a vessels departs from the terminal. Base Case A scenario in which the status quo of the system is analysed. The basecase serves a starting point and comparison to other scenarios. Berth A place alongside the quay in which a vessel is moored. Berthing schedule A schedule providing information of the estimated arrival and departures of ship per berth. Berth schedules are often extended with the assigned cranes per ship. Dwell Time The time containers are stored at the terminal. Empty container MT A container without any cargo in it. Estimated time of arrival ETA An estimation of when a ships is scheduled to arrive at the terminal. Estimated time of completion ETC An estimation of when the terminal will complete servicing a ship. Estimated time of departure ETD An estimation of when a ship is scheduled to depart the terminal. Experiment A number of simulation runs in which a number of scenarios is studied. Inland port An intermodal terminal directly connected to a seaport. Gang Group of stevedores (workforce) that work in a container Key performance indicator Knot Landside KPI terminal. An indicator that represent those aspects of an organizational performance that are most critical for the current and future success of the organization. Measure of ship speed. One nautical mile (1,852 meters) per hour. Referring to activities happening on the side of the terminal that is most away from the water. Moves per hour Mph Indicates the number of moves a piece of terminal equipment did per hour. In this study a move can be either 1 or 2 containers. Peel off method Quay Method where container in a stack all have the same destination. Landside transport can receive the top container instead of a specific one, removing all digging moves. A structure built in parallel to the bank of waterways to allow for the mooring of ships. Quay crane QC Cranes located on the quay. Reefer Refrigerated container. Rubber tyred gantry crane RTG A mobile gantry crane with rubber tyred wheels. Used in the yard of container terminal. Scenario A set of input variables that is being studies using the simulation model. xiv

15 Terminal operating TOS Control system of a terminal. system Turnaround time TAT The time between the arrival and departure of a ship at the terminal. Twenty-foot equivalent TEU Standard measure of a 20 foot container unit Seaside Referring to operations happening at or closely by the water. xv

16 1 Introduction Container transhipment terminals are looking for ways to increase the volume of containers they can handle without increasing the handling time of ships at the terminal or the handling costs per container. This research aims to deepen the understanding on how container transhipment terminals can achieve this goal trough integration of an inland port and through improved information exchange in the container transport chain. Research questions on the design of an inland port and information exchange are proposed. The questions are answered using the design science research methodology and using a discrete event simulation model, expert interviews and literature research as main research methods. A case study on a transhipment terminal in the port of Algeciras is used to provide context of the research problem. 1.1 Research Problem Increased ship size and an significant growth of containerized maritime transport cause the need for transhipment terminals to increase their throughput volumes while keeping their handling costs and handling time per container at the same level. The productivity and performance of a container terminal can be improved through integrating an inland port and by improving information exchange between different transport partners in the container chain(roso, Woxenius, & Lumsden, 2009, p344; Gharehgozli, Roy, De Koster, 2016, p139). This research addresses multiple knowledge gaps in the design of integrating an inland port and information exchange alternatives and their effect on productivity, performance and the achieved throughput at the terminal. Background The globalisation of world trade led to a significant growth of containerized maritime transport. Nowadays more than 90% of the international transport is moved by ships and most is packaged in containers (Fransoo & Lee, 2013, p253). This increase in traffic puts pressure on operations in deepsea terminals. Bigger ships and call sizes contribute further to this pressure on deep-sea terminals (Nguyen & Notteboom, 2018, p1). Increasing ships size is an important part of the strategy of carriers to reduce their transport cost by achieving economies of scale. As a consequence of this cost reduction strategy, carriers are not willing to pay a higher rate for the handling of a container nor to accept a lower container handling speed from the terminal. Transhipment terminals are less stable in their demand compared to gateway terminals (Ducruet & Notteboom, 2012, p401). This makes them even more vulnerable to the market power of the carrier and increasing competition among transhipment terminals Problem Statement Container transhipment terminals need to increase the volumes they handle given the constraints of handling costs and time. A container terminal can do this by improving its throughput capacity, by means of increasing the number of and improving the productivity of terminal equipment and increasing the yard capacity of a terminal. Spatial development problems and costs constraints can render such solutions to be unfeasible. Productivity improvements in a container terminal can be achieved without increasing the theoretical throughput capacity by integrating an inland port and improving information exchange in the container transport chain. An inland port is an inland facility with a close relationship with a terminal in a port (Rodrigue, Debrie, & Fremont, 2010, p2). It enables the deep-sea terminal to handle more volume without increasing the ship turnaround time at the terminal or the number of resources needed at the terminal. The scope of this research does not include the specific design of the inland port itself. This research is demarcated to the question how a given inland port should be integrated 1

17 in the container transport chain and especially in the operations of the deep-sea terminal. The question which containers should be handled at the inland port and which should be kept at the inland port fall for example within this scope. In literature it is unknown how an inland port should be designed that is dedicated to increasing the productivity and performance of a transhipment terminal. The studied inland port has two main differences with most discussed inland ports in literature. The first difference is regarding the function of the inland port. The two main functions of inland ports are to improve productivity of the coupled deep-sea terminal and to improve hinterland accessibility (Roso, Woxenius, & Lumsden, 2009, p344). Inland ports are normally designed to serve both functions. In this research the only function of the inland port is to increase the productivity of the deep-sea terminal. The second difference is regarding the container split in the deep-sea terminal. To increase productivity in a deep-sea terminal normally only import, export and empty containers are moved towards the inland port 1. This research will focus on transhipment terminals specifically. For transhipment container terminals a strategy of exchanging long-stay transhipment container might be interesting given the large number of transhipment containers. Long-stay transhipment container are those containers that have a large dwell time within the yard of the deep-sea terminal. It is currently unknown whether exchanging long-stay transhipment containers also can contribute to an improvement in productivity. Information exchange is a well-researched topic in the improvement of supply chains. It is concluded that information exchange has both a positive effect on the service level as the productivity of supply chains (Li & Lin, 2006, p1642). Literature on information exchange in the container transport chain is still limited (Gharehgozli, Roy, De Koster, 2016, p139). It is known that the willingness to improve information exchange is dependent on the relationship between transport partners. An overview of effective and, from a social perspective, feasible information exchange is missing. Also their effect on the productivity, performance and achieved throughput at the terminal is unknown. Many different indicators are known to measure the performance and productivity of a container terminal (Henesey, 2006, p141). In this thesis the performance of the terminal is measured by the ship turnaround time in the terminal. This way it represents both the waiting time at the terminals as the productivity of the STS cranes. The productivity of the terminal is measured by the average number of moves per gang. This is preferred for two reasons. Firstly 50 to 70 percent of the handling costs in manual container terminal is labour related (Saanen & Dekker, 2006, p90). Secondly the gang productivity is more independent from the ship turnaround time than measuring the productivity of the STS cranes or the yard. Ship turnaround time and gang productivity together provide good insight in the state of a terminal. 1.2 Research questions The knowledge gaps described in the problem statement translate to the next main research question: How can the volume handled in a transhipment terminal be increased trough intergration of an inland port and improved information exchange without increasing ship turnaround time and decreasing gang productivity? As discussed in the problem statement handling more volume in a container terminal increases ship turnaround time in the terminal or decreases productivity of the gangs. Four sub questions are presented that guide research on how the ship turnaround time and productivity of the gangs can be 1 Source: VP Navis Atom, Navis 2

18 improved given the current volume in the terminal. This improved ship turnaround time and gang productivity are used to increase the volume in the transhipment container terminal to a point where the initial ship turnaround and gang productivity are reached again. The relationship between the four sub questions is shown in Figure 1. The four sub questions are: 1. What information exchange alternatives will be most effective on reducing the ship turnaround time in the terminal and increasing gang productivity and will have most support among the transport partners in the container transport chain? 2. What are the main mechanisms explaining how ship turnaround time and gang productivity are affected by the integration of and inland port and improved information exchange? 3. What is an effective high level design of integration of an inland port dedicated to decreasing the ship turnaround time in a transhipment container terminal? 4. What is the effect of improving information exchange between transport partners on the ship turnaround time and the gang productivity in a container transhipment terminal? 1.3 Research flow diagram A research flow diagram is shown in Figure 1. It shows the relationship between the different research questions. For every research question, the used research method and the research output is displayed. It can be seen that the first research question will be answered using a literature review and expert interviews. The output of this research question is a set of information exchange alternatives. The output of the second research question is a conceptual model that show the mechanisms of how an inland port and the identified information exchange alternatives affect the ship turnaround time and productivity of the gangs. This question is also answered using literature review and expert interviews. The conceptual model will be used to build a simulation model. Using a case study to provide the context for the simulation model, the effect of different design alternatives for integration of the inland port and the information exchange alternatives will be evaluated. These effects are visualized and tested for statistical significance. Now an effective design of the integration of an inland port and effective information exchange alternatives to reduce the ship turnaround time and improve gang productivity are known, the main questions can be answered by doing additional experiments using the simulation model and analyzing the statistical significance of the results. 3

19 Research Question Research output Main Question: How can the volume handled in a transshipment terminal be increased trough implementation of an inland port and improved information exchange without increasing the ship turnaround time and decreasing the gang productivity? Sub Question 1: What information exchange alternatives will be most effective on reducing the ship turnaround time in the terminal and increasing gang productivity and will have most support among the transport partners in the container transport chain? Literature Review Expert interviews Research method Information exchange alternatives Sub Question 2: What are the main mechanisms explaining how ship turnaround time and gang productivity are affected by the implementation of and inland port and improved information exchange? Literature Review Expert interviews Simulation Model Data Analysis Conceptual Model Simulation Model Data Analysis Sub Question 3: What is an effective high level design of an inland port dedicated to decreasing the ship turnaround time in a transshipment container terminal? Sub Question 4: What is the effect of improving information exchange between transport partners on the ship turnaround time and the gang productivity in a container transshipment terminal? High level design of inland port Results of effect information exchange Main Question: How can the volume handled in a transshipment terminal be increased trough implementation of an inland port and improved information exchange without increasing the ship turnaround time and decreasing the gang productivity? Simulation Model Data Analysis Figure 1 - Research flow diagram 1.4 Research methodology and methods In Figure 2 the research methodology and used research methods are shown. In this research the design science research methodology is used (Peffers, Tuunanen, Rothenberger, & Chatterjee, 2007). This research consists of six steps with iterations. In each step the shown research methods are used. The six steps are discussed briefly. 4

20 1. Identify problem & Motivate Problem demarcation 2. Define objectives of a solution Set of KPI s 3. Design and development Conceptual model 4. Demonstration Experimentation Results 5. Evaluation Conlusions 6. Communication Documentation Expert interview Scientific Literature Expert interview Scientific Literature Expert interview Scientific Literature Simulation Model Experimental Design Expert interview Data Analysis Figure 2 - Research methodology & methods 1. Problem identification and motivation. In this activity the problem is identified. The current state of the transhipment container terminal and its interfaces with the transport container is analysed and demarcated. This is executed supported by expert interviews and literature research. This step is documented in chapter 2 and Define objectives of a solution. This activity is about defining the design requirements of the design that is going to be delivered. In this research the design requirements are increasing or decreasing certain KPIs. These KPIs have been identified based on interviews with experts and scientific literature. This is documented in chapter Design and development. This activity is about designing the new system. In this research different alternatives on a strategic level have been identified that are useable in deepen the understanding of interesting mechanism inside the system. This step is supported by a literature research on container terminal operations and information exchange in the container transport chain. Different interviews have been conducted both with functional experts from Navis as external experts. This step will lead to a conceptual model of the container transport chain and conceptual models of the proposed adaptions. 4. Demonstration. The impact of the proposed adaptions on the specified KPIs will be demonstrated in this step. a simulation model is used to replace real world experimentation. An experimental design will be used in combination with the simulation model to estimate the effects of the different adaptions the systems on the KPIs most efficiently. This activity is discussed in chapter 5 and Evaluation. In the evaluation activity the results of the simulation model are analysed. The results of different experiments are discussed with experts from Navis to consider the useblitity of the model. Based on the results of the case general conclusion will be drawn on how transhipment container temrinals can be improved through information exchange and intergration of an inland port. 6. Communication. The last activity is about communicating the problem, solution and scientific relevance. This will be done by documenting the research in this thesis and by presenting the results. 5

21 1.5 Deliverables Next to the knowledge that is researched in this project, three more specific deliverables are specified. These are: A set of information exchange alternatives that will be most effective on improving ship turnaround time and gang productivity of a transhipment terminal and will have most support among the different transport partners A conceptual model showing how an inland port and the demarcated information exchange alternatives might improve ship turnaround time and gang productivity. List of design principles for the design of inland ports dedicated to improving the ship turnaround time and gang productivity in a transhipment container terminal. 1.6 The case: A transhipment terminal in the port of Algeciras This research is supported by a case study on the integration of an inland port for a transhipment terminal in Algeciras. The port of Algeciras is located in the strait of Gibraltar, providing it with strong geographical advantages to serve as hub in the Asian European ocean trade lane. It is one of the busiest transhipment ports in Europe. Almost 90% of the containers handled are being transhipped on the seaside, while only 10% is transported to the hinterland. The studied terminal suffers from a high yard occupancy, hurting their productivity and constraining their throughput capacity. The port authority of Algeciras is looking for ways to increase the competitiveness of the container transhipment terminals in Algeciras. As part of this project the port authority wants to know if integration of an inland port to the studied terminal can reduce its yard occupancy and if it will increase its productivity. In most cases the inland port is only used for import and export containers. Because the small share of these containers in this terminal, it is being considered to use this inland port as storage facility for empty and long-stay transhipment containers. Moving these containers to the inland port and back will cause additional operational cost. To know if these costs are justified, the port Authority wants to know how much additional volume can be handled in this terminal when it is integrated with an inland port. This specific case is provided by Navis. Navis is the biggest provider of terminal operating systems (TOS) to container terminals. To improve their products, Navis closely collaborates with stakeholders in the container transport chain. One of these collaborations is with the port authority of Algeciras. Navis supports this research with functional container transport chain expertise and case data. Macomi is specialised in simulation and advanced data analytics. They are also experienced with simulation of container terminal operations. Macomi supports this research with providing guidance on a regular basis and sharing expertise about the simulation in a container terminal context. 1.7 Structure This thesis consists of 7 main chapters and extensive appendix. The thesis starts with the problem statement, research questions and methodology in chapter 1. In chapter 2 the current body of knowledge is analysed using scientific literature. Chapter 3 provides a problem analysis resulting in the researched design alternatives and a set of KPIs. In chapter 4 the conceptual and simulation models are specified and discussed. Chapter 5 describes the case study, quality of the gathered data and the verification and validation of the developed models. In chapter 6 the experimental design is discussed. Chapter 7 elaborates on the analysis of the results of the experiments. The thesis ends with chapter 8 where the answers on the research questions, the discussion, limitations of the research and recommendations for future research can be found. 6

22 2 Literature Analysis In this chapter the current literature regarding inland ports, information exchange in the container transport chain and measuring the performance of terminals is analyzed. Section 2.1 the characteristics of an inland port, its main functions and the related mechanism known in literature. Multiple information exchange alternatives in the transport chain are discussed in section 2.2. Measuring improvement in a terminal is explained in section 2.3. This section concluded with an brief conclusion in section Inland ports, Inland terminals, dry ports and extended gate. Different labels for inland facilities where containers are being handled are used in literature. Some use the term inland terminal(rodrigue, Debrie, & Fremont, 2010,p1). Leveque and Roso (2002, in Veenstra, Zuidwijk, & Van Asperen, 2012, p20 ) use the term dry port and define it as an inland intermodal terminal directly connected to seaport(s) with high capacity transport mean(s), where customers can leave/pick up their standardised units as if directly to a seaport. Veenstra et al (2012, p20) propose the extended gate concept that differs from a dry port in the fact the (sea)terminal can control the exchange of containers with the inland port. Rodrigue et al (2010,p2) propose the generic term of inland port to describe those inland facilities with a close relationship with a port terminal. They define tree main criteria which are fundamental for inland ports. The core activity of inland ports must be the handling of containers. Added value services like consolidation, light manufacturing and transloading can also take place at the inland port. The inland port must have a dedicated and high capacity transport link between the terminal in the port and the inland port. This can either be trucks, a train or barges. The inland port must enable economies of scale in the inland distribution of containers as otherwise a container can be better picked up at the port directly. The third criteria is not mentioned in the definition by Leveque and Roso (2002, in Veenstra, Zuidwijk, & Van Asperen, 2012, p20 ). In this thesis the more generic term inland port will be used. Two main functions of inland ports The two main function of inland ports that can be derived from literature are improving hinterland accessibility by reducing transport costs and improving the productivity of a deep-sea terminal by transferring activities from the deep-sea terminal to the inland port. Around 25% of the transport costs of a container are hinterland transport costs. These costs can be reduced by a third with consolidation of these transport flows ( Stopford, 2002,p13; Notteboom & Rodrigue, 2005, p1; Veenstra et al., 2012, p25; Wilmsmeier, Monios, & Lambert, 2011, p1380). The second main function is transfer activities from the terminal towards the inland port to improve the productivity of the core activities (Roso, Woxenius, & Lumsden, 2009, p344). Terminals will try to optimize these processes as far as possible (Monios, 2011, p16). One way of achieving this is by providing customs clearance at the customs location (Wilmsmeier, Monios, & Lambert, 2011,p1380). The inland port can also serve as a storage empty depot and long-stay containers (van Schuylenburg & Borsodi, 2010, p4). These main function relate with the geographical distance between the terminal and the inland port. Roso et al. (2009, p ) distinguish between distant, midrange and close inland ports. Distant inland ports are more aimed at improving the hinterland accessibility, while close dry ports are more aimed at improving the productivity by mitigating capacity and congestions problems. 7

23 Relationship between inland port and productivity deep-sea terminal The performance of the yard equipment has a direct effect on the performance of the quayside (Hanh Dam; Le-Griffin & Murphy, 2006, p3). The productivity of the yard equipment decreases when the yard occupancy increases (Kemme, 2012, p570; Saanen & Dekker, 2006, p82). This is caused by the fact that the number of rehandles increase when the yard occupancy is high. Rehandles are unproductive moves needed to get access to a container that is currently blocked by other containers (Carlo, Vis, & Roodbergen, 2014, p415). This process is commonly referred to as digging containers. Inland ports are used to decrease the yard occupancy in a terminal and increase the yard productivity. When transhipment containers are being exchanged with the inland port, additional (productive) moves are caused as these containers need to be moved back to the deep-sea terminal. In this case a trade-off must be made between the increase in productive moves to achieve a certain decrease in unproductive moves. This mechanism is showed in Figure 3. Figure 3 - Effect inland port on ship turnaround time in terminal Knowledge gap of close terminal controlled inland ports Inland ports can be classified according their geographical distance with the deep-sea terminal and if they are controlled by the deep-sea terminal as discussed before. A classification of different inland ports can be found in Table 2. Most inland port are not deep-sea terminal controlled. An example of such inland ports are BCTN Ablasserdam, the dry port of Madrid and Isaka dry port Tanzania, which cover all distance classification (Monios, 2011, p10; Roso, Woxenius, & Lumsden, 2009, p342; van Schuylenburg & Borsodi, 2010). TCT Venlo is an example of a mid-range inland port that is terminal controlled (Veenstra, Zuidwijk, & Van Asperen, 2012, p15). No examples could be found of inland ports that have the main function of improving the productivity of the deep-sea terminal where the deep-sea terminal could control the exchange of containers. The proposed inland port coupled to the transshipment terminal in Algeciras will be in this category. More about this inland port will be explained in chapter 6. Table 2 - Classification of inland ports Controlled by terminal Distance: Close Mid-range Distant No terminal control BCTN Ablasserdam Dry port of Madrid Isaka dry port tanzania Terminal controlled Inland port Algeciras TCT Venlo 8

24 2.2 Information exchange in the container transport chain Information exchange can be divided into sharing new information and improving the quality of information currently exchanged. Improving the quality of information is easier to implement as a certain level of trust between the implementing transport partners already is established. Improving the accuracy of the estimated time of arrival, real time sharing of the berthing window and a truck appointment system are information exchange alternatives discussed in transport chain literature. Information sharing vs Quality of information Coordination is needed between the different members in the transport chain in order to improve the productivity of the chain (Lind, Haraldson, Karlsson, & Watson, 2015, p500). An important part of collaboration is the exchange of information. Improving information exchange can be achieved by improving information sharing or improving the quality of information (Li & Lin, 2006, p1642). Information sharing is defined as the extent to which critical, often proprietary, information is communicated to one s partner (Mohr & Spekman, 1994, p139). Organizations often view the disclosure of internal information as a loss of power and thus competitive advantage (Li & Lin, 2006,p1646). This makes the implementation of information sharing alternatives hard as most stakeholders in a container transport chain not only have to cooperate, but also compete with each other. Additional information sharing can only be achieved among stakeholders that already have established a certain level of trust (Mohr & Spekman, 1994, p138). Information quality consists of the accuracy, adequacy, credibility and timeliness of the information that is being exchanged (Monczka, Petersen, Handfield, & Ragatz, 1998, p559). Delay and distortion reduces the value of the information that is being shared and can impact the performance of a system in a negative way (Li & Lin, 2006,p1643). Research on information exchange in the container transport chain Information in the container transport chain can be categorized in chronological order. This is discussed in section Information exchange that is explicitly discussed in the container transport chain literature is the improvement of the estimated time of arrival, sharing the berthing window and truck appointment system. These will be discussed in from section onwards Categorization of information in the container transport chain Menger (2016, p17) proposes to categorize information in a chronological order. Four levels can be identified: Fixed information. Fixed information does not change over time. Examples of fixed information are the number of a container and its type. Historical information. Historical information is about events that happened in the past. An example of historical information is the time a container was discarded from a vessel. Status information. Status information represents the current conditions of a container. It can be for example that a container is currently waiting in the yard of a deep-sea terminal. Predictions / Future events. This information contains predictions about future situations. The estimated time of arrival is an example of a future prediction Accuracy of the ETA Operations research in the container transport chain often assumes exact estimations regarding the arrival time of ships. A clear example is research on the berth allocation problem and the quay crane assignment problem. Most studies either assumes static or dynamic arrival times (Bierwirth & Meisel, 9

25 2015, p620). When assuming a static arrival time, all ships are already in the port when the berthing schedule is made (Imai, Nishimura, & Papadimitriou, 2001, p404). With a dynamic arrival time ships have different arrival times. It is explicitly assumed that these arrival times are known when the berth schedule is created (Imai, Nishimura, & Papadimitriou, 2001, p406). There is discussion in literature whether the estimated time of arrival (ETA) is really as accurate as assumed. Lind et al. (2015,p501) state that the ETA often corresponds with the actual time of arrival (ATA). Others claim that there are discrepancies between the ETA and ATA and that this hurts the productivity and performance in the terminal (Fancello et al., 2011, p. 2; Xu, Chen, & Quan, 2012, p. 123). Improvements of the accuracy up to 25% have been reported (Parolas, 2016, 69, figure 21). Two mechanisms on how an ETA arrival affects the performance and productivity of a container terminal are discussed. The first is the allocation of workforce. Based on the ETA of inbound ships a workforce planning is made. Discrepancies between the ETA and ATA lead to overmanning and undermanning in the terminal, decreasing productivity and performance (Fancello et al., 2011,p2). The second mechanism discussed in literature is the allocation of the berth and ship to shore cranes based on the ETA. An inaccurate ETA cause the need for the terminal operator to continuously reschedule the allocation of cranes, resulting in poorer performance compared to the baseline crane assignment plan (Xu, Chen, & Quan, 2012, p125). Figure 4 shows how a more accurate ETA effects the ship turnaround time through improved gang utilization. Figure 4 - Effect real time sharing berthing window and more accurate ETA on ship turnaround time Sharing of the berth window Sharing of the berth window enables a carrier to adapt its speed based on the predicted state of the terminal at arrival. This can decrease waiting time and increase the utilization of the resources of the terminal (Lang & Veenstra, 2010). Waiting of ships at the area is considered waste, as a ship could have sailed at a lower speed (Lang & Veenstra, p478). Reducing the speed of the ship can save significantly on the bunker consumption of the ship. The bunker costs of a ships can contribute more than half to the total operating cost (Notteboom, 2006, p36). Reducing speed can save up to 20% of the total operating costs of the ship (Maloni, Paul, & Gligor, 2013, p151). 10

26 Figure 4 shows how sharing of the berthing window affects the ship turn around through reducing the waiting time and improving the productivity in the terminal. The study of Lang & Veenstra (2010) is mainly focussed on the cost effects for ships. Also they assume the ETA to be exact to the ATA Truck appointment system Introducing a truck appointment system will force truck operators to share their information on when they will pick up a container. This information can be used in two ways by the terminal to decrease the number of unproductive moves. The first is to prepare the container during a housekeeping move when the yard equipment would be idle. This method does not decrease the number of unproductive moves, but decreases their effect on the performance of the yard (Van Asperen, Borgman, & Dekker, 2013, p. 557). The second way is to determine a better location for the container when it is being rehandled. This means that a container will not be put on top of a container that will be picked up before the rehandled container will. This method decreases the number of rehandles caused by the retrieval of import containers by 50% (Zhao & Goodchild, 2010, p343). The effect of fewer unproductive moves on the ship turnaround is shown in Figure 5. It can be seen that fewer unproductive moves lead to less utilization of the yard equipment, improving the productivity of the both the cranes as the berth. No literature could be found on the quantitative effects of a truck appointment system on the productivity of the terminal and the ship turnaround time. Figure 5 - Effect truck appointment system on ship turnaround time 2.3 Measuring improvement of a container terminal Different alternatives have been identified which could improve a transhipment container terminal. To quantify the improvements caused by these alternatives, it must be known how operations in a container terminal can be measured exactly. Many different performance indicators of container terminals are mentioned in literature. Henesey (2006, p141) distinguishes four main categories of performance indicators. These are measures of productivity, production, utilization and level of service. Productivity means the output per unit of input. Productivity of a container terminal is about the effective use of equipment, land and labour (Dowd & Leschine, 1990, p108). The productivity of each piece of equipment and part of a terminal can be measured. Berth productivity is defined as the number of moves per hour for the total gross time that the ship is being serviced at the berth. The berth productivity is mainly dependent on the crane productivity, the number of moves per hour per deployed ship to shore (STS) crane and the number of deployed STS cranes per ship. The measure of production is the total throughput of a container terminal. This often measured in either containers or in TEU. The utilization is defined as a percentage that a resource is used or 11

27 utilized (Henesey, 2006,p274). The utilization of the berth, quay equipment and yard equipment are often used as performance indicators. The service time is one indicator of the level of service. The average service time lacks uniformity among different terminals. Some measure it as the total time the ships is berthed. Others measure it as the total time the ship is serviced (Le-Griffin & Murphy, 2008, p7). Henesey (2006, p159) uses the ship turnaround time as an indicator instead of the service time. It is defined as the departure time minus the arrival time of the ship. He includes the waiting time at the port as a separate indicator. It is defined as the time that a vessel has to wait before a berth is available. 2.4 Conclusion of literature analysis An inland port is an inland facility that is closely coupled to a deep-sea terminal. The two main functions of an inland port are improving productivity and the hinterland accessibility of the deep-sea terminal. The improvement in productivity is mainly achieved by decreasing the yard occupancy. In terminals where the productivity of the yard is constraining the productivity of the berth, a lower yard occupancy improves the productivity of the yard removing this constraint. The distance between the inland port and deep-sea terminal is closely related to its main function. In literature there is currently a research gap on how a terminal controlled inland port dedicated to decreasing the productivity of a transshipment terminal should be integrated. Improving information exchange in the container transport chain can be achieved by sharing new information and improving the quality of information exchanged. Information exchange alternatives can only be implemented between transport partners that have established a certain level of trust between each other. Sharing information is often conceived as a loss of power. Examples of information exchange alternatives discussed in literature are: improving the accuracy of the estimated time of arrival (ETA), sharing the berth window and a truck appointment system. Now the current state of the literature regarding inland ports and information exchange in the transport chain is analyzed, a more thorough analysis will be made of the problem regarding the implementation of both topics. This is discussed in the next chapter. 12

28 3 Problem identification The most important stakeholders involved in this part of the container transport chain want container transhipment terminals that handle much volume with high handling speed and low handling costs. This is discussed in section 3.1. The needs are translated to specific KPIs for the container transhipment terminal. An integration of an inland port will be designed to increase the productivity and performance of the terminal by reducing the yard occupancy. In section 3.2 the demarcation of the container transport chain and the processes within the container terminal is explained.the most important design choices regarding the inland port are whether it should include customs facilities, the exchange strategy, the mode of transport between both terminals and the use of buffer stacks. These are further explained in section 3.3. Section 3.4 discusses the identification of information exchange alternatives. Experts state that future event information has the highest impact on the productivity and performance of the terminal. Out of multiple alternatives, improving the accuracy of the estimated time of arrival for ships and real time sharing of the berth window are selected. These alternatives have most support among the implementing transport partners. This selection is made based on the trust relationship between the implementing partners and the preference to improve the quality of information above sharing new information. In section 3.5 a synthesis of the problem identifc 3.1 Stakeholders involved in container terminal operations Here the key performance indicators for a transhipment container terminal are defined based on the needs of the most import stakeholders. Section provides a brief overview of these stakeholders and the most important needs towards a transhipment container terminal. Based on these needs a set of key performance indicators is defined in section In section it is explained that the likeliness of stakeholders exchanging information increases when there is a financial dependency between them. It shows that this dependency exist on the seaside. Overview of stakeholders and needs Transhipment container terminals fulfil an essential transshipping and decoupling role in a container transport chain. Many stakeholders are involved in terminal operations. The most important stakeholders and their needs towards the transshipment terminal are analyzed here. These needs will be translated into KPIs. Appendix B contains a more extensive analysis of stakeholders and their needs. The shipper A hipper is an economic agent that produces and ships goods (Holguín-Veras, Xu, de Jong, & Maurer, 2011). They need container terminals that support reliable container transport chains with short transit times and low transport (Kent & Parker, 1999). The carrier Carriers sell transport to shippers from the place of origin to either a container terminal or in the case of carrier haulage a place more inland. They need reliable terminals with low handling rates and high handling speed. The terminal operator The main interest of the terminal operator is to increase profitability by increasing their throughput. To stay competitive they need to ensure a high performance in the handling of containers, while keeping their rates at a reasonable level. The port authority The port authority is the landlord, regulator and operator of nautical services in the port. Typical operator function they fulfil are pilotage and towage (Verhoeven, 2010). Their main interest is to 13

29 increase regional economic growth and the number of jobs by increasing the volume handled in the terminals of a port. Synthesis of needs It can be concluded that the needs of the most important stakeholders are aligned. The main stakeholders in the transport chain need a transhipment terminal that is capable of handling much volume with short handling time and low handing costs. The needs are shown in Table 3. Table 3 Needs regarding the transhipment container terminal from most important stakeholders Need N1 N2 N3 Description Low handling time Low handling costs Capable of handling much volume General factors and KPI s within an container terminal Based on literature described in section 2.3 multiple factors are identified that are used to measure the state of the terminal. These factors are used to build a conceptual model that shows the interdependencies between different parts of the terminal. The factors are shown in Figure 7. The factors are sorted according the classification of Henesey (2006). Out of these identified factors two KPIs are chosen. The most important stakeholders in the container transport chain need container terminals to be able to handle much volume with a high handling speed and low handling rates. In this thesis the handling time of the terminal is measured by the ship turnaround time in the terminal. This way it represents both the waiting time at the terminal as well as the productivity of the STS cranes. The handling costs of the terminal is measured by the gang productivity. This is preferred for three reasons. Firstly, measuring the real costs using a simulation model would require access to sensitive data from a real terminal. This data was not accessible in this research. Secondly, 50 to 70 percent of the handling costs in manual container terminal is labour related (Saanen & Dekker, 2006, p90). Thirdly, the gang productivity is more independent from the ship turnaround time than measuring the productivity of the STS cranes or the yard. Independence of KPIs is preferred as they would otherwise measure the same relationships within the terminal. Ship turnaround time and gang productivity together provide good insight in the state of a terminal. The last identified need, the number of volume handled, will be treated like an external variable in the conceptual model. The KPIs are underscored in Table 4. 14

30 Table 4 - Factors & KPIs Factors & KPI s Unit Need Category 1. Average Ship terminal turnaround Hours N1 Level of service time 2. Average Ship waiting time at terminal Hours N1 Level of service 3. Average moves per hour yard equipment Moves / (hour per piece of yard N2 Productivity 4.Average productive moves per hour yard equipment equipment) Moves /hour per piece of yard equipment N2 Productivity 5. Average moves STS crane per gang Moves STS crane / N2 Productivity per hour (gang * hour) 6. Average utilization rate RTG % N2 Utilization 7. Average Yard occupancy % N2 Utilization 8. Average Max Yard occupancy % N2 Utilization The proposed factors are frequently used in scientific literature and in industry. However how these factors are interpreted and measured differs. Table 5 explains how these factors will be measured in this study. STS cranes or RTGs are only considered available when there is workforce assigned to it. Table 5 - Measuring the factors and KPIs Factor & KPI s 1. Average Ship terminal turnaround time 2. Average Ship waiting time at terminal 3. Average moves per hour yard equipment 4.Average productive moves per hour yard equipment Measured as Actual time of arrival at a terminal (ATA) Actual time of departure at a terminal (ATD) Actual time of arrival at a terminal (ATA) - Time of Berth Total number of containers that are being operated by all cranes / total available hours of STS cranes Total number of containers that are being handled before an horizontal move towards either the quay or the gate by a RTG / the total available hours of RTG cranes Total number of moves at the berth / the number of gang hours assigned Hours of available RTG / hour of utilized RTG 5. Average moves STS crane per gang per hour 6. Average utilization rate RTG 7. Average Yard occupancy Number of TEU in yard / total theoretical capacity of a yard 8. Max Yard occupancy Max yard occupancy measured Financial dependencies in the transport chain Transport partners are more likely to exchange information with transport partners with whom they have financial dependency 2. Figure 6 shows that in the container transport chain the terminal operator receives money from the carrier on the seaside for their services and not from truck companies on the landside. This results in more information exchange on the seaside than on the landside. An analysis 2 Source: Member of the board of SDMG 15

31 of the information exchange on the seaside compared to the landside can be found in appendix C. Information exchange alternatives are more likely to be implemented on the seaside than on the landside because there is more financial dependency between the transport partners. Figure 6 Financial dependencies in the container transport chain 3.2 Demarcation of the container terminal and transport chain In this section the transport container chain around the container terminal is demarcated. The transport chain is demarcated from inbound ships 26 hours before their planned arrival at the terminal until the moment that a container leaves the terminal again. In any terminal many relationships and interactions are present. The scope of this research is on the relationship between yard occupancy, productivity of the yard and the berth as well as the ship turnaround time in the terminal. Also, the waiting time of ships and the productivity of the gangs are included. Demarcation of the physical transport of a container in the transport chain In this research the physical transport of a container through the chain is demarcated from the moment that an inbound ship is planning to arrive at the terminal within 26 hours until the moment that the container leaves the terminal by either the seaside or the landside. Landside arrivals enter the systems when a truck transfers the container to the yard equipment. Gang nomination of the terminal is dependent on the ETAs that are being exchanged in the last 26 hours before arrival of the ship. Therefore it is decided to include this part of the transport chain in this research. Demarcation of terminal operations relationships In a container terminal many relationships are present. This research focusses on the relationships shown in Figure 7. Also the relationship between the factors and KPIs defined in section and the factors are shown. The system is demarcated to the relationship between the yard occupancy and the ship turnaround time. The relationship between improved information and the productivity of the gangs is also included. These relationships are derived from literature and described in chapter 2. The challenge in this research, with regards to integration of an inland port is to design it in such a way that it improves the productivity of the yard and reduces the ship turnaround time maximally. 16

32 Figure 7 Relationship between factors and KPI s 3.3 Design options of an inland port In this research the integration of an inland port to a deep-sea transshipment terminal will be designed with the purpose of reducing the ship turnaround time at the deep-sea terminal. The inland port can be used to handle or store import, export, empty and transhipment containers. For transport between both terminals trucks, an external train and a shuttle train are considered. It is also considered whether designated stack should be used to enable peeling operations. Optimize for skipping of the main stack in the deep-sea terminal In order to reduce the yard equipment utilization rate it is essential that the movement of containers do not cause too many moves in the deep-sea terminal itself. In all explored designs the terminal operating system (TOS) will prioritize main yard skipping moves. A container can skip the main yard when transport towards the inland port is immediately available at the inland port at the moment it is being discarded by a STS crane. An example is a truck picking up a container destined for the inland port that is being discarded from a ship directly at the berth instead of taking a similar container from the main yard. In this case it is assumed that the trucks used for transport between the deep-sea terminal and inland port are allowed to collect containers at the STS cranes. Container exchange strategy To reduce the yard occupancy in a deep-sea terminal, either import/export containers, empty containers or long-stay transhipment container be moved to the inland port. The tree options are discussed in this section. 17

33 Import / Export containers Inland ports are often used to handle import and export containers. Deep-sea terminals can push import container quickly towards the inland port to alleviate congestion (Veenstra et al., 2012, p15). Export containers can be kept as long as possible at the inland port before sending them to the deepsea terminal. Both reduce the yard occupancy in the deep-sea terminal Empty containers Inland ports can be used to store empty containers. Container terminals need to hold a minimum number of empty containers at the terminal to fulfill late request for empty containers from carriers. Redundant containers can be moved to the inland port to reduce yard occupancy. The possibility of storing empty containers at the inland port also makes the use of the inland port more attractive for the handling of import and export containers as these operations often involve picking up or returning an empty container (van Schuylenburg & Borsodi, 2010, p4; Veenstra, Zuidwijk, & Van Asperen, 2012, p23) Transhipment containers The third option is to use the inland port to store long-stay transhipment containers. This option is not discussed in literature. This option will create two additional productive moves in the yard of the deepsea terminal for every transhipment container that is stored at the inland port. Storing long-stay transhipment containers at the inland port might be a good exchange strategy for transhipment terminals, because of the high number of transhipment containers in their container split. Customs facilities at inland port Some inland ports are part of a customs zone where customs services are available (Veenstra, Zuidwijk, & Van Asperen, 2012, p22). Moving customs clearance toward the inland port is one of the opportunities to transfer activities from the deep-sea terminal towards the inland port (Roso, Woxenius, & Lumsden, 2009, p344). This is mainly because inspection of containers creates at least two additional moves in the yard 3. When transhipment containers are being stored at the inland port, custom facilities should be available there, in order to avoid paying potential import and export duties 4. Transport link between container terminal and inland port A high capacity link is needed between an inland port and the container terminal (Rodrigue, Debrie, & Fremont, 2010, p2). Based on the context in the case of Algeciras and literature, multiple options are derived. These options are used to study the impact of some main properties of these alternatives on the effectiveness of an inland port. These main properties will be discussed at each alternative Truck Trucks are a flexible and cheap way of transport. It can create congestion problems within and around the terminal (Roso et al., 2009, p343). This causes unreliable transit times for trucks (Veenstra, Zuidwijk, & Van Asperen, 2012, p17). In the specific case of Algeciras, trucks between the inland port and the deep-sea terminal can only transport containers at nighttime to avoid additional congestion in the city center. This is further discussed in section 5.3. This property is used to study the effect of having downtime in the transport link. It can be studied by comparing the truck alternative with the external rail alternative. In the external truck alternative trucks are also used for the transport towards the train terminal. Because this terminal is located in the port, the transport link towards the inland ports does not have any downtime. 3 Terminal operations expert, Navis 4 Program Manger Logistics, Port authority Port of Rotterdam 18

34 External Train Trains can be used when enough containers can be consolidated between the inland port and the container terminal (Roso, Woxenius, & Lumsden, 2009, p344). Most inland ports are connected to the sea terminal by train often with the purpose to replace truck transport. (Veenstra, Zuidwijk, & Van Asperen, 2012, 26). Using trains reduces the congestion on the roads around the port (Roso, Woxenius, & Lumsden, 2009, p343). In the researched case the train terminal is not located in the terminal directly, but in the port, so additional transport between the deep-sea terminal and the train terminal by truck is needed. The main advantage of this is that containers can be pushed towards the train terminal at any time, as this terminal is located in the port itself. As discussed in the previous section, this mechanism is used to study the effect of downtime of the transport link on the effectivity of the inland port Shuttle train A shuttle train differs from an external train in the way that the whole train is dedicated to transport between the deep-sea terminal and the inland port. A high frequency service between the two terminals can be achieved by this shuttle train (Roso, Woxenius, & Lumsden, 2009, p342). In the researched case the shuttle is loaded and unloaded in the terminal directly. This is further explained in section 5.3. Loading and unloading of the shuttle causes additional moves for the yard equipment as the same equipment is used for both activities. When a shuttle is being loaded in the terminal, containers can be moved directly from the quay to the shuttle. When there is no shuttle in the container terminal, containers have to be stored in the yard. By comparing the results of the shuttle train alternative with the train alternative the impact of these additional moves can be assessed. Buffer stacks to enable peel off operations Buffer stack are designated stacks that only contain container that will be transported towards the inland port. Buffer stacks enable peel off operations, where the truck always receives the top container in the stack, eliminating all digging moves in this stack (JOC, 2014). The yard equipment is thus making less unproductive moves handling containers exchanged with the inland port. Assigning buffer stacks is a trade-off between the performance of the yard toward the inland port and the performance of the yard towards the seaside. The capacity of the normal yard will be reduced as some stacks will be reserved as a buffer stack. This can increase the number of unproductive moves in the main yard as more flexibility in the yard is preferred (Saanen & Dekker, 2006,p86). Overview of design options inland port In Table 6 an overview is given of the different design options for the inland port in an morphological chart. 19

35 Table 6 - morphological chart design inland port Exchange Strategy Import / Export Empty containers Long-stay transhipment containers Customs at Yes No inland port Mode of Trucks External train Shuttle train transport Buffer stacks Yes No 3.4 Design alternatives for information exchange An extensive analysis on the current information exchange in the transport chain in the context of the studied terminal in the port of Algeciras is executed. The identified information exchange has been clustered based on the type of information and whether it is needed on the seaside or the landside. Experts stated that information of the type future events have the most impact on improving the productivity and performance of a container terminal. Based on literature research and expert interviews the next four information exchange alternatives are identified: Improving the accuracy of the ETA of ships Real-time sharing of the berth window Sharing container availability predictions Implementing a truck appointment system Based on the trust relationship between the implementing transport partners and the preference of improving the quality of information over sharing new information, improving the accuracy of the ETA and real-time sharing of the berth window have been selected as alternatives with most support among the implementing transport partners. These two information exchange alternatives will be further researched. Identification of information currently exchanged To identify the most important information exchange that is currently exchanged in the container transport chain, the container chain through the port of Algeciras was analyzed with functional experts from Navis. The extensive analysis can be found in appendix C. The identified information exchange has been clustered using the chronological classification explained in section This classification has been further simplified into a class of future events and a class of status, historical and fixed information. The second classification is about if the information exchanged is needed on the seaside or the landside in the transport chain. Experts 5 stated that improving the exchange of future event data will have a much higher impact on the productivity of the terminal than focusing on status, historical and fixed data. This research will therefore be demarcated to improving future event data. Comparing the current future event data exchanged with the proposed future event data discussed in literature (see section 2.2.2), it stands 5 Vice President of Navis Atom, Member of the board SMDG 20

36 out that there is currently no sharing of the berth window and truck appointment system in the studied terminal Figure 8 - Classification of information currently exchanged Description of selected information exchange alternatives In the previous section it has been discussed that improving the exchange of future event information will have most impact on the productivity and the performance in the terminal. Based on the literature discussed in section and multiple interviews with experts, both from inside Navis as external, four information exchange alternatives are defined that are estimated to improve the productivity and performance in the terminal. These alternatives are improving the accuracy of the estimated time of arrival of ships, real time sharing of the berth window, sharing predictions on the availability of the containers in the terminal and a truck appointment system Improving the accuracy of the estimated time of arrival of ships Two ways on how the ETA accuracy affects the operations in a container terminal have been identified. The first is the allocation of workforce. Based on the ETA of inbound ships a workforce planning is made. Discrepancies between the estimated time of arrival and actual time of arrival lead to overmanning and undermanning in the terminal, decreasing productivity and performance (Fancello et al., 2011,p2). The second mechanism is the allocation of the berth and STS cranes based on the ETA. An inaccurate ETA cause the need for the terminal operator to continuously reschedule the allocation of cranes, resulting in poorer performance compared to the baseline crane assignment plan (Xu, Chen, & Quan, 2012, p125). A literature analysis on the accuracy of the ETA of ships can be found in section Real time sharing of berth window Sharing of the berth window enables a carrier to adapt its speed based on the predicted state of the terminal at arrival. This can decrease waiting time and increase the utilization of the terminals resources (Lang & Veenstra, 2010). Waiting time can be converted to lower bunker costs as it enables decreasing the speed, without consequences on the servicing of the ship when arriving at the terminal. 21

37 The bunker costs of a ships contribute to more than half of the total operating cost (Notteboom, 2006, p36). Reducing speed can save up to 20% of the total operating costs of the ship (Maloni, Paul, & Gligor, 2013, p151). Ships can also be advised to increase their speed slightly when their ship can be serviced earlier at the terminal. This will benefit the terminal as it will increase the utilization of its resources (Lang & Veenstra, p 480). A more productive terminal is also in the interest of the carrier as it will affect the rates it has to pay for the handling of containers. A literature analysis on sharing of the berth window can be found in section Share container availability prediction Trucks operators would like predictions on container availability from the terminal (B. Wiegmans, Van Arem, & Behdani, 2017). The container is available when it is released by customs and is waiting for pick-up in the terminal. In several interviews 6,7 with terminal experts it was mentioned that it is likely that truck operators could better allocate their resources if they have better predictions on when a container becomes available. Containers terminals do have the information to make these predictions. One expert 6 mentioned the possibility that the dwell time of import containers could be reduced when truck operators know the availability of containers earlier. Reducing dwell times for import containers will have a positive impact on the yard occupancy of the container terminals and thus on its efficiency. This would provide the incentive for the terminal operator to share the container availability predictions with its hinterland partners. It must be noted that the relationship between the information about container availability at the terminal and the dwell time is uncertain Truck appointment system on terminal Most terminals have much uncertainty regarding the pick-up time of import containers by truck operators. When a container is being picked up without prior notification, it potentially causes unproductive rehandling moves (Van Asperen, Borgman, & Dekker, 2013, p. 557). Introducing a truck appointment system will force truck operators to share their information when they will pick up a container. This information can be used in two ways to decrease the number of unproductive moves. Firstly, by doing the rehandling moves in the time the yard equipment would be idle (Van Asperen et al., 2013, p. 557). Secondly, by determining a better location of the container when it being rehandled, to avoid the need to rehandle it again shortly after. A literature analysis on truck appointment systems can be found in section Selection of information exchange alternatives The willingness for sharing information with a transport partner in the transport chain is connected towards the amount of trust between the transport chain partners (Jones, Fawcett, Fawcett, & Wallin, 2010, p706). The four identified information exchange alternatives are all situated on two specific relationships in the transport chain. Sharing container availability predictions and implementing a truck appointment system are both between the terminal operator and the truck companies on the landside. Sharing more accurate and timeliness vessel ETAs and berthing windows is information exchange between the terminal operator and the carrier on the seaside. As discussed in section there is more trust between the terminal operator and the carrier than between the terminal operator and truck companies, because there is a financial dependency 6 Vice president, Navis Atom 7 Logistic Consultant, TNO 22

38 between the carrier and the terminal operator. Carriers are also more consolidated than truck companies, making it easier to build a trusting relationship. It is more likely that transport partners will improve the quality of the information they currently exchange than they will share new information. Container availability predictions are currently not shared with the truck companies. There is some information being exchanged about the trucks that are going to pick up containers in the terminal. This could be upgraded to a truck appointment system. The ETA of the vessel and the berthing window are also currently exchanged, but the quality of the information could be improved. The classification of the different information exchange alternatives is shown in Figure 9. Alternatives that are based on increasing information quality and are among transport partners that have a trust relationships have most support for implementation. Based on this classification improving the accuracy of the ETA and the timeliness of the berthing window will be further researched. Figure 9 - Classification of information exchange alternatives 3.5 Synthesis problem identification Figure 10 shows the identified means, the relationships between the factors and the identified KPIs. A inland port decreases the yard occupancy and, in case of storing transhipment and empty containers, increases the number of productive moves in the yard of the deep-sea terminal. Both factors affect the utilization rate of the yard equipment reducing the productivity of the berth and in the end, the ship turnaround time in the terminal. Two information exchange alternatives have been selected. A more accurate estimated time of arrival of the ships will increase the productivity of the gangs. Real time sharing of the berthing window will also increase the productivity of the gangs and will also decrease the waiting time of ships. 23

39 The quantitative effects of the different design alternatives for integration of the inland port and the identified information exchange alternatives on the KPIs are unknown. In the next chapter conceptual models and a simulation model will be specified that will be used to estimate these effects. Figure 10 - System diagram of a container terminal 24

40 4 Model Specification Based on the system diagram from the previous chapter, a conceptual model is specified that is used to study the relations of interest further. The requirements of this model are explained in section 4.1. Section 4.2 describes the main assumptions of the model. The model focusses on the relationship between the yard occupancy and the ship turnaround time at the terminal and how this relationship is affected by integration of an inland port. Modelling the relationship between the yard occupancy and the ship turnaround is explained in section 4.3. Section 4.4 elaborates on how integration of the inland port is modelled. Several algorithms are developed that represent decision making affected by the information exchange alternatives discussed in the previous chapter. The way information is modelled is described in section 4.5. An important part of the model is the ship arrival generator. This generator is specified in section 4.6. Finally the conceptual model has to be implemented in a simulation paradigm and tool. The choice for discrete event simulation is discussed in section 4.7. Section 4.8 describes the implementation in Simio. 4.1 Requirements of the model In chapter 3 multiple design choices regarding the design of the inland port and the information exchange are identified. These alternatives will be studied using a simulation model. The simulation model will be useful for its purpose when it fulfils a set of requirements. These requirements are: 1. The model should capture the main dynamics between the productivity of the STS cranes, the productivity of the yard equipment and the yard occupancy. 2. The model should be able to capture the exchange and transport of containers between the deep-sea terminal and the inland port and the impact of these moves on the yard equipment of the deep-sea terminal. 3. The model should be able to capture the effects of the accuracy of the ETA on gang nomination and the allocation of STS cranes. 4. The model should be able to capture the effects of the real-time sharing of the berthing window on the speed of inbound ships and its consequences on the productivity of the gangs and the waiting time of ships at the terminal. 4.2 Main Assumptions To build a useful model within the time scope of this research, assumptions are made inevitable. The most important and high-level assumptions are discussed in this section. In other subsections in this chapter more specific assumptions are discussed. All container are assumed to be either standard 20 feet of 40 feet general cargo containers. Reefers are excluded from the model. The terminal gate is not a bottleneck in the system. Neither the productivity of the yard nor the STS cranes is constrained by horizontal moves within the terminal. The inspection rate of transhipment, import and export containers are equal. All inspected containers are released. All ships are of equal priority to the terminal operator. All ships behave the same. Equipment does not break down nor does workforce arrive late. The availability of the berth is constrained by the availability of STS cranes and berths and not by the length of the quay 25

41 4.3 Main objects and relationships This section describes the way the container terminal is modelled. Section describes the main objects and their relationships. Section explains how the relationship between the yard occupancy and handling time of productive moves in the yard is modelled. Main objects and relationships The most important objects and relationships between these objects in the systems are depicted in Figure 11. The whole chain is related to the container object. Several classes are distinguished based upon the origin, destination and current state (empty / full) of the container. Inside the chain the container is transported between different locations by a transporter. This can either be a ship, train or truck. When a container is not being transported, it is stored in a stack. Terminal handling equipment is needed to move containers between transporter or between transporters and stacks. The productivity of handling equipment depends on the occupancy of the stacks. These handling operations are only performed inside deep-sea terminals, rail terminals or in the inland port. A more detailed explanation of the transporter and yard equipment objects can be found appendix E. Figure 11 - Class diagram of most main objects and relationships Modelling effect yard occupancy on productive moves yard equipment In the model, the important relationship between the yard occupancy and yard equipment productivity is present. When the yard occupancy is high, the yard productivity decreases because of unproductive rehandling moves (see section 2.1.2). Unproductive moves are not modelled separately, instead productive moves by the yard equipment take longer in a congested yard. 26

42 Based on expert s opinion 8, the average handling time of a productive container move increases from 100% at a yard occupancy of 60% to 150% at a yard occupancy of 85%. Based on the same discussion, the relationship between the yard occupancy and average handling time of a productive container is assumed to be exponential. The effect of an additional container in a stack increases when the yard occupancy is higher. This is because a container stacked upon one other container creates one additional potential digging move. When a container is stacked upon three other container, it creates three additional potential digging moves. Using this estimation and assumption, a relationship between the yard occupancy and the average handling time multiplier is estimated. This estimation is only used for a yard occupancy between 60% and 100%. Based on the same discussion with the expert, it is assumed that beneath 60% congestion of the yard has a limited effect. The relationship is displayed in Figure 12. The sensitivity of the model towards this estimation is tested with a sensitivity analysis in section 7.5. It is concluded that the sensitivity of the results of the model towards this estimation is limited and that the lack of precision for these estimations does not hurt the usefulness of the model. Figure 12 - Relationship yard occupancy and handling time multiplier productive move in yard 4.4 Modelling of the inland port The model of the deep-sea terminal and inland port are composed of the same objects. Only external trucks, train and shuttle train objects are defined for the transport between the inland port and the deep-sea terminal. The main difference in the system with an integrated inland port compared to a system without an integrated inland port, is the way some of the containers move through the system. Some containers will be handled or temporarily stored at the inland port, reducing the yard occupancy in the deep-sea terminal. The flow of containers between the different objects identified in Figure 11 are displayed In Figure 13. The flow of containers is displayed per type (import, export, transhipment and empty). Flows that are currently in the system are represented with uninterrupted arrows. Optional flows that can be created in the system based on different design choices are represented with interrupted arrows. 8 Terminal operations expert, Navis 27

43 Figure 13 - Flow of containers of a deep-sea terminal coupled to an inland port The way that a container flows through the system is decided by the terminal operator based on information from the TOS. In the model the behavior of the terminal operator is simplified into decision rules. These decision rules consider the next properties of a container: Type of container Owner of the container Dwell time of container The exchange strategies are dependent on the next properties of the exchange strategy and the container terminal: Current number of empty containers in deep-sea terminal Availability of customs at inland port The type of containers to exchange Threshold dwell time for storing transshipment container at the inland port. These specific properties relate to the design alternatives described in section Modelling of information exchange alternatives Now that the physical system is modelled, the focus shifts to modelling information exchange inside this physical system. In section it is concluded that the accuracy of the estimated of arrival of ships is limited. Based on a real world dataset it can concluded that there are discrepancies between the ETA and ATA. These discrepancies are modelled using a biased random walk. Four algorithms are developed that take this ETA as input. The first is used to incorporate the effect of the ETA on gang nomination. This is described in section Two algorithms are developed that model the allocation of cranes to ships based on the ETA. These are discussed in section In section the last algorithm is discussed that is used to model the impact of real-time sharing of the berth window on in bound ships. 28

44 Modelling of inaccurate ETA Based on a dataset containing multiple ETAs communicated over time of around 40 ships calling the port of Algeciras, a model to represent the inaccuracy of the ETA is created. The dataset contains multiple estimations of the arrival time per ship. These estimation are from different sources and are used in real world decision making. This dataset is shown in Figure 14. Each color represents another ship. It is clear that some measurement errors are present in this dataset. Striking is a group of points where the difference between the ATA and ETA is roughly the same as the number of hours that the ETA was received before the ETA. The hypothesis is that a systematic measurement error was made here. These points where corrected using a cluster algorithm (see appendix F). These points now become white noise around the x-axis. When this hypothesis would be proven to be false and the points where actually correct, it would affect the outcome of this study as follows: The ETA is less accurate than shown in this study. Improving the accuracy of the ETA will improve the ship turnaround time and gang productivity more than shown in the results in this study. Real-time sharing of the berth window will improve the ship turnaround time and gang productivity more than shown in the results of this study. Some additional points where removed from the dataset. These points are received after the ship is already berthed. Also straight lines, consisting of set of point with equal distance on the x-axis of the same ships are considered measurement errors. These points are removed. In section 0 the circumstances under which the data was collected are discussed. Figure 14 - Visualization of uncleaned data The discrepancies between the estimated time of arrival and the actual time arrival are depicted in Figure 15. The ETAs of three ships are connected with a line to emphasize the path of the ETA over time. It is concluded that the communicated ETAs are not exact to the ATA. Most estimations have a deviation of around five hours, but also bigger discrepancies are present. There are more data points with a positive deviation than a negative one. It can be concluded that there is some inaccuracy in the estimations of the time of arrival and that these estimations tend to underestimate the time of arrival. 29

45 Figure 15 - Relationship accuracy ETA and time till ATA (cleaned) The inaccuracy of the ETA is modelled as an hourly change in arrival time. This hourly change in arrival time can be seen as a biased random walk. The random walk represent the possibility of both small and big changes in the arrival time. The random walk is slightly biased, because the arrival time has a bigger possibility of being postponed than advanced. A Bernoulli distribution is added to the random component to increase the probability that there is no delay nor advancement compared to the previous estimation. This is based on the observation that the ETA sometimes is equal for periods longer than one hour in the dataset. The model that is used to represent inaccuracy of the estimations of the time of arrival is shown in (1). The parameters of the Bernoulli and normal distribution are the results of generating multiple set of 100 random walks based on different parameters. The outcomes where visually compared to Figure 15. The chosen parameters model the ETA based on this dataset best. y t = y t 1 + B(1, 0.5) N(0.1, 1) (1) Figure 16 shows 40 paths of estimations of the time of arrival generated over time. The set seems useful to generate delays and advancements on the arrival time to represent the inaccuracy of the ETA. 30

46 Figure 16 - Delays and advances or arrival time over time Modelling of the nomination of gangs based on the ETA The availability of STS cranes and yard equipment is depended on gang work schedule 9. Once a day this work schedule is established for the next 24 hours of operations in the terminal. The schedule is created based on the service level agreement (SLA) of ships that are estimated to be serviced during the 24 hours that the work schedule covers. The next assumptions are made regarding the drafting of the work schedule. The work schedule is made based on the last updated ETA of the ships. The servicing time at the terminal is estimated when the schedule is made based on the number of moves of the inbound ship and the STS crane performance of the terminal the previous day. The number of gangs that will be reserved is equal to the number of STS cranes in the SLA. If a ship is being operated during a part of a gang shift, gangs are reserved for the whole shift. This means that when a ship arrives at the berth the last hour of a gang shift, the gangs are nominated for the whole shift. The maximum of available gangs is capped, based on the specific terminal that is being modelled. No difference is made between weekdays and weekend. Modelling of the allocation of cranes based on the ETA The ETA is important when allocating STS cranes to ships. Two algorithms are developed that allocate cranes based on the estimated time of arrival. The first algorithm is used when a ship arrives at the terminal. This algorithm is discussed in section The second algorithm is used to reallocate cranes that are available but estimated to remain idle during their working shift. This algorithm is discussed in section Allocation of cranes at arrival of ship When a ship arrives at a berth of a terminal, the terminal operator has to decide how many STS cranes will be assigned to the ship. The operator will try to at least satisfy the number of cranes which is agreed upon in the service level agreement (SLA). A terminal operator will assign more cranes to the 9 Source: Terminal Operations Expert, Navis 31

47 ship when additional resources are available 10. The highest priority of the terminal operator is to fulfil all SLAs. For the algorithm assumptions are made. These assumptions are made based on similar assumptions in Bierwirth & Meisel (2015) and discussions with multiple operation experts from Navis. The assumptions are: The cranes that are allocated at the start of operating a ship are time invariant. These cranes keep operating the ship the whole time that the ship is berthed. All STS cranes are equal and can operate all ships. Cranes are able to pass each other. When less than half of the agreed cranes are available, the ship will anchor at the waiting area. The terminal will assign at least the number of STS cranes in the SLA if enough cranes are available. If more than half of the agreed cranes are available, but less than the agreed number, all available cranes will be allocated. In this case the ship will be serviced immediately. The terminal operator will only allocate extra cranes if the allocation of the extra ship does not cause conflicts with the SLA of inbound ships. The designed algorithm is displayed in Figure 17. The algorithm starts when a ship arrives at the terminal. It will first determine how many STS cranes are available. Given the mentioned assumption cranes will be assigned to the ship or the ship will be send to the waiting area. When more STS cranes are available the algorithm consists of two loops which determine the maximum number of cranes that can be assigned without creating conflicts with the SLAs of inbound ships. The first loop controls the number of cranes that is assigned to the ships and starts with all the available STS. The second loop controls the time inside the loop. For every timestep it is checked whether conflicts arise because of gang shift changes and arriving ships. If none conflicts are predicted during the estimated time of operation, additional cranes are assigned to the ship. If conflicts are predicted, the algorithm tries again with one crane less until it reaches the number in the SLA Allocation of cranes at start gang shifts To avoid cranes that are idle during a whole work shift a second algorithm is implemented. Without this algorithm gangs can stay idle due to the time invariance property of the algorithm that allocates cranes at arrival. Cranes that are estimated to stay idle during a whole shift will be assigned to a vessel for the duration of that shift. The next assumptions are made for this algorithm: Cranes will be assigned to the ship with the earliest estimated time of departure given that there is at least four hours of work. The allocation of cranes is constrained by the location of the STS crane. The probability that a crane can operate is ship is equal to the number of idle cranes divided by the total number of cranes The structure of the algorithm is similar to the algorithm explained in section Source: Terminal Operations Expert, Navis 32

48 Modelling real time sharing of the berth window The berthing window of ships can change due to delays in seaside operations in the terminal and changing ETAs of inbound ships. Every 15 minutes a new berthing schedule is generated based on the ETAs of the inbound ships and the current state of the terminal. When communication the berthing window with the ships is enabled in the model, the ships will reduce their speed if they arrive before their berth window to decrease their waiting time at the terminal. If the allocated resources and the berth are earlier available than the ship arrives at the terminal, the terminal operator will advise the vessel to speed up. For the algorithm that determines the berthing window the next assumptions are made: The allocation of STS cranes to ships is time invariant. Cranes do not switch between ships during operations. All STS cranes are equal and can operate all ships. Cranes are able to pass each other. No more cranes than agreed in the SLA will be assigned. Ship will not reduce their speed lower than 10 knots. This assumption reflects the decreasing effect of reducing speed on the bunker costs. At some point the carrier will not reduce their speed any further. It is assumed that this point is 10 knots. Ships will not increase their speed above 20 knots. This assumption reflects the increasing effect of improving speed on the bunker costs. It is assumed that above 20 knots the increase in bunker costs is so high that the carrier will not increase its speed. Ships will only reduce speed in order to avoid waiting times at the terminal. Ships will increase their speed when their agreed resources are available earlier in the terminal. Updating the berthing window is done according the algorithm in Figure 18. The algorithm starts with the current state at the terminal and the estimated time of arrivals of inbound ships. The algorithms has a own state variable that keeps track of time. In every timestep a new berthing window is created where the arrival of new ships, completion of currently serviced ships and changes in the number of available gangs is processed. This way the algorithm keeps track of how many resources are predicted to be available at what time in the terminal. This is used to advice ships to lower their speed if they will arrive at the berth when no cranes or place at the berth are available. It can also advice ships to speed up to avoid idle resources at the terminal. A resource is considered idle when there are gangs available for the resource, but there is a lack of work. 33

49 Determine number of available STS Cranes in terminal Reserve cranes for ships in waiting area Decide number (Q) that can maximally assigned to the ship Less 50% of the cranes in SLA ship are available Send ship to waiting area Q <= cranes In SLA Assign Q to ship Q > cranes in SLA Calculate ETC based on Q Berth Vessel Process changes in crane allocation in timestep Set Q = Q - 1 Reset Time (T) Less than 0 cranes available 0 or more STS Cranes available Changes occur due to departure of ships, arrival of ships and changes in the available gangs Set T = T + timestep T < ETC T => ETC Figure 17 - Allocation of cranes at arrival of ships algorithm 34

50 Figure 18 - Algorithm sharing of the berthing window 35

51 4.6 Modelling the arrival of ships The model is sensitive towards the way ships are being generated. The goal of a ship arrival generator is to include as much variance possible in the arrival scenarios generated, without generating to much unrealistic scenarios. This way the design alternatives can be tested for their robustness against different pro forma berthing schedules. Section describes the different classes of ships and their properties. In section it is explained why a randomized equidistance approach is used to model the pro forma berthing schedule and the related arrival of ships. Section explains why the increase of volume handled in the terminal is modelled by increasing the number of ships that call the terminal instead of increasing the call sizes. Classes of ships and service level agreement Based on a real berthing window of the transhipment terminal discussed in the case and expert opinion 9 different classes of ships for the arrival generator are defined. These classes are defined based on the number of moves they cause at the terminal. Every class has a number of cranes that at least should operate it, according to the service level agreement, and a range of cranes that can maximally operate it. It is assumed that the probability of drawing a certain number from the specified ranges is for all numbers equal. An overview of the classes of ships and their range of specifications is shown in Table 7. Table 7 - Classes of ships Class Number of moves Cranes in SLA Maximum Cranes A distribution must be chosen that determines the percentage of loading (or discarding) container on a ship when generating ships randomly. Analysis of different distributions have been performed in appendix G. A uniform distribution of the percentage of loading moves between 45% and 55% of the total moves is chosen to include some variance without generating to many unrealistic scenarios. Generation of the pro forma berthing schedule Variance in the generated pro forma berthing schedules is desirable as the solution should be robust for a wide range of possible berthing schedules. The goal of the ship arrival generator is to include enough stochasticity to generate a wide range of possible pro forma berthing schedules without generating to much unrealistic berthing schedules. An unrealistic pro forma berthing schedules is for example a schedule where too much ships arrive at the same time. Normally a terminal operator would spread these ships in their pro forma berthing schedule 11. The way the interarrival times of ships are modelled have a big impact on the productivity of the terminal (Polman, Asperen, Dekker, & Arons, 2003, p1739). The effect of both a Poisson process as a form of equidistance is simulated and analysed in appendix G. Both methods are discussed by Polman et al (2003). It is chosen to model the interarrival time with a randomized equidistance approach, because a Poisson distribution generates to many extreme scenarios causing too much variance in the KPIs. An equidistance based approach also reflects reality better as a terminal operator can make sure 11 Terminal operation experts, Navis 36

52 that the pro forma berthing window is balanced 9, preventing for example that too much mega vessels need to be serviced at the same time. Increasing volume by increasing the number of ships calling the terminal The amount of volume handled in the terminal can be increased in two ways: bigger ships enabling bigger call sizes or more ships. Both ways have a different effect on the studied system and on the KPIs in the model. Carriers are currently increasing the ships sizes to achieve bigger economies of scale. Terminals have trouble handling these bigger call sizes as it increases the peaks in the yard occupancy of terminals. They also need bigger STS cranes to handle these ships. In the model bigger ships will increase the terminal turnaround time, because they need to be serviced longer. It is also expected that the bigger ships will decrease the effect of information exchange, because there will be a higher utilization of the terminals resources. The higher utilization of the terminals resources will be also improve the productivity of the gangs. Increasing the number of ships with the current fleet mix is also likely to decrease the effect of information exchange. More ships calling the terminal will increase the queue for the berth, making information less valuable. Because the utilization of the STS cranes will be higher, also the gang productivity will be higher. The ship turnaround time might increase, but only because of possible longer waiting times. Compared to increasing the call size of ships, the service time is in this case not influenced because the call size is the same. This is important because the ship turnaround time is the most important KPI and measured per ship. This KPI can be better compared in scenarios with the same call size. It is chosen to model additional volume by increasing the number of ships calling the terminal. 4.7 Choice for discrete event simulation & Simio In the previous sections conceptual models are developed that incorporate causal relationships between different objects and processes in the terminal, inland port and on the seaside. These conceptual models need to be implemented in a tool to run the model over time. This can either be done with an analytical model or a simulation model. The scope of this research includes multiple parts of the container terminal. Besides, the effect of the sub model of the yard on the sub model of the berth is an important relationship in this study. There are no closed-form analytical models to study integrated container terminals, making simulation the established approach for this research (Angeloudis & Bell, 2011, p524). The most common simulation paradigms are Discrete Event, Agent Based and System Dynamics. Borshchev & Filippov (2004, p. 3) distinct these three formalisms based on the level of abstraction that is used in the model and whether they deal with continuous or discrete processes. This distinction is shown in Figure 19. Some of the causal relationships defined in the conceptual model are dependent on the properties of an object on a low level. An example is the decision if a container should be moved to the inland port, which is based on multiple properties of a container. Some relationships are dependent on more aggregate states. An example is the relationship between the time a productive move will take and the aggregated state of the yard occupancy. The model is mainly focussed on discrete processes and states. The flow of container trough the model and the use of resources along the its way to the model is an important property of the model. These parts are best captured in the Discrete Event paradigm. 37

53 Figure 19 - Paradigms in Simulation Modelling (Adapted from Borshchev & Filippov (2004, p3, Figure 3)) Two development approaches are considered: Generic simulation tools (GST) and simulation programming libraries (SPL). GSTs are commercial of the shelve packages that are used for developing simulation models. They often have visual design features and are flexible when there are clear material flows in the system. GSTs contain a lot pre-defined logic improving the speed of development of the simulation model, but limiting the model structure. Well-known discrete event GSTs are Arena and Simio. SPLs provide a simulation framework which can be extended with parts developed in generic programming languages. SPL are more flexible in the model structure that can be used, but are less easy to use (Angeloudis & Bell, 2011, p526). It is chosen to use a GST, because more can be modelled in the time scope of the project and the structure of the model is quite common for a discrete event model. Simio V10 is chosen as simulation tool as the researcher is experienced using it. 4.8 Implementation in Simio The specified model is implemented in Simio version 10. Most parts of the specified model could be implemented without to many problems because of the extensive options available in Simio. Only implementation of sharing the berth window algorithm was troublesome, because Simio lacks functionality to sort a list of entities easily based on certain states. The sorting function in the algorithms was replaced by a recursive function implemented with a combination of the search and execute building blocks. Figure 20 shows an overview of the Simio model in interactive mode. The quay (red), yard (yellow) and inland port (blue) are distinguished using colours. Between the inland port and yard of the deep sea terminal multiple modes of transport are modelled. On the seaside, left in the model, multiple labels are shown reflecting current states of the model. They show the work schedule, current productivity per STS crane and the estimated time completion of ships at the berth. Appendix H describes how the model is implemented in Simio. 38

54 Figure 20 - Simio model in interactive mode 39

55 5 Case study: transhipment terminal in the port of Algeciras Context is needed to run experiments with the model specified in the previous chapter. In this chapter a case study on a transshipment terminal in the port of Algeciras is discussed. The specifications of this terminal are explored and the way an inland port should be integrated in this case is specified. The quality of the data used in in this case study is discussed. Finally the developed models are verificated and validated using the context of this case study. 5.1 Transhipment terminal on the European Asian trade lane The container transport chain on the seaside consists of two different kind of services. The first consists of the mega vessels transporting containers between the main gateway ports in Asia and Europe. These mega vessels, with a capacity up to 21 thousand TEU, call Algeciras as a hub to tranship containers among themselves or to tranship them to feeder ships. A small portion of the containers carried by the mega vessels actually gets imported inland (or exported) through the port of Algeciras. Algeciras is known as one of the hubs in the global transport network of Maersk. Most container handled in the port are from Maersk. The second part consist of smaller ships called feeders that transport containers between the smaller ports in the Mediterranean and West-Africa and the hub in Algeciras. This way the smaller ports are connected to the network of the big carriers. In the model both types of services are included. It is assumed that both have the same container split. Ships sailing towards Algeciras will enter the system 26 hours in advance, because from this time on gangs can be nominated for this ship. The ships leave the demarcated system again when operations at the terminal is finished. Figure 21 - Demarcation of supply chain (seaside) 5.2 Specifications of the terminal The yard occupancy of studied transshipment terminal is high, causing many rehandles when unstacking a container. This increases ship turnaround time and reduces gang productivity in the terminal, making the terminal less attractive for carriers to handle their containers. The studied terminal has a yard with an Asian lay-out 12 using RTG as their main yard equipment. Workforce must be nominated to the stevedoring company. Each day at noon the following four shifts must be nominated. The shift all take six hours and start at 14.00, 20.00, and The high level 12 See Appendix A for background information on the Asian yard lay-out. 40

56 specifications of the terminal can be found in Table 8. It can be seen that this terminal has 18 STS cranes. Only 14 of them can be used simultaneously due to a limited number of available gangs. Table 8 - Specifications studied terminal in Algeciras Variable Capacity yard Max number of STS cranes that can be used simultaneously Number of STS cranes 18 RTG per STS crane 3.5 Parameter TEU 14 RTG used on landside 5 Quay Length 2300M Container Split The container split is determined based on public available container throughput data from the Port Authority of Algeciras (APBA, 2018). The containers have been divided into being an import, export, transhipment and empty container. Also the share of 40s in the total number of containers has been determined. It is assumed the share of 40s is equal among all container types. The share of 40s in the studied transhipment terminal is 40%. The container split parameters have been separated based on if they enter the terminal on the land or seaside. The container split for containers entering from the landside is shown in Table 9. The container split for containers entering from the seaside is shown in Table 10. Table 9 - Container split landside Algeciras Container Type Percentage Empty 14% Export 86% Table 10 - Container split seaside Algeciras Container Type Percentage Empty 14% Import 4% Transhipment 82% Fleet mix The fleet mix is determined based on a 7 days berthing window of the studied terminal. The ships in the berthing window are classified based on the number of moves they bring to the transhipment terminal and whether it is Maersk vessel or not. In Table 11 the ship split of the transhipment terminal is shown. It is observed that 44 ships were serviced in a week at the transhipment terminal. In section the specifications of the different classes are discussed. 41

57 Table 11 - Fleet mix of studied transhipment terminal Class Observed percentage Observed percentage of being a Maersk vessel within this class 1 55% 73% 2 29% 70% 3 14% 83% 4 2% 100% 5.3 Extending the container terminal with an inland port The logistics network of Andalucía, a public company owned by the regional government of Andalucía is initiating an inland port 20 kilometres away from the port of Algeciras. The studied terminal is considering to use this inland port to reduce the yard occupancy in their terminal. Their main reason is that there is no storage growth possibility at the terminals location. The transhipment terminal is considering to handle import and export containers, empty containers and long-stay transhipment containers at this location. Three different modes of transport have been identified to move containers between the container terminal and the inland port. The different routes of these transport modalities have been displayed in Figure 22. Trucks A pool of dedicated trucks is reserved for the transport between the container terminal and the inland port. These trucks will collect the containers at either the buffers, the main stack in the terminal or at the berth and will drop them off at the main stack of the logistic zone. Trucks can only be used between 20:00 and 08:00 because of local congestion reasons. This means that the transport link has a down time of 12 hours a day. From a cost perspective a truck is the cheapest option. The logistic network of Andalucía (see Appendix B) will provide a pool of trucks with a fixed price of 50 euros per single trip between the inland port and the studied terminal. External Train At the inland port a rail terminal will be established which will be connected to a rail terminal in the port of Algeciras. An external train company will provide transport between these two places. The train has a capacity of 60 TEU. Transport between the rail terminal in the port and the terminal will be provided for by a dedicated pool of trucks from the ports stevedoring company. In the rail terminal some space is available where containers can be stored when waiting for the train. This could provide benefits for the deep-sea terminal as it can be used as an additional storage location for containers. There are no investment costs for the terminal when the external train is used. However the operation costs will be higher compared to the truck alternative. The trucks that are used for transport towards the train terminal in the port must be driven by workforce from the stevedoring company which are more expensive than other truck drives. Additional costs have to made to pay the train operator for transport and handling of the containers. 42

58 Shuttle Train A third is options is operating a dedicated shuttle train between the studied terminal and the inland port. The big difference with the external train is that the shuttle train will depart directly from the terminal. This way the train can be loaded directly from the stacks inside the terminal using RTGs. The need to load and unload the shuttle train with the available yard equipment could prove to be a downside to the shuttle train alternative as it requires additional moves to be made by the yard equipment. The success of the shuttle train depends on the number of yard skipping moves that can be made as well as the flexibility it provides in transport of containers compared to the truck alternative. From a cost perspective the shuttle train is undesirable as it requires high investment costs. A railroad to the terminal is already there, but a small loading and discarding section must be established. Figure 22 - Geographical representation of transport alternatives 43

59 5.4 Quality of the gathered data The quality of the data is an important determinant for the validity of the results. In this research four main sources of data were used: a dataset with ETAs from 40 ships that called the port of Algeciras a seven days berthing window of the studied transhipment terminal Aggregated container throughput data from the Port of Algeciras Estimations by experts from Macomi and Navis The quality of the gathered data will now be discussed per source. During a 10 days period in the autumn of 2017 the communicated estimated time of arrivals of 40 ships were recorded by Navis. The sample size of 10 days is too small to perform seasonality analysis. According to expert opinion 13 with local knowledge of the port of Algeciras, there is no reason to believe that the differences in the ETAs is different in autumn than on other times of the year. It must be noticed that the it was known by the participants that their ETA and ATA were recorded. This could have introduced the observer effect, where participant changed their behaviour because they knew it was being recorded. In this dataset multiple estimations and the actual time of arrival of each ship is present. The dataset contains the time each ETA was received by the system and the source of the estimation. The recorded sources are the port management system, the TOS of studied transhipment terminal and Marine Traffic. The dataset was very useful due to the recorded metadata. In some of the data points a systematically measurement error was present. This was corrected using a cluster algorithm. The berthing window of the studied transhipment terminal contains seven days of information about the type of ships that called the terminal and their service time at the terminal. From this berthing window several classes of ships could be specified. The berthing window also provided information about the number of STS cranes used in the terminal and the utilization of these cranes which is used in validation. The sample size of 7 days is rather small and prevents seasonality analysis on the arrival of ships. Based on expert opinion 14, there is no reason to believe that this week cannot represent an average week of operations in the transhipment terminal. Aggregated container throughput data is used to determine the container split of the studied terminal. As these number are the official throughput numbers, they are assumed to be reliable, but this cannot be verified. An important limitation of this dataset is that the shown throughput data represent all container terminals in Algeciras. The container split from the studied terminal can be different from the aggregated container split in the port. The dataset is the best estimator available of the container split of the studied transhipment terminal. Some high-level parameters had to be estimated by experts from Navis and Macomi. Most high level estimations regarding the modelling of a generic container terminal are validated in internal simulation studies. Case specific estimations where made by experts from Navis that had considerable knowledge of the port of Algeciras. Uncertain estimation of high level parameters have been discussed with multiple experts. Because estimation errors still will be present, sensitivity analysis will be applied to the most important parameters. 13 Terminal Operations Expert, Navis 14 Terminal operations expert, Navis 44

60 5.5 Verification & Validation In this section the usefulness of the model is assessed by verification and validation of the model. Both the development of the conceptual as the simulation model were based on extensive and numerous discussions with multiple experts from Navis and Macomi. It is concluded that the conceptual model is useful for this research. Verification is testing if the conceptual model is correctly implemented in the simulation model. This was tested using tracing and extreme value testing. The model was validated by comparing the model results with limited case data that was available. Both the structure of the simulation model as the generated results were discussed with expert from Navis. Based on these step it can be concluded that the simulation model is useful for this research. Validation of the conceptual model Evaluating whether the conceptual model is useful for its purpose is done continuously during model development. During the development of the conceptual model there have numerous and extensive discussion with experts from both Macomi and Navis. The algorithms for sharing of the berth window, the nomination of the gangs and the allocations of cranes where based on discussion at Navis. The final version of these algorithms where face validated by an expert from Navis. The algorithms where considered useful. However it was stated that only scheduling cranes at the arrival of the ships could be too unrealistic as cranes normally can change ships in-between shifts. Based on this feedback another algorithms was created. This algorithm reschedules gangs that are estimated to be idle for the whole shift. Verification of the simulation model Verification is testing whether the conceptual model is correctly implemented in the simulation model (Sargent, 2010, p5). During model development the model has been tested iteratively after the implementation of every new part. The final model has been tested using two different ways of dynamic testing. These are tracing and extreme value test. These test are discussed in appendix I. Based on these test the confidence level is high enough that model is correctly implemented in the simulation program. Validation of the simulation model For validation the results of the model are compared with the available historical data and by expert validation. As discussed in section 5.4 limited historical data was available for validation. Using the berthing window from the studied terminal, two results of the model where compared with the berthing window. The first was the number of the assigned cranes. In the berth window on average cranes per shift were assigned. In the (base case) model on average cranes per shift were assigned. The difference between the averages is within the boundary that would hurt the usability of the model. As discussed in appendix c the averages are also not different statistically speaking. The second comparison made with historical data is the expected ship turnaround time in the berth window compared with the average ship turnaround time from the model. The quality of expected ship turnaround from the berth window is limited as it only tells the expected turnaround time while the actual one might deviate. Besides the expected turnaround is reported very roughly per gang shift. Therefore all expected turnaround times are rounded to six hours. Nevertheless both values where compared. The average expected turnaround time in the berth window was hours. The average turnaround time in the simulation model was hours. This is around 30% lower than the expected turnaround time from the berth schedule. As discussed in appendix I.3 this difference is statistically significant different with p-value lower than

61 This results was discussed with a terminal expert from Navis. The difference could be partly explained by other activities at the berth like bunkering and repairs that are not included in the simulation model. This cannot explain a difference of 30%. Another possible explanation is that there is a difference between the expected berthing time and the actual berthing time in the terminal. This seems reasonable, given the rough estimations reported in the berth window. A third explanation would be that the model underestimates the turnaround time in the terminal. This is not a problem as the model is used to compare the ship turnaround time across different scenarios and not to predict the specific turnaround time. It can be concluded that the model is useful for researching the relative change in ship turnaround time caused by additional information exchange and the integration of an inland port, but that it might not be useful for exact prediction of the ship turnaround time in these scenarios. The results of this study remain trustworthy. The last validation method used was expert validation with an terminal operations experts from Navis. This validation session consisted of two phases. First the structure of the model was explained. It was explained how containers move through the model and what logic was added to the model. After the structure of the model was clear, it was run in interactive mode to show the movement of ships and containers through the model. The second phase was a discussion about the model and its results. The expert asked question about how certain mechanism were incorporated and why some parts of terminal operations were not included in the model. A specific example of a mechanism that was discussed was the relationship between a high yard occupancy and the performance at the berth. The relationship was discussed using results generated by the model. These results were considered trustworthy by the expert. Finally there was a discussion about the demarcation of the model. The discussion was mainly about omitting horizontal (truck) moves in the terminal and the modelling of the yard as one stack instead of multiple. The expert agreed that normally RTGs would be the bottleneck in the yard and that trucks therefore where not necessary to be modelled, especially as there was only one stack. The expert stated that adding multiple stack to the model could be a nice extension and could increase the usability of the model, but that the model was useful with how it was currently modelled. Based on the verification steps taken, validation of the conceptual model and validation of the simulation model it is concluded that the model is sufficiently useful for the scope of this research. 46

62 6 Experimentation In the previous chapters the problem is demarcated, the models are developed and the context for the experiments is discussed. In this chapter, the experiments used to explore the system of interest are discussed. Each experiment consists of a number of scenarios. The experimental setup of the experiments is discussed in section 6.1. The base case is the status quo of the studied terminal. There is no inland port nor is any of the information exchange alternatives implemented. The input variables for the base case are shown in 6.2. Starting at the base case, a funnel approach is used to design the experiments. Section 6.3 describes the experimental design. Finally, experiments for a sensitivity analysis have been defined in Warm up period, Runtime and replications Three important aspects of the experimental setup need to be determined before experiments can be executed. These are the warm-up time, the runtime of the model and the number of replications. The warm-up period is set at 21 days. This warm-up period is required as the simulation model starts without any containers in the yard and without any ships at the berth. These 21 days are required to reach the steady state of the model. A steady state of the model is the state where the influence of the starting conditions on the measured output of the model is limited. Therefore the gathered statistics of the first 21 days are omitted from the output. The runtime of the model is set to 100 days. The number of replications ran per experiment is 40. A thorough analysis that supports the choice of a warm-up period, runtime and number of replications can be found in appendix I. 6.2 Base case The base case is the starting point for the experimentation and analysis. In this study the base case is the status quo of the studied transshipment terminal. There is no inland port and the information exchange is not improved. The input variables shown in Table 12 are used. These input variables are kept constant among the different experiments unless a change is specified. These variables originate from the case data in chapter 5. 47

63 Table 12 - General input variables experiments Input Variable Value Unit Use Inland port? False Share berth window? False Quality of information B(1,0.5) * N(0.1,0.75) Inspection rate 0.01 Maersk container on Maersk 0.9 vessel Maersk container on non 0.3 Maersk vessel Share Maersk in export 0.75 containers Share of 40 s 0.4 Capacity yard terminal TEU Inland port Stack capacity TEU Inland port RTG capacity 6 Dwell time import / export 5.5 days Dwell time transhipment 5.5 days Max number of STS cranes 14 allocated Number of STS cranes 18 RTG per STS 3.5 Number of productive moves 11 per hour STS cranes Average ships a week 44 Average boxes a day from landside Experimental Design The experimental design follows as funnel approach as shown in Figure 23. The starting point is the base case scenario as discussed in the previous section. One funnel will perform the experimentation with the information exchange alternatives. The other funnel will design an inland port by exploring the effect of customs facilities at the inland port, several container exchange strategies, the mode of transport and the use of buffer stacks. The alternatives are discussed in section 3.3. and shown in Table 6. The growth potential of both the information exchange alternatives, the designed integration of an inland port as both measures combined will be explored. Also a sensitivity analysis based on the base case and the final results is executed. The funnel approach is chosen as it is a computational efficient way of executing experiments. However this approach has some limitations compared to a more systematic and automated approach. This approach is sensitive towards humans bias and fatigue which may results in neglecting interesting parts of the solutions space (Lee et al, 2015). 48

64 Figure 23 - High level experimental design Improving accuracy ETA and share berthing window The goal of this set of experiments is to explore both the main as interaction effects of improving the accuracy of the ETA, real time sharing of the berth window and the percentage of additional volume handled in the terminal on the proposed KPIs. The quality of the ETA can be improved in two ways in this model. The ETA can either match the previous ETA more often or the difference between two successive ETAs can be less when they do not match. It is chosen to model improved accuracy as a higher probability that two successive ETAs are the same. As explained in section this is controlled with the input variable of a binomial distribution. As we are interested in the interaction effect a full orthogonal design has been used. Table 13 shows the first part of this design where the increase of volume over all experiments is 0%. The same design is repeated for an increase in volume of 5% and 10%. Table 13 - Experimental design information exchange experiments (first part) Scenario Improved Accuracy ETA (%) Real time berth window sharing Increase in volume (%) Base Case 0 False 0 BW Sharing 0 True 0 ETA: 25% 25 False 0 ETA: 25% + BW sharing 25 True 0 ETA: 50% 50 False 0 ETA: 25% + BW sharing 50 True 0 ETA: 75% 75 False 0 ETA: 75% + BW sharing 75 True 0 ETA: 100% 100 False 0 ETA: 100% + BW sharing 100 True 0 The results of this experiment are discussed in section 7.2. Based on these results it is chosen to test the growth potential of real time sharing of the berth window. 49

65 Customs at the inland port The goal of this experiment is to get insight in whether a transhipment terminal also benefits from doing the customs clearance of import and export containers at an inland port. It must be noticed that this experiment is relevant when no transshipment containers will be stored at the inland port. Storing transhipment containers at an inland port requires the presence of custom facilities (see section 3.3.3). The experiments are run with overcapacity in truck transport. The experimental design is shown Table 14. Table 14 - Experimental design customs experiment Scenario Customs at inland Exchange strategy Transport Link port? Import / Export False Import / Export 100 Trucks Import / Export + True Import / Export 100 Trucks customs The results of this experiments are discussed in section It is concluded that in the case of handling import and export containers at the inland port, no customs facilities should be located at the inland port. Exchange strategy The goal of this set of experiments is to determine which exchange strategies are useful to increase productivity and decrease the ship turnaround time in the terminal. All exchange strategies are tested with overcapacity on truck transport. Table 15 shows the experimental design used to determine useful exchange strategies. Table 15 - Experimental design exchange strategies Scenario Exchange Strategy Lower threshold empty container (box) Upper threshold empty container (box) Threshold exchange transhipment containers (days) Import / Export Import / Export Empty 1 Import / Export Empty Empty 2 Import / Export Empty Transship (2,5%) Import / Export transhipment Transship (5%) Import / export transhipment Transship (10%) Import / export transhipment Transship (15%) Import / export transhipment of The results of this experiment are discussed in section Based on this experiment it is concluded that storing empty containers at the inland port is not a useful strategy in this case. Storing long-stay 50

66 transhipment containers is a useful strategy to reduce ship turnaround time and improve gang productivity. Transport between inland port and deep-sea terminal The goal of this experiment is to gain insight in the importance of the specifications of the transport link between the deep-sea terminal and the inland port. By comparing the truck alternative with the shuttle train and external train alternative, insight is gained on the importance of having a link that provides transport 24 hours a day. By comparing the shuttle train with the truck and external train alternative, insight is gained on the effect of additional moves by the yard equipment that needs to be done when loading the shuttle train. The results will also give insight in the best results for the case specific. This experiments have been executed with an exchange strategy where 2.5% of the transhipment containers are exchanged. The number of trucks, trains, and shuttle trains needed are determined with prior experimentation. It is interesting to see whether the results change as the terminal handles more volume. The experiments are executed with both the current volume as an increase of 10%. Table 16 - Experimental design mode of transport Scenario Mode of transport Number of transport Increase in volume (%) units Truck Truck 25 0 External Train External Train 10x (a day) 0 Shuttle Train Shuttle train 2x 0 Truck + 10%V Truck External train + 10%V External Train 12x (a day) 10 Shuttle train + 10%V Shuttle Train 2x 10 In section the results are discussed. It is concluded that the trucks are used in the next experiments. Use of buffer stacks in deep-sea terminal The goal of this experiment is to gain insight in the trade-off between optimizing the stacks in a transhipment terminal toward the seaside or towards the transport link with the inland port. Buffer stacks will improve the handling speed of containers towards the inland port, but this will hurt the handling speed of containers towards the seaside as the main yard capacity is reduced. In this experiment 2.5% of the transhipment containers are stored at the inland port. The mode of transport is trucks. The experiments are run with two levels of volume that need to be handled. This way the robustness of the results against the level of volume can be determined. The capacity of the buffer stacks are based on prior experimentation and is 500 TEU. Table 17 - Experimental design buffer stacks Scenario Buffer stack used? Capacity Buffer Increase in volume (%) Stack T2.5% False 0 0 T2.5% + buffer stack True T2.5% + 10%V False 0 10% T2.5% + buffer stack + 10%V True % 51

67 In section the results are discussed. It is concluded that buffer stacks should not be used when integrating an inland port to a container transshipment terminal. Growth potential deep-sea terminal Based on the results of the experiments described in the previous sections, it is explored how much additional volume can be handled in the deep-sea terminal through improved information exchange and integration of an inland port given the turnaround time and gang productivity constraints. Three sets of experiments are defined to explore the growth potential when both means are implemented separately and when they are combined. The experiments are full orthogonal. Table 18 shows the input parameters and the their input levels that are used in the three experiments to explore the growth potential of the deep-sea terminal given the mentioned constraints. Table 18 - Experimental Design growth potential Set of experiments Berth Window shared Exchange strategy Additional volume (%) Potential of berth True - False sharing Potential of inland port - Import / export Transhipment: 2.5% - 5% -7.5% - 10% Potential of both True - False Import / export measures combined Transhipment: 2.5% - 5% -7.5% - 10% The results of this experiment are discussed in section Sensitivity analysis Now the experiments are executed, a sensitivity analysis is executed to determine the sensitivity of the results towards certain assumptions and estimations. First the a global scan is performed by varying uncertain variables with the same percentage in the base case. This is described in section Secondly a more extensive sensitivity analysis is performed to determine the robustness of the estimated growth potential against the uncertainty of some of the estimations. This is described in section Global sensitivity scan of uncertain variables In this section the sensitivity of some input parameters that are kept constant in defined experiments are explored. The goal of this experiment is to find towards which variables the results of the model are most sensitive. The chosen variables have a level of uncertainty that is caused by potential estimation or measurement error. The effect of increasing and decreasing the values of the selected parameters with 20% will be tested. The sensitivity analysis is executed on the base case. The variables with their tested estimation range are shown in Table

68 Table 19 - parameters global sensitivity scan Variable Estimation range Change (%) Handling time multiplier at 1.4 ± yard occupancy of 85% Dwell time containers 132 ± 26.4 hours 20 RTG productivity 11 ± 2.2 mph 20 RTG per assigned STS crane 3.5 ± STS crane productivity 30 ± 6 mph 20 Share of 40 s 40% ± 8 percent point 20 The results of this experiment is discussed in section 7.5. Robustness of the growth potential towards assumptions and estimations The goal of the second sensitivity analysis is to get a more thorough understanding of how robust the results are given the uncertainty in the assumptions and estimations. For each variable first the uncertainty is determined. This is shown in Table 20. The estimation about how much the handling time of container in the yard increases when the yard is 85% occupied compared to 60% is most uncertain. This variable will be tested with a point estimation that is 50% lower of 50% higher. The average dwell time of a container is also uncertain. Therefore this variable is tested with an increase or decrease of 1.5 days compared with the used dwell time in the experiments. The parameters of the RTG productivity is often used in the simulation of container terminals. Therefore it is chosen to vary this variable with 20%. The same deviation is tested for the share of 40s. This number is based on all containers that are handled in the port of Algeciras and might deviate little for the studied terminal. Finally it is determined that the number of RTGs per assigned STS crane is either 3, 3.5 or 4. Therefore these levels are tested. Table 20 - Level of uncertainty in variables Variable Estimation range Change (%) Handling time multiplier at 50% ± 0.25 percent point 50 yard occupancy of 85% Dwell time containers 132 ± 36 hours 27.3 RTG productivity 11 ± 2.2 mph 20 RTG per assigned STS crane 3.5 ± Share of 40 s 40% ± 8 percent point 20 The handling time multiplier at a yard occupancy of 85% is a point estimation that is used to fit an exponential function. Changing this point estimation also changes the parameters in this exponential function. The different function based on different point estimations are shown in Table 21. The functions are visualized in Figure 24. Table 21 - Functions based on point estimation handling time multiplier Point estimation multiplier at occupancy of 85% Function 1.25 (low) 0.59 * e 0.89x 1.5 (medium) 0.38 * e 1.62x 1.75 (high) 0.26 * e 2.24x 53

69 Figure 24 - Yard occupancy and handling time multiplier with low, medium and high point estimation The sensitivity analysis will be performed on testing the ship turnaround time of the next three scenarios: Base case Integration of inland port when exchanging import, export and 2.5% of the transshipment containers while handling 5% additional volume. Sharing of the berth window while handling 5% additional volume. In section 7.4 it is explained that these three scenarios have statistically speaking no different turnaround time, supporting the conclusion that the latter two scenarios enable the handling of 5% additional volume. The goal of this sensitivity analysis is to find out how robust this conclusion is towards the uncertainty regarding the assumptions and estimations. 54

70 7 Analysis of results In this chapter the results of the experiments defined in the previous section are visualized, tested for statistical significance and interpret. This chapter starts with a discussion on the used methods for testing for statistical significance between the scenarios in section 7.1. The effects of improved information exchange on the ship turnaround time and the productivity of the gangs is discussed in section 7.2. In section 7.3 a high level design of integrating an inland port with a transshipment terminal is established. The growth potential of the information exchange alternatives and inland port is discussed in 7.4. The robustness of the these results against uncertain estimations and assumptions is tested in section Used methods for analysis In this chapter the results of the experiments are visualized and tested for significance. Testing of statistical significance starts with testing of equal variance in the results of the KPIs of the compared scenarios. Levene s test of equal variance is used. The independent T-test is used for comparing two scenarios. When the variance of the two compared is not equal, Welch s T-test is used. ANOVA is used to test for statistical significance when comparing multiple scenarios. In this case Tukey honest significance difference test (HSD) is used as post hoc test for pairwise comparison of the scenarios (Keppel & Wickens, 2004, p5). The most important adjusted p-values from Tukey s HSD are reported in this chapter. For all test a confidence level of 95% is used. R software is used for the analysis of the results. 7.2 The effect of information exchange Figure 25 shows a bar plot of the different information exchange alternatives with regard to the relative increase in gang productivity compared with the base case under three levels of volume. Multiple insights are derived from this bar plot. The productivity of the gangs can be improved by increasing the accuracy of the ETA and by sharing the berth window real time. Sharing of the berth window is about as effective as improving the accuracy of the ETA with 75%. Looking at the interaction effects it shows that combining sharing of the berth window with a more accurate ETA, improves the productivity more than sole implementation of one of the alternatives. This is logical as both alternatives improve the match between resources at the terminal and the work caused by the arrival of ships, but sharing of the berth window also spreads workload when it is higher than the capacity of the terminal. This can for example happen when to many ships arrive at the same time and more than 14 cranes are needed. In this case sharing of the berth window can decrease the speed of an inbound ship to spread the workload. The effect of sharing of the berth window on the productivity of the gangs decreases as the ETA becomes more accurate. This is logical as less adapting of the speed of ships is needed when there is a better match between available resources at the terminal and the workload through a more accurate ETA. The effect of the information exchange alternatives decrease when the number of ships calling the terminal increases. When more ships call the terminal, the terminal will more often work at its maximum capacity. In this case the terminal will look more like a queuing system, where there are no advantages in knowing the workload in the future. 55

71 Figure 25 - Effect information sharing alternatives on productivity gangs Figure 26, where boxplots of alternatives are shown with regard to the average moves per hour per gang, shows that a statistically significant 15 improvements in the productivity of the gangs is achieved when the ETA is improved by more than 50%. Sharing the berth window, without any improvements in the accuracy of the ETA, improves the productivity also statistical significant. Sharing the berth window with an improved ETA of 25% is statistically speaking not different from sharing the berth window without an improved ETA or sharing the berth window with a 50% more accurate berth window. Table 22 shows the most important p-values of this experiment. It shows that sharing the berth window is clearly statistical significant. It also shows that the impact of improving the ETA with 25% is not statistical significant. Table 22 - Most important P-values of information exchange experiment on productivity gangs Scenarios compared Adjusted P-value ETA 0 ETA 25 (both no BW) ETA 0 ETA 25 (with BW) ETA 0 (no BW) - ETA 0 (BW) Letters are used to visualise the statistical significance of the results. Boxplot with the same letter are statistically speaking not different. Boxplots with different letters have a difference in their mean that is statistically significant. Multiple letters can be assigned to one boxplot. 56

72 Figure 26 Statistical significance of information exchange alternatives on productivity gangs Figure 27 shows that the median of the average ship turnaround time is lower in the scenarios where the berth window is shared. The differences in means between most scenarios are not statistical significant. Improving the accuracy of the ETA keeps the ship turnaround more less the same. Only sharing the berth and a perfect accurate ETA together, reduces the ship turnaround time significantly speaking. Analysis in appendix K.1 shows that sharing the berthing window significantly reduces the waiting time of ships at the terminal. It also shows that there are significantly less gangs nominated when the berth window is shared and the accuracy of the ETA is improved by more than 50%. This might explain why in sharing of the berth window scenarios, the decrease of the ship turnaround is not statistically significant, while the increase in productivity of the gangs is. Figure 27 Statistical significance of information exchange alternatives on ship turnaround time The results of this experiment show that an inaccurate ETA is decreasing the productivity of the gangs inside the terminal. This supports the claim by Fancello et al (2011, p. 2) that the inaccuracy of the ETA 57

73 hurts productivity in the terminal and that the ETA should be improved. This insight questions the assumption that the ETA of ships is perfect, which is often done in operation research regarding the container transport chain (see section ). It shows that this assumptions does not hold and that the inaccuracy of the ETA has a statistically significant result on the productivity inside the terminal. The practical difference of 2.5% between the current accuracy of the ETA and an exact ETA is rather small. When researchers are willing to accept this inaccuracy in their results, the assumption can be defensible. Now it proven that the inaccuracy of the ETA does hurt the productivity in the terminal, the question remains how much it can be improved and what its effect is on the productivity of the gangs and the turnaround time in the terminal. In section it is discussed that the maximum improvement found in literature is 25% established by Parolas (2016). The results show that there is no statistical significant difference in the scenarios where the accuracy is improved with 25% compared to the base case. More research should be put into increasing the accuracy of the ETA. The effect of improving the accuracy with 25% on the additional volume that can be handled in the terminal will not be estimated as the improvement in productivity and turnaround time is not statistically significant. The results of sharing the berth window are in line what is reported by Lang & Veenstra (2010). Sharing of the berth window can generate operational benefits for both the carrier as the terminal operator. The terminal operator can utilize its resource better, which is proven by an increase in productivity of the gangs. The carrier saves bunker costs because waiting times are reduced. In contrary to this study Lang & Veenstra (2010) focus more on the operational costs while this study focusses on the operational processes. 7.3 Designing an inland port This section discusses the effect of customs at the inland port, the exchange strategy, the mode of transport and the use of buffers on the ship turnaround time and gang productivity in the deep-sea terminal. Customs facilities The goal of this experiment is to gain insight in the question whether customs facilities at the inland port improves the ship turnaround time when only handling import and export containers there. This experiment is not needed for scenarios where transhipment containers are stored at the inland port. In section it is explained that custom facilities are needed at an inland port when transhipment containers are stored there, because otherwise import and export duties have to be paid. Figure 28 shows a boxplot of the average ship turnaround time in a scenario with customs facilities at the inland port and a scenario without customs facilities at the inland port. It hardly makes any difference of having customs at the inland port when only handling import and export containers. Table 17 shows that with a p-value of 0.7 the null hypothesis should be accepted that there is no difference between the ship turnaround time in both scenarios. More on this statistical test can be found in appendix K.2. Based on the results and the t-test, it is concluded that moving custom facilities to the inland port does not improve the ship turnaround time when only the import and export containers of a transhipment terminal are handled there. It shows that the volume of import and export containers in a transhipment container terminal is too limited to make a difference by moving customs for these containers. This refines general inland port literature. Roso et al (2009, p. 344) for example see moving custom clearance as one the opportunities of improving the productivity of deep-sea terminal in the 58

74 context of a general container terminal. This experiment show that this does not hold for transhipment terminals when only handling import and export containers at the inland port. Figure 28 - The relationship between customs and ship turnaround time Table 23 - Two sample T-test for customs experiment Degrees of freedom T-value P-value Exchange strategies In section it is discussed that handling import and export container at the inland port a good way to increase the productivity of the terminal, but that their potential might be limited in a transhipment container terminal. Therefore it is tested whether also empty container or transhipment container should be stored at the inland port Storing empty container at the inland port Two strategies where empty containers are stored at the inland port are simulated. In both strategies also import and export containers are handled at the inland port. Figure 29 shows a boxplot with the average turnaround times of three scenarios. It shows that the average turnaround time is higher in scenarios where empty containers are exchanged. Storing empty containers also causes outliers in the results. There is no statistical difference when empty containers are stored compared to only handling import and export containers at the inland port. It can be concluded that in this case storing empty containers at the inland port does not improve the performance of the terminal. This results is unexpected as Roso et al (2009, p. 341) mention that the storage of empty containers should be available at an inland port. The scenarios where empty container are stored at the inland port are inspected in interactive mode to evaluate why the ship turnaround time is higher. It is discovered that the demand from the seaside for empty containers is very capricious. Empty container are only stored shortly at the inland port. This causes many additional moves in the yard of the deep 59

75 sea terminal compared to their effect on decreasing the yard occupancy. It can be concluded that storing empty containers at the inland port only improves ship turnaround time when they have a high dwell time at the container terminal. Figure 29 - Effect storing empty containers on turnaround time Storing transhipment containers at the inland port Figure 30 shows the effect of storing transhipment containers at the inland port on the ship turnaround time in the terminal. The handling of import and export containers is also shown as reference. It can be seen that exchanging more transhipment containers improves the turnaround time until the point where sharing more containers increases the turnaround time again. In this case storing 10% of the transhipment containers at the inland performs significantly better than only sharing import and export containers. The reported p-values in Table 24 show that transhipping 5% of the transhipment containers reduces the ship turnaround time almost statistical significant compared with only handling import and export containers. Storing 15% of the transhipment containers is significantly worse than just handling import and export containers at the inland port. 60

76 Figure 30 - Effect exchange strategy on ship turnaround time Table 24 - Most important p-values: effect exchange strategy on turnaround time Scenarios compared Adjusted p-value Import / Export Transhipment (2.5%) Import / Export Transhipment (5%) Import / Export Transhipment (10%) Import / Export Transhipment (15%) The mechanism behind this improved turnaround time at the terminal is shown in Figure 31, Figure 32 and Figure 33. Figure 31 shows that the yard occupancy in the deep-sea terminal decreases significantly when transhipment containers are stored at the inland port. It is notable that the yard occupancy increases again at the moment too many transhipment containers are stored at the inland port. This is explained by Figure 32. The utilization of yard equipment is so high than the waiting time for unstacking of container increases. Container stay longer in the yard, increasing the yard occupancy again. Figure 33 shows that the productivity of the STS increases aligned with the decrease in yard occupancy. It shows that storing 10% of the transhipment containers improves the performance of the STS cranes statistically significant compared to only handling import and export containers. The RTG utilization is not statistically different among most alternatives. This can point at the fact that it takes less time to make a productive move, but that more moves have to be made when more transhipment containers are being exchanged. 61

77 Figure 31 - Effect exchange strategy on yard occupancy Figure 32 - Effect exchange strategy on utilization RTG's 62

78 Figure 33 - Effect exchange strategy on productivity STS cranes Figure 34 shows the effect of the exchange strategy on the productivity of the gangs. Table 25 reports the p-values of the effect of the transhipment strategies compared to import and export containers. It can be seen that the productivity of the gangs follow the productivity of the STS cranes. Exchanging 10% of the transhipments containers improves the productivity of the gangs statistically significant comparing it to handling only import and export containers. Table 25 - P-values: impact exchange strategy on gang productivity Compared Scenarios Adjusted P-value Import / Export Tranship (2.5%) Import / Export Tranship (5%) Import / Export Tranship (10%) Import / Export Tranship (15%) Figure 34 - Effect exchange strategy on productivity gangs 63

79 It can be concluded that storing long-stay transshipment containers at the inland port is an effective exchange strategy in order to reduce the ship turnaround time and gang productivity at the deep-sea terminal. Based on the results the hypothesis can be accepted that the temporarily absence of a longstay container in the main yard has more effect on the productivity in the yard than the additional moves of transporting the container towards the inland port. Storing long-stay transshipment containers at an inland port is not discussed in general inland port literature. These results therefore extends current literature with a new concept and its effect on the deep-sea terminal. Reflecting on the results of exchanging empty and transshipment containers some additional insights can be derived. Storing empty containers reduced the productivity of the deep-sea terminal as the dwell time of these empty was observed to be too low. Storing long-stay transshipment containers did improve the productivity because the dwell time was high enough. It is derived that storing empty containers at the inland port can still be a useful strategy when the dwell time of empty containers is long enough. In the studied case, the dwell time of empty containers was to low and thus it reduced the productivity of the system. Mode of transport In this experiment the mode transport is determined for the case. By comparing main properties of the case specific transport alternatives some insights can be gained that can be useful for integrating inland ports at other terminals. Figure 35 shows boxplots displaying the average turnaround time of the ships of the different scenarios. Table 26 reports the p-values of comparing the means of the different modes of transport within the same level of volume. It can be seen that with the current level of volume handled in Algeciras all the transport alternatives perform roughly the same. Statistically speaking there is no difference between the different transport alternatives. When stressing the terminal with 10% additional volume some small differences arise. The train alternative and the trucks alternative have roughly the same mean, but the variance of the truck alternatives is a bit higher. The mean of the shuttle alternative is a bit higher than the trucks alternative which could point the fact that the additional moves in the deep-sea terminal caused by the shuttle have more impact in an congested terminal. This is supported by a decreasing p-value for the statistical tests involving the shuttle scenario when handling more volume. The differences in the mean among the different modes of transport are still not statistically significant when handling more volume. As discussed in section 5.3. trucks are the cheapest transport alternative in the case. Because there is no significant difference in the performance of the terminal using the different modes of transport, it is advised to use trucks in this case. In section it is discussed that by comparing the truck and train alternative insights can be gained on the effect of downtime in the transport link. Because the train alternative with no downtime does not perform statistically significant better than the truck alternative with 50% downtime, it is concluded that downtime does not affect the performance of the deep-sea terminal. It is also discussed that when comparing the external train with the shuttle train alternative, the effect of additional moves in the deep-sea terminal can be explored. Because the performance of both alternatives is statistically the same, it can be concluded that the additional moves in the yard do not affect the performance of the terminal. 64

80 Figure 35 - Effect mode of transport and additional volume on turnaround time Table 26 - P-values: experiment mode of transport Additional volume (%) Compared mode of transport Adjusted p-value 0 Trucks External Train 1 0 Trucks Shuttle Train External Train Shuttle Train Trucks External Train 1 10 Trucks Shuttle Train External Train Shuttle Train 0.13 The effect of buffer stacks Buffer stacks are dedicated stack for containers that will be moved towards the inland port. Buffer stacks enable peeling moves, where the yard equipment always can pick the top container, omitting all digging moves in this stack. The downside is less yard capacity for the other containers as no containers with another destination than the inland port can be placed in these stacks. Figure 36 shows the effect of using buffer stacks on the ship turnaround time given a number of volume that needs to be handled. The results show that in both case using buffer stacks increases the ship turnaround time in the terminal. Table 27 shows that in both volume experiments, the scenarios with and without buffer stacks are statistically different with very low p-values. More on these statistical test is discussed in appendix K.3. It can be concluded that in a transhipment terminal no yard capacity should be allocated to optimize handling of containers towards the inland port. The effect of the peeling-off method as used in Long beach is limited (JOC, 2014). These results agree with the statement of Saanen & Dekker (2006) that as much flexibility in the yard is needed for congested terminals. This means that no additional stacks should be reserved for containers towards the inland port. 65

81 Figure 36 - Relationship buffer stacks with turnaround time Table 27 - T - tests for the use of buffer stacks Experiment Degrees of freedom T-value p-value Current Volume % additional volume Growth potential Now it is known that sharing of the berth window and integration of in inland port can improve the productivity of the gangs and reduce the ship turnaround time in the deep-sea terminal, the growth potential of both alternatives is determined. The growth potential is the number of additional volume that can be handled inside the deep-sea terminal without increasing the ship turnaround time or reducing the gang productivity inside the terminal. In this section the most important results are reported. Section discusses the growth potential achieved through integration of an inland port. The growth potential caused by sharing of the berth window is reported in section In section both measures are combined. Section reports the conclusion and discusses the results. More results on the growth potential achieved through integration of an inland port is presented in appendix K.4. Growth potential of an inland port Figure 37 shows the effect of integrating an inland port while handling more volume on the ship turnaround time. Table 28 reports the p-values of comparison of the mean turnaround time of scenarios with additional volume with the base case. It shows that when 2.5% additional volume must be handled in de the deep-sea terminal, the average ship turnaround time increases. Statically speaking there is no difference between this scenario compared to the base case. The same applies when comparing the scenario where 5% additional volume is handled and where 2.5% of the transhipment containers are stored at the inland port. When storing 10% of the transhipment containers while handling 7.5% more volume, the average turnaround time is lower than in the base case. When transshipping 10% while handling 10% more volume, the ship turnaround time is 66

82 statistically significant higher than the base case. It is concluded that 7.5% additional volume can handling in the deep-sea terminal trough integration of an inland port without increasing the ship turnaround time. Figure 37 - Effect implementation inland port and handled volume on turnaround time Table 28 - P-values: mean of ship turnaround base case compared with scenarios handling additional volume Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) Base Case (2.5%) Base Case (0%) Tranship 2.5% (5%) Base Case (0%) Tranship 10% (7.5%) Base Case (0%) Tranship 10% (10%) Figure 38 shows the effect of integration an inland port and handling additional volume on the productivity of the gangs. It can be seen that when handling 2.5% additional volume in the base case, the average productivity is a bit lower, but that there is statistically speaking no difference. The p- values for these statistical tests are reported in Table 29. For all the transhipment scenarios that exchange transshipment containers on top of import and export containers, the improvement in gang productivity is statistically significant. 67

83 Figure 38 - Effect implementation inland port and handled volume on gang productivity Table 29 - P-values: mean of gang productivity base case compared with scenarios handling additional volume Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) Base Case (2.5%) Base Case (0%) Tranship 2.5% (5%) Base Case (0%) Tranship 10% (7.5%) Base Case (0%) Tranship 10% (10%) Given the productivity and turnaround constrains, it can be concluded that a 7.5% additional volume can be handled in the deep-sea terminal through integration of an inland port. This is an increase of 5 percent points compared with the maximum throughput volume that can be handled in the base case. This number can potentially be increased when nominating even more workforce, reducing the productivity of the gangs and the ship turnaround time. This however is not researched in this thesis. Growth potential of real time sharing of the berth window Figure 39 shows the effect of real time sharing of the berth window and handling additional volume in the deep-sea terminal on the ship turnaround time. It shows that when 2.5% additional volume is handled in the base case, the average turnaround increases. Statistically speaking there is no difference in the in the average ship turnaround of the base case compared with the base case where 2.5% additional volume is handled. The p-value of this statistical test is reported in Table 30. When the berth window is shared 5% additional volume can be handled in the container terminal, without violating the turnaround time constraint. Also in this case the average is higher than the average turnaround time of the base case where no additional volume is handled. This difference is again not different statistically speaking. 68

84 Figure 39 - Growth potential created by real time sharing of the berth window given the turnaround time constraint Table 30 - Tukey multiple comparisons of means turnaround with different volumes and BW sharing Scenario Adjusted P-value Base Case (0%) Base Case (2.5%) Base Case (0%) BW sharing Base Case (2.5%) BW Sharing 1 Figure 40 shows the effect of real-time sharing of the berth window and the handling of additional volume on the productivity of the gangs. It is concluded that handling an additional 5% when sharing the berth does also not violate the gang productivity constraint. It can be seen that when 2.5% addition volume is handled without sharing the berth window the productivity is bit lower compared to the base case with 0% additional volume. As seen in Table 31, the difference is not statistical significant. In a scenario where 5% additional volume is handled with real time sharing of the berth window, the productivity of the gangs is statistically significant higher compared to the base case without additional volume. Figure 40 - Growth potential created by real time sharing of the berth window given the gang productivity constraint 69

85 Table 31 - Tukey multiple comparisons of means gang productivity with different volumes and BW sharing Scenario Adjusted P-value Base Case (0%) Base Case (2.5%) Base Case (0%) BW sharing Base Case (2.5%) BW Sharing Given the ship turnaround and gang productivity constrains it can be concluded that, compared with the current volume, 5% additional volume can be handled in the deep-sea terminal when sharing the berth window real time. This is 2.5% more than the maximum number of volume that can be handled in the base case. This number can potentially be increased when nominating even more workforce, reducing the productivity of the gangs and the ship turnaround time. This is not in the scope of this thesis. Growth potential of combining both alternatives Figure 41 shows the impact of implementing an inland port and sharing the berth window on the average turnaround time given a percentage of additional handled volume. Compared to Figure 37 it shows that 5% additional volume at the deep-sea terminal can be handled with only real time sharing of the berth window. In this scenario the average turnaround time is slightly higher, but statistically speaking there is no difference. It also shows that when the berth window is shared and 10% of transhipment containers is stored at the inland port, 10% additional volume can be handled given the turnaround time constraint. In this case the average turnaround time is also slightly higher, but again statistically speaking there is no difference. The p-values of the means of the scenarios with additional volume compared to the mean turnaround time of the base case are reported in Table 32. Figure 41 - Growth Potential inland port and berth window sharing Table 32 - P-values: Means ship turnaround of scenarios with additional volume compared with the base case Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) Base Case (2.5%) Base Case (0%) BW Sharing (5%) Base Case (0%) Tranship 10% (7.5%) Base Case (0%) Tranship 10% + BW sharing (10%)

86 Figure 42 shows the productivity of the gangs for the identified scenarios that can handle additional volume given the turnaround constraint. The p-values of comparing the mean gangs productivity across the most important scenarios are reported in Table 33. It shows that all scenarios do not violate the productivity constraint. The improvement in productivity could potentially be converted into a lower turnaround time in the terminal, when more gangs are nominated. This could enable more growth potential. It can be concluded that in this case a maximum of 10% additional volume can be handled without violating the turnaround time and productivity constraint when exchanging 10% of the transhipment containers and sharing the berth window real time. This is 7.5% percent point more than maximally could be handled in the base case. Figure 42 - Growth potential inland port and berth window sharing Table 33 - P-values: Means gang productivity of scenarios with additional volume compared with the base case Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) Base Case (2.5%) Base Case (0%) BW Sharing (5%) Base Case (0%) Tranship 10% (7.5%) Base Case (0%) Tranship 10% + BW sharing (10%) Conclusion & discussion of growth potential experiments The results of these experiment show that in the studied case 2.5% additional volume can be handled without violating the ship turnaround time and gang productivity constrains. 5% additional volume can he handled by either real-time sharing of the berth window or by integration of an inland port where import, export and 2.5% of the transhipment containers are handled. These alternatives enable thus a growth potential of 2.5% on top of the current growth potential of the base case. It is interesting to compare the results of sharing of the berth window with integration of an inland port. Integration of an inland port is an alternative that requires high investments in the development of the inland port. Also there are additional transport costs and handling costs of the containers at the terminal, especially when storing long-stay transhipment containers at the inland port. Sharing of the berth window is mainly aimed at reducing waste inside the transport chain. Compared to the inland port 71

87 there are hardly any investment or additional operational costs. Real-time sharing of the berth window is mostly retained by the willingness of the terminal to share their information and the carrier to act upon it. These results show that when these social blockers are removed, waste can be reduced in both their operations. Inland ports are widely discussed in scientific literature. No literature was found that estimates the growth potential of a deep-sea transhipment terminal trough integration of an inland port. The percentage of additional volume that can be handled through integration the inland port is important to evaluate the financial feasibility of an inland port. These results provide insights in the growth potential of an inland port given different exchange strategies. 7.5 Sensitivity Analysis This section contains the results of two sensitivity analyses. The first is a global analysis which determines towards which uncertain variable the model is most sensitive. This is discussed in section In section the robustness of the growth potential achieved by integration of an inland port or sharing the berth window is tested against the most uncertain estimations. Global sensitivity model Figure 43 shows the sensitivity of the terminal turnaround time estimated by the model to input parameters that are considered constant over the different experiments. It shows that the model is most sensitive to variables that directly affect the performance of the yard. The model is more sensitive to a decrease of productivity in the yard than to an increase of the productivity in the yard. This is caused by the fact that the terminal enters a state where its throughput capacity is too low for the volume that needs to be handled. When the replication time would go to infinity, the turnaround time will also go to infinity in this state. The sensitivity towards the yard productivity parameters is not concerning as it as expected. It shows that the model is sensitive towards the productivity in the yard, which is one of the main requirements of the model (see section 4.1). The most uncertain estimation, the relationship between the yard occupancy and the RTG productivity, has the least influence on the performance of the model of the tested variables. These results strengthen the idea that the model is useful for this research. Figure 43 - Tornado chart sensitivity simulation model7.5.2 Robustness of the results 72

88 Robustness of the results In this section the robustness of the achieved growth potential through integration of an inland port or improved information exchange is tested. It chosen to test the design of the inland port that enables the handing of 5% additional volume as it enable comparison with sharing of the berth window which also enables 5% additional volume. These scenarios are compared with the base case to evaluate if they still have a growth potential of 5% when the uncertain variables are changed. This way their robustness against these uncertainties can be assessed. Each experiment is discussed separately. In section a general conclusion about the robustness of the growth potential achieved by both alternatives is drawn Robustness against the number of RTGs per STS crane Figure 44 shows that when an additional 0.5 RTG per STS crane is available, the ship turnaround of both scenarios is not statistically different compared to the base case. In the scenario where the berth window is shared, the mean turnaround time is roughly the same as in the base case. In the scenario with an inland port and additional volume, the mean turnaround is slightly higher, although the difference is not significant statistically speaking. Table 34 reports the corresponding p-values. Figure 44 - Robustness against a high number of used RTGs per STS crane Table 34 - P-values: robustness against high number of used RTGs per STS crane Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) BW Sharing (5%) Base Case (0%) Tranship 2.5% (5%) Figure 45 shows that when 3 instead of 3.5 RTGs per STS crane are available, sharing of the berth window with 5% additional volume has a higher mean of the turnaround time compared to the base case. The difference between both means is statistically significant. The mean in the scenario where 73

89 there is an inland port is also higher compared to the base case, although the difference is not statistically significant. Table 35 shows the corresponding p-values. Figure 45 - Robustness against a low number of RTGs per used STS crane It can be concluded that the results of integration of an inland port are robust against the uncertainty about the number of RTGs that are used per STS crane. The results of additional volume that can be handled through sharing of the berth window is not robust against the uncertainty in the number of RTGs used per STS crane. When 3 instead of 3.5 RTGs are used per STS crane, sharing of the berth window does not enable the handling of 5% additional volume without increasing the ship turnaround time. Table 35- P-values: Robustness against a low number of RTGs per used STS crane Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) BW Sharing (5%) Base Case (0%) Tranship 2.5% (5%) Robustness against productivity of the RTGs Figure 46 shows that when the productivity of the RTGs is 20% higher, the mean of the turnaround time when sharing the berth window decreases. In this case the mean of the turnaround when an inland port is integrated, slightly increases. Statistically speaking, there is no difference in the mean turnaround time of both scenarios compared to the base case. The corresponding p-values can be found in Table

90 Figure 46- Robustness against a high productivity of the RTGs Table 36 - P-values: Robustness against a high productivity of the RTGs Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) BW Sharing (5%) Base Case (0%) Tranship 2.5% (5%) Figure 47 shows that both the variance as the mean increases of the berth window scenario when the productivity of the RTGs is decreased with 20%. The difference in mean turnaround time is statistical significant. Also the mean and the variance of the turnaround time in the inland port scenario increase compared with the base case. The difference of the means is not statistical significant, although the p-value is very low. The corresponding p-values are reported in Table 37. Figure 47 - Robustness against a low productivity of the RTGs 75

91 Table 37 - P-values: Robustness against a low productivity of the RTGs Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) BW Sharing (5%) Base Case (0%) Tranship 2.5% (5%) It can be concluded that the 5% additional volume that can be handled through integration of an inland port is robust against the uncertainty in the productivity of the RTGs. It must be noticed that the increase in ship turnaround time in the scenario where the productivity of the RTGs is low, is almost statistical significant. The growth potential of 5% of sharing of the berth window is not robust against the uncertainty of the productivity of the RTGs. When the productivity is low, the turnaround time in this scenario is statistically significant higher than in the base case Robustness against share of 40s Figure 48 shows that both the mean as the variance of both the berth window and inland port scenario increases compared to the base case when the share of 40s is increased with 20%. The mean of the ship turnaround time of both scenarios do not differ statistically significant from the base case. The corresponding p-values are reported in Table 38. Table 38 - P-values: Robustness against a high share of 40s Figure 48 - Robustness against a high share of 40s Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) BW Sharing (5%) Base Case (0%) Tranship 2.5% (5%) Figure 49 shows that when the number of 40s decreases, the mean and variance of the ship turnaround in the berth window scenario increases compared to the base case. The difference in the mean is statistical significant. The mean and variance of the turnaround in the inland port scenario also increases compared to the base case, although the difference in mean is not statistical significant. The corresponding p-values can be found in Table

92 Table 39 - P-values: Robustness against a low share of 40s Figure 49 - Robustness a low share of 40s Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) BW Sharing (5%) Base Case (0%) Tranship 2.5% (5%) Based on the results it can be concluded that the growth potential estimated for the sharing of the berth window is not robust against the uncertainty in the number of 40s. The mean ship turnaround increase as the share of 40s increases with 20%. This is explained by the fact that more moves have to be made in the yard as the RTGs do not perform twin moves. The yard is constraining the performance of the STS cranes more when additional volume is handled. Sharing of the berth window does not reduce this bottleneck. This bottleneck in reduced by integration of the inland port. Therefore the growth potential of the inland port is robust against the uncertainty of the share of 40s in the container split Robustness against dwell time Figure 50 shows that the mean and variance of the turnaround time increases in the sharing of the berth window scenario when the dwell time of containers is 27% higher. The difference in the mean is statically significant. In the inland port scenario the mean and variance of the turnaround time is higher compared to the basecase, altough in the difference is not significant statistically speaking. The correpsonding p-value can be found in Table

93 Figure 50 - Robustness against a high dwell time of the containers Table 40 - P-values: Robustness against a high dwell time of the containers Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) BW Sharing (5%) Base Case (0%) Tranship 2.5% (5%) Figure 51 shows that when the dwell time of container is 27% lower, the mean turnaround time of in the sharing of the berth window scenario is statistically significant lower. The mean turnaround time in the inland port scenario is slightly higher, but the difference is not statistically significant. The corresponding p-values are reported in Table 41. Figure 51 - Robustness against a low dwell time of containers 78

94 Table 41 - P-values: Robustness against a low dwell time of the containers Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) BW Sharing (5%) Base Case (0%) Tranship 2.5% (5%) The growth potential of the berth window is not robust against uncertainty regarding the dwell time of the containers. This is explained by the relationship between the dwell time and the yard occupancy. A higher dwell time of containers increases the yard occupancy, decreasing the productivity in the yard. Because of this bottleneck, the turnaround time of ships increases when more volume must be handled. Sharing of the berth window does not remove this bottleneck, decreasing the number of additional volume that can be handled. Integration of an inland port does remove this bottleneck partly. Therefore the growth potential estimated for the inland port scenario is robust against the uncertainty of the dwell time. It is notable that in the case the dwell time is 20% reduced, the turnaround time in the berth window sharing scenario is significantly lower. This shows that sharing of the berth window improves the ship turnaround time, when the STS cranes are not constrained by the performance of the yard Robustness against relationship yard occupancy and handling time multiplier Figure 52 shows the average ship turnaround time across the three scenarios when the impact of the yard occupancy on the handling time multiplier is significantly lower. The mean turnaround time in the berth window sharing scenario decreases compared to the base case. The mean turnaround time in the inland port increases slightly. Both differences are not statistically different compared to the base case. The corresponding p-values are reported in Table 42. Figure 52 - Robustness against a low impact of the yard occupancy on the handling time multiplier Table 42 - P-values: Robustness against a high impact of the yard occupancy on the handling time multiplier Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) BW Sharing (5%) Base Case (0%) Tranship 2.5% (5%)

95 Figure 53 shows the ship turnaround time of three chosen scenarios when the impact of the yard occupancy on the handling time multiplier is significantly higher. It shows that in the scenario where the berth window is shared, the mean turnaround time is statistically significant higher compared to the base case. In the inland port scenario, the turnaround time is also higher, although not statistically significant. The p-values are reported in Table 43. Figure 53 - Robustness against a high impact of the yard occupancy on the handling time multiplier Table 43 - P-values: P-values: Robustness against a high impact of the yard occupancy on the handling time multiplier Compared Scenarios (additional volume) Adjusted P-value Base Case (0%) BW Sharing (5%) Base Case (0%) Tranship 2.5% (5%) The growth potential of sharing of the berth window is not robust against the uncertainty about the impact of the yard occupancy on the ship turnaround time. This is because when this impact is high, the yard will be a bottleneck in the performance of the STS cranes, increasing the turnaround time. When more volume is handled, the yard occupancy increases. Sharing of the berth window does not remove this bottleneck, so no additional volume can be handled. Integration of the inland port does remove this bottleneck, so additional volume can be handled without increasing the ship turnaround time. It is interesting to see that the p-value of the inland port scenarios is much higher when the impact of the yard occupancy is high than when its low. It shows the impact of integration an inland port on the ship turnaround time increases as this relationship is stronger Conclusion and discussion of robustness of the results In this section the robustness of growth potential of sharing of the berth window and integration of an inland port are tested against uncertain estimations. The uncertainty of these estimations is discussed in section The variables all mainly affect the performance in the yard. The results of the different experiments are consistent with each other. When changing the variables in a way that reduces the productivity of the yard further, the growth potential of sharing the berth window decreases. This is for example the case when the dwell time increases. The yard occupancy rises which lowers the productivity in the yard, constraining the additional volume that can be handled by the 80

96 terminal. Sharing of the berth window does not remove this bottleneck. Therefore its effect on enabling growth potential is limited. Integration of the inland port does remove this bottleneck partly. Its estimated growth potential is robust against the uncertainty in the estimations. Testing the sensitivity towards the relationship between the yard occupancy and the handling time multiplier in the yard, shows that the effect of integrating an inland port reduces as this relationship becomes weaker. When this relationship is too weak, doing additional moves because of storing transhipment containers at the inland port will not pay-off. 81

97 8 Conclusion Now that the results are analyzed, the research question from chapter 1 can be answered. This is done in section 8.1. The main research question is answered in section The findings of this research are discussed based on current literature analyzed in section 8.2. Section 8.3. discusses the scientific and societal relevance of this research. The limitations of both the case study as the used conceptual and simulation model are discussed in section 8.4. Suggestion for further research are presented in section 8.5. This thesis concluded with a personal reflection on the research process, reported in section Answer to research questions The aim of this research is to deepen the understanding of how container transhipment terminals can increase the number of volume they handle without increasing the ship turnaround or reducing the gang productivity in the terminal through integration of an inland port or improved information exchange. This research uses the design science research methodology and a discrete event simulation model, expert interviews and literature research as main research methods. A case study on a transhipment terminal in the port of Algeciras is used to provide the context of the research problem. The main research questions is supported by multiple sub questions. These sub question will be answered first in section to section The main research question is answered in section Answer to sub question 1 What information exchange alternatives will be most effective on reducing the ship turnaround time in the terminal and increasing gang productivity and will have most support among the transport partners in the container transport chain? In section it is explained that information in the container transport chain can be divided into four types of data: future event data, status data, historical data and fixed data. Based on interviews it is decided that future event information has most potential to reduce the ship turnaround time and improve productivity of the workforce in a container terminal. Based on interviews and scientific literature multiple future event information exchange alternatives are identified in section These alternatives are improving the accuracy of the estimated time of arrival, real-time sharing of the berth window, sharing predictions on the availability of import containers in the terminal and implementation of a truck appointment system. In section it is discussed that the support towards an information exchange alternative is dependent on whether the proposed alternative is about sharing new information or about improving the quality of information currently exchanged. Organizations view the disclosure of additional information as a loss of power and competitive advantage (Li & Lin, 2006,p1646). Improving information that is currently exchanged has therefore more support among transport partners than sharing new information. The expected support is also dependent on the financial dependency between the transport partners that should implement the information exchange alternative. Transport partners within the transport container chain not only cooperate, but often also compete with each other. Therefore additional information exchange can only be achieved among stakeholders that already have established a certain level of trust. This level of trust is aligned with the financial dependency between the transport partners. The carrier is paying the terminal operator, while the truck companies are not. Improving information exchange on the seaside of the terminal therefore has more support among the related transport partners than the landside. 82

98 Improvement of the accuracy of the estimated time of arrival of ships and real time sharing of the berthing window are both alternatives that require support from the carrier and the terminal operator. Here the trust relationship is present. Improving the accuracy of the ETA and the timeliness of the berth window are both alternatives aimed at improving the quality of information instead of sharing new information. It is therefore concluded that improving the accuracy of the ETA and sharing the berth window real time are two information exchange alternatives that will have effect on the terminal and will have support among the implementing transport partners. Answer to sub question 2 What are the main mechanisms explaining how ship turnaround time and gang productivity are affected by the integration of and inland port and improved information exchange? In section it is described that the inland port influences the productivity and performance of the deep-sea terminal trough decreasing the yard occupancy. A high yard occupancy causes unproductive moves in the yard, decreasing the productivity of the yard equipment (Kemme, 2012, p570; Saanen & Dekker, 2006, p82). The performance of the yard equipment has a direct effect on the ship turnaround time (Hanh Dam; Le-Griffin & Murphy, 2006, p3). Inland ports are used to decrease the yard occupancy of a deep-sea terminal and to increase the yard productivity (Roso, Woxenius, & Lumsden, 2009, p344). In this research also the storage of long-stay transhipment containers at the inland port is researched. In contrary to import and export containers transhipment container cause additional (productive) moves in the deep-sea terminal. In this case it is uncertain whether the increase in productive moves can be justified by the decrease of unproductive moves in the terminal. A more accurate ETA of ships at the terminal will affect the productivity of the terminal in two ways. Terminal operators use the estimated time of arrival of ships to predict the workload for each work shift. Wrong workload predictions cause under or overmanning of the terminal (Fancello et al., 2011, p143). Overmanning hurts the productivity of a terminal, while undermanning decreases the performance in the terminal. In the model a gang nomination algorithm is present which nominates gangs based on the ETA of inbound ships. The ETA of ships is an important input variable for the allocation of cranes to ships (Bierwirth & Meisel, 2015, p618, figure 5). Deviation between the estimated time of arrival and the actual time of arrival cause the need for the terminal to continuously reschedule the allocation of cranes, resulting in a poorer performance compared to the baseline crane assignment plan (Xu, Chen, & Quan, 2012, p125). This mechanism is captured using two crane allocation algorithms that allocate cranes to ships when ships arrive at the terminal and at the start of each gang shift. These algorithms take the estimated time of arrival, estimate time of departure and the number of cranes discussed in service level agreement as input. Using both algorithms, the effect of the accuracy of the ETA on the ship turnaround time and gang productivity can be studied. Real time sharing of the berthing window enables the carrier to increase or decrease the speed of their ships within certain boundaries based on the predicted availability of the berth and STS cranes. This mechanism is captured with an algorithm that continuously updates the expected berth window based on the current state at the terminal and the estimated time of arrival and estimated time of departure of inbound ships. Based on this berthing window, the terminal operator advices the carrier to decreasing the speed when terminal resources are not available on arrival. This way the carrier saves bunker, resulting in less transport costs. When terminal resources are estimated to be idle during an certain period, the terminal operators advices the carrier to speed their ship. This will improve gang productivity which can result in lower handling rates at the terminal for the carrier. 83

99 Answer to sub question 3 What is an effective high level design of integration of an inland port dedicated to decreasing the ship turnaround time in a transhipment container terminal? Using a simulation model in the context of a transshipment terminal in Algeciras, multiple insight in the integration of inland port aimed at decreasing the ship turnaround time of a transhipment terminal are obtained. The simulation model is focused on the relationship between the yard occupancy and handling time of a container in the yard of the deep-sea terminal. The handling time is dependent on the aggregate state of the yard, instead of the exact location of a container. It is assumed that the gate of the terminal and the horizontal moves within the terminal are not constraining the performance of the yard equipment. Handling of reefers is excluded from the model. The insights obtained from experimentation with the simulation model are translated to a set of design principles for integration of inland ports in the operations of transshipment terminals. When the inland port only handles import and export containers, moving customs facilities to the inland port does not improve the ship turnaround time. When also storing transhipment containers in the inland port, customs facilities should be present to avoid paying import duties on the transhipment container. Import and export containers should always be part of the exchange strategy as these containers do not produce additional moves in the yard when implementing an inland port. Storing long-stay transhipment container improves the ship turnaround time of a transhipment terminal. In the studied case the ship turnaround time could be improved by storing between 0% and 10% of the transhipment containers at the inland port. When too much transhipment containers at the inland port are stored, it increases the ship turnaround time again. The underlying mechanism is as follows. Reducing the yard occupancy results in less unproductive moves. This removes the constraints of the yard productivity on the productivity of the STS cranes. The marginal impact of storing more transhipment containers at the inland port decreases as their effect on the average dwell time in the deep-sea terminal becomes less, while still producing two additional moves. At some point the productivity of yard equipment will decrease again, constraining the STS cranes. At this point the ship turnaround time will increase. Downtime in the transport link of around 50% does not decrease the ship turnaround time. This was learned by comparing the train with a truck alternative. In the train alternative containers could be moved from the terminal 24 hours a day, while in the truck alternative containers could only be moved 12 hours a day to the inland port. The differences in turnaround time in of these two scenarios was not statistically significant. Using dedicated stacks in the deep-sea terminal for containers destined for the inland port will decrease the ship turnaround time of the terminal. These dedicated stacks are meant to make the handling of containers, which will be moved to the inland port, in the yard of the deep-sea terminal more efficient. These stack reduce the capacity in the main yard for all other containers. The loss of capacity in the main yard has a bigger effect on the ship turnaround time than the time that is won by handling the container more efficiently. Answer to sub question 4 What is the effect of improving information exchange between transport partners on the ship turnaround time and the gang productivity in a container transhipment terminal? Based on multiple experiments with a simulation model insights have been generated on the effect of improving the accuracy of the estimated time of arrival and real time sharing of the berthing window 84

100 on the ship turnaround time and the productivity of the gangs in a deep-sea transhipment terminal. The most important assumptions regarding the modelling of information exchange are certain level of inflexibility in nominating gangs and allocating cranes. Once a day gangs for the next four shifts of six hours each have to be nominated. The allocation of cranes is time invariant during servicing of the ships. This means that for most ships, the number of cranes that is servicing the ship is constant over the different gang shifts, limiting optimization of the terminals resources. Improving the accuracy of the ETA improves the productivity of the gangs as there is better alignment between the nomination of gangs and the workload at the terminal. A perfect ETA would increase the productivity in the terminal by around 2.5% achieving a similar decrease in the number of workforce needed per day in the terminal. Real time sharing of the berthing window improves the productivity of the gangs also with around 2%. The impact of real time sharing of the berthing window on the productivity of the gangs declines as the ETA is more accurate. Sharing the berthing window real-time and improving the ETA combined, increase the productivity more than these means separately. The effect of both measures on the productivity of the gangs decline when the volume handled in the terminal increases. Real time sharing of the berthing window reduces the ship turnaround time in the terminal trough reducing the waiting time of ships. An improved productivity of the gangs does not decrease the turnaround time as less gangs are nominated to fulfil the predicted workload. In the studied case a decrease of 4% to 5% in the ship turnaround time is achieved by sharing of the berthing window. This decrease is not statistically significant. Answer to main research question How can the volume handled in a transhipment terminal be increased trough implementation of an inland port and improved information exchange without increasing ship turnaround time and decreasing gang productivity? Supported by the answers of the sub question an answer to the main research question can be formed. This research focusses on two main ways of increasing the volume that can be handled in container transhipment terminals without hurting the ship turnaround time and gang productivity. First improving the accuracy of the estimated time of arrival and real time sharing of the berth window have been selected as information exchange alternatives that have most effect on the productivity and performance of a container terminal and will have most support among the implementing transport partners. A more accurate ETA will improve the productivity of the terminal by reducing the overmanning and undermanning the terminal and by improving the allocation of the STS cranes. Real time sharing of the berthing window reduces the waiting time of ship, because ships get notified to decrease their speed in order to save bunker costs when the berth is not available upon the expected arrival time. When the berth is available earlier than the arrival of the ship, ships will get notified to increase speed, improving the utilization of the terminals equipment and workforce, resulting in less handling costs at the terminal. An inland port can improve ship turnaround time and gang productivity of the deep-sea terminal by storing and handling containers at the inland port. This reduces the yard occupancy, improving the productivity of the yard equipment and the STS cranes. Next to import and export containers longstay transhipment containers can be stored at the inland port to improve the performance of the terminal. Customs facilities should only be implemented at the inland port when transhipment containers are stored there. No dedicated stacks in the yard of the deep-sea terminal for transport towards the inland port should be implemented as it will decrease ship turnaround time. 85

101 The gain in ship turnaround time and gang productivity at the current volume can be used to increase the volume that is handled in deep-sea terminal. In this specific case an inland port enables the handling of more volume at the deep-sea terminal than real time sharing of the berth window. Both measures can be combined to improve the throughput capacity even further. In the base case a maximum of around 2.5% additional volume can be handled given the ship turnaround and gang productivity constraints. When sharing the berth window around 2.5% additional volume can be handled inside the terminal, providing a growth potential of 5% compared to the current volume handled. A similar results can be achieved by storing 2.5% of the transhipment containers at the inland port. When the yard is constraining the performance of the STS cranes more, the effect of integrating an inland port becomes higher. In this case the effect of sharing the berth window becomes smaller. An inland port can enable the handling of 5% additional volume compared to the base case when 10% of the transhipment containers is stored at the inland port. When combing an inland port with sharing of the berth window up to 7.5% additional volume can be handled without hurting gang productivity and ship turnaround time. These results are case specific and are among others dependent on dwell times, productivity of the yard equipment and the container split in the terminal. 8.2 Discussion This research shows how the number of volume handled in a transshipment terminal can be increased by integration of an inland port and through additional information exchange without increasing the ship turnaround time or decreasing the gang productivity at the deep-sea terminal. As discussed in chapter 2, inland ports already have been studied extensively, but not in the context of an inland port controlled by a transshipment terminal. The results of this study show that the way inland ports should be integrated differs for transshipment terminals compared to general terminals. Transhipment terminals are characterized by handling a low share of import and export containers. General inland port literature like Veenstra et al (2012) and Roso et al (2009) are mainly discussing inland ports where import and export containers are handled. This study shows that the change in container split at the deep-sea terminal influence the validity of recommendations made in general inland port literature. Roso et al (2009, p. 344) see moving custom clearance as one the opportunities of improving the productivity of deep-sea terminal. This study show that this does not hold for transhipment terminals when import and export containers will be handled at the inland port, because their volume is to limited. This study also shows that storing empty containers at the inland port is not always an useful strategy. In addition to current general inland port literature, this study adds the concept of storing long-stay transhipment containers at an inland port. As long as the dwell time of the stored container is high enough, the additional moves inside the deep-sea terminal can be justified by the positive effect on the yard occupancy. This is in particular interesting for transhipment terminals, but can be extended to any container terminal with long stay transhipment containers. Operations research regarding the container transport chain often assume perfect information. A clear example is research on the berth allocation problem and the crane allocation problem that assume the ETA to be exact the ATA (Bierwirth & Meisel, 2015). This study shows that this assumption does not hold. Estimated time of arrivals differ from the actual time of arrivals. This research shows that inaccurate ETAs reduce productivity of the gangs in the deep-sea terminal. This support research by Fancello et al (2011) that the accuracy of the ETA should be improved in order to utilize workforce at the terminal better. The results of sharing the berth window are in line what is reported by Lang & Veenstra (2010). Both show benefits for both the terminal operator as the carrier. Lang & Veenstra (2010) focus in their study on the minimisation of a general costs function. Their study is mainly from the perspective of a the carrier, using a deterministic model for the terminal. This study extends their finding by focussing more on the operational processes inside the terminal. It shows that the terminal 86

102 operator can increase the productivity of the gang as sharing of the berth window enables him to partly correct mismatches between the workload and available resources. This study also shows that the impact of sharing the berth window is dependent on the accuracy of the ETA, as less mismatches are created when the ETA is more accurate. In this study the effect of information exchange and an inland port is studied using the same case data and the same simulation model. This way a more useable comparison can be made about the effects of both measures than by comparing inland port studies with information exchange studies. This study shows that 2.5% additional volume can be handled in the deep-sea terminal by either integrating an inland port and exchanging import, export and 2,5% of the transhipment containers or by sharing of the berth window. The growth potential that can be achieved by implementing both measures is dependent on how strong the STS are currently being constrained by the yard. A bigger bottleneck in the yard increases the growth potential of integration of an inland port. It reduces the growth potential of the berth sharing alternative. Integration of an inland port is expensive as additional infrastructure needs to be build. However less support is needed from partners in the transport chain to integrate an inland port. Sharing of the berth window is much cheaper than integrating an inland port. However it is more dependent of the willingness of the transport partners to share their information and act upon it. 8.3 Relevance of thesis The findings it this research are both relevant for science as for society. In the scientific relevance is explained. In the societal relevance is discussed. Scientific Relevance The main scientific contribution of this this thesis is twofold. The first is the identification of information exchange alternatives in the container transport chain that are socially feasible and estimating their effects on the performance and productivity of a container terminal. The second main scientific contribution is a list of design principles that help to design inland ports dedicated to improving the performance and productivity of a nearby transhipment container terminal. The amount of literature on information exchange in the container transport chain is limited. When discussing information exchange in the container transport chain, most literature focusses on a single information exchange alternative and ignores the feasibility from a social perspective. This thesis gives an overview of different information exchange alternatives, both from literature as from expert interviews. The alternatives are clustered based on the estimated impact of the information type and based on their feasibility from a social perspective. It is concluded that improving the accuracy of the estimated time of arrival of ships and real time sharing of the berth window are the two alternatives which impact the performance and productivity at the terminal and are socially feasible. The effect of these two alternatives are more thorough estimated using a simulation model. Improving the accuracy of the ETA and real time sharing of the berth window improves productivity and performance in a terminal considerably. However the effect of the currently achieved improvement of the accuracy of the ETA is literature of 25% is limited. Literature describes inland ports as a mean to both improve both the hinterland accessibility as improve the productivity in the terminal. The most applied strategy is handling import and export containers at the inland port. There is a knowledge gap on how an inland port should be designed that is dedicated to improving the productivity of a transhipment container terminal and which can be controlled by this terminal. This thesis shows that the high level design of an inland port dedicated to improving the productivity of a transhipment terminal is different from the ways inland ports are normally designed. The most notable differences are that when designing for a transhipment terminal 87

103 customs should not be moved to the inland port when only handling import and export containers there. Also, storing long-stay transhipment containers at the inland port can improve the productivity and performance of the terminal significantly. Societal Relevance From a societal perspective the results of the thesis are interesting. A lot of essential information is already exchanged between transport partners, but transport partners in the container transport chain are restrained in improving their information exchange further. This thesis shows that considerable improvements in the productivity of a container terminal can be achieved when transport partners improve the quality of information they already exchange, with transport partners they already trust and cooperate with. Other transport partners in the transport chain can benefit from increased productivity in the terminal as it enables them to handle more volume or profit from lower rates. The results of this thesis can be a starting point for companies to discuss the improvement of information exchange, especially about real-time sharing of the berth window. Congested yards of transhipment container terminal caused by spatial development problems is a common problem. This thesis provides insights on how inland ports can be designed specifically for transhipment terminals. These insight can be used a starting point to design the inland port more detailed. 8.4 Limitations This thesis presents extensive research on the effects of improved information exchange and the implementation of an inland port on the productivity and the performance of a container transhipment terminal. However some limitations to this research apply. Section discusses limitations regarding the case. Limitations regarding the conceptual and simulation model are discussed in section Limitations regarding the case In this study a case study on a transhipment terminal in the port of Algeciras is used. Despite the fact that this is a typical transhipment container terminal, some remarks have to be made on how well insights from this case study can be generalized to any transhipment terminal. The studied terminal is a RTG equipped container terminal, meaning that they are very dependent on workforce compared to automated RMG equipped terminals. In contrary to RTG terminals, the capacity of the yard equipment in a RMG terminal is more or less constant as no gangs need to be nominated. In this case more RMGs are available per STS crane when the berth is not fully utilized. This reduces the effect of a full yard on the performance of the berth. In this case the implementation of an inland port and improved information exchange could have less effect on the productivity and performance in the terminal. The second remark is about the gang nomination system in Algeciras. Because of the inflexibility of this system the estimated time of arrival is more important compared to terminal where the allocation of workforce is more flexible. This gang nomination system is common in Spain, but the allocation of workforce might be more flexible in other transshipment terminals in the world. These container terminals would have less benefits from an improved accuracy of the ETA. Limitations regarding the conceptual and simulation model Despite the usefulness of both the conceptual as the simulation model some limitations apply. As discussed in the validation session of the model, the yard is modelled on a very high level. This makes the time needed to take a container from the stack dependent on the aggregated state of the yard instead of the state of the specific container. Extending the model in such a way that the unstacking time is dependent on the state of the specific container could make it more useable. The second 88

104 limitation is the limited way costs are taken into account in this study. The handling costs of a container is only represented by the productivity of the gangs. As a consequence all costs made outside the deep-sea terminal are neglected. It is assumed that these costs do not influence decision making in the system. In the real system this is not the case. An example is deciding whether a ship should speed up when it knows that the berth is available. In reality this decision would be cost driven. It would be dependent on how much the additional bunker would cost compared the amount of money that can be saved at the terminal or even further in the container transport chain. This same applies for integration of the inland port. Even while it outside the scope of this research, it is important to be aware of the fact the choice of implementing an inland port will be very dependent on its costs. If the total costs of handling the containers at the inland port and for transport between the inland port and the deep-sea terminal are more than the additional revenue that can be made by handling more volume in the deep-sea terminal, an inland port will never be implemented. Despite that this model does not estimates these costs, it is a useful tool to estimate how much more volume can be handled in the deep-sea terminal. This is valuable information in calculating the financial feasibility of integrating an inland port. 8.5 Future work Based on the limitations of this research in the previous sections multiple suggestion for future work are made. The first is to extend this research with another case study, preferably with an automated RMG equipped terminal that has more flexibility in nominating their gangs for the STS cranes. It would be interesting to see the differences and similarities in results between both cases. Another interesting topic would be to extend the model with a more extensive representation of the yard. In this case more stacks should be modelled in the terminal and a representation of the TOS should be added to decide where containers should be stacked. Another interesting topic for research would be to further research joint operational planning by sharing of the berth window. Instead of simple decision rules for the carrier based on the shared berth window, a cost optimization function should be developed that takes the cost of a network of ships and terminals into account when advising inbound ships on their speed. Based on research with this cost function, incentive schemes can be developed that can smoothen the way for implementation of joint operational planning between carriers and terminals. The last suggested research topic is to research the effect of more performance and productivity at a transhipment terminal from container chain perspective instead of looking only at the effect inside the terminal. During this research it was discussed with multiple experts that improvement of performance at the terminal could bring benefits to certain partners in the container chain. Being able to monetize these effects could enable a terminal to increase its performance in the chain. This research would need considerable support from all partners in the transport chain in order to provide the right data to make this research a success. 8.6 Reflection I will conclude this thesis with some personal remarks on the process of this research. Scoping and demarcating this research was hard. The starting point was to research information exchange between an inland port and a deep-sea terminal. During my research these two parts drifted away from each other. Resulting on information exchange on the seaside of the terminal and integrating an inland port on the landside of the terminal. In retrospect, it would have saved me a lot of work if I had chosen either one of this topics early on. However, reaching the end of this thesis I believe that there is also benefit from researching both subjects in the same study: the ability to compare the effects of both means more trustworthy. 89

105 During model development, I experienced that operations at a terminal is much more complex than one initially would think. This caused the need to simplify my model more than I was planning to do. Simulation is an well-established approach in the research of container terminals. Complex container terminal models with a high level of detail are built by simulation companies. Compared to these models, my model has a high level of abstraction. Nevertheless I am sure that my model is still a useful tool in studying both the integration of an inland port and additional information exchange. I am even confident that some of the information exchange concept I have added in my model can be a further improvement for those detailed models that already exist. One of the two main topics in this thesis is improving information exchange in the container transport chain. An important finding is that there is little support from transport partners in the transport chain to share their information. Unfortunately, this is something I have encountered during this research. Little detailed information about the case study was available. A clear example is the relationship between the yard occupancy and the performance of the STS cranes. This relationship could be derived from data that is recorded in every terminal and it could have improved the validity of parts of this study. It was not possible for me to acquire this data. I am thankful towards the experts from both Navis as Macomi that helped me overcome this problem. Their estimations truly helped this study forwards. In this research I truly liked exploring the mechanisms behind the integration of an inland port and information exchange in the transport chain by talking to experts, developing a simulation model and analyzing the results. My biggest personal challenge is to document these findings in a clear, concise and scientific way. I have learned very much about the mechanisms that influence the productivity and performance of a deep-sea terminal trough integration of an inland port and improved information exchange. I hope this thesis provides in the transfer of this knowledge to anyone interested. 90

106 References Angeloudis, P., & Bell, M. G. H. (2011). A review of container terminal simulation models. Maritime Policy & Management, 38(February 2015), APBA. (2018). Stats. Retrieved January 15, 2018, from Bierwirth, C., & Meisel, F. (2015). A follow-up survey of berth allocation and quay crane scheduling problems in container terminals. European Journal of Operational Research, 244(3), Borshchev, A., & Filippov, A. (2004). From System Dynamics to Agent Based Modeling. Simulation, 66(11), Retrieved from v04.pdf Carlo, H. J., Vis, I. F. A., & Roodbergen, K. J. (2014). Storage yard operations in container terminals: Literature overview, trends, and research directions. European Journal of Operational Research, 235(2), Dekker, R., Voogd, P., & Van Asperen, E. (2007). Advanced methods for container stacking. Container Terminals and Cargo Systems: Design, Operations Management, and Logistics Control Issues, 586, Dowd, T. J., & Leschine, T. M. (1990). Container terminal productivity: A perspective. Maritime Policy and Management, 17(2), Ducruet, C., & Notteboom, T. (2012). The worldwide maritime network of container shipping: Spatial structure and regional dynamics. Global Networks, 12(3), Fancello, G., Pani, C., Pisano, M., Serra, P., Zuddas, P., & Fadda, P. (2011). Prediction of arrival times and human resources allocation for container terminal. Maritime Economics and Logistics, 13(2), Fransoo, J. C., & Lee, C. Y. (2013). The critical role of ocean container transport in global supply chain performance. Production and Operations Management, 22(2), Fremont, A. (2007). Global maritime networks: The case of Maersk. Journal of Transport Geography, 15(6), Giannopoulos, G. A. (2004). The application of information and communication technologies in transport. European Journal of Operational Research, 152(2), Henesey, L. E. (2006). Multi-Agent Systems for Container Terminal Management. Technology. Blekinge Institute of Technology. Retrieved from e=pdf Holguín-Veras, J., Xu, N., de Jong, G., & Maurer, H. (2011). An Experimental Economics Investigation of Shipper-carrier Interactions in the Choice of Mode and Shipment Size in Freight Transport. Networks and Spatial Economics, 11(3),

107 Imai, A., Nishimura, E., & Papadimitriou, S. (2001). The dynamic berth allocation problem for a container port. Transportation Research Part B: Methodological, 35(4), JOC. (2014). Peel off method gaining momentum as port congestion figher. Retrieved March 15, 2018, from Jones, S. L., Fawcett, S. E., Fawcett, A. M., & Wallin, C. (2010). Benchmarking trust signals in supply chain alliances: moving toward a robust measure of trust. Benchmarking: An International Journal, 17(5), Kemme, N. (2012). Effects of storage block layout and automated yard crane systems on the performance of seaport container terminals. OR Spectrum, 34(3), Kent, J. L., & Parker, S. R. (1999). International containership carrier selection criteria. International Journal of Physical Distribution & Logistics Management, 29(6), Keppel, G., & Wickens, T. D. (2004). Simultaneous Comparisons and the Control of Type I Errors. Design and Analysis: A Researcher s Handbook, Kim, K. H., & Gunther, H.-O. (2007). Container Terminals and Cargo Systems. Berlin heidelberg: Springer. Lang, N., & Veenstra, A. (2010). A quantitative analysis of container vessel arrival planning strategies. OR Spectrum, 32(3), Langen, P. W. De, & Chouly, A. (2004). Hinterland Access Regimes in Seaports. Ejtir, 4(1988), Le-Griffin, H. D., & Murphy, M. (2006). Container Terminal Productivity: Experiences at the Ports of Los Angeles and Long Beach. Container Terminal Productivity, Feb(1), Le-Griffin, H. D., & Murphy, M. (2008). Assessing Container Terminal Productivity, 29. Retrieved from Final Report_0_0.pdf Li, S., & Lin, B. (2006). Accessing information sharing and information quality in supply chain management. Decision Support Systems, 42(3), Lind, M., Haraldson, S., Karlsson, M., & Watson, R. T. (2015). Port collaborative decision making closing the loop in sea traffic management. 14th International Conference on Computer Applications and Information Technology in the Maritime Industries, (2009), Retrieved from Lind.pdf Maloni, M., Paul, J. A., & Gligor, D. M. (2013). Slow steaming impacts on ocean carriers and shippers. Maritime Economics and Logistics, 15(2), Menger, I. (2016). Information Exchange between Deep Sea Container Terminals and Hinterland Parties. TU Delft. Retrieved from uuid:df65f8c2-3c27-43ce-b9a3-768d964eef51 Mohr, J., & Spekman, R. (1994). Characteristics of Partnership Success : Partnership Attributes, Communication Behavior, and Conflict Resolution Techniques Author ( s ): Jakki Mohr and Robert Spekman Stable URL : Accessed : : 92

108 40 UTC Yo. Strategic Management Journal, 15(2), Monczka, R. M., Petersen, K. J., Handfield, R. B., & Ragatz, G. L. (1998). Success Factors in Strategic Supplier Alliances: The Buying Company Perspective. Decision Sciences, 29(3), Monios, J. (2011). The role of inland terminal development in the hinterland access strategies of Spanish ports. Research in Transportation Economics, 33(1), Nguyen, L. C., & Notteboom, T. (2018). The relations between dry port characteristics and regional port-hinterland settings: findings for a global sample of dry ports. Maritime Policy and Management, pp Notteboom, T. (2006). Chapter 2 Strategic Challenges to Container Ports in a Changing Market Environment. Research in Transportation Economics, 17(6), Notteboom, T. E. (2006). The Time Factor in Liner Shipping Services. Maritime Economics & Logistics, 8(1), Notteboom, T. E., & Rodrigue, J. (2005). Port Regionalization : Towards a New Phase in Port Development Port Regionalization : Towards a New Phase in Port, (JULY 2005), Parolas, I. (2016). ETA prediction for containerships at the Port of Rotterdam using Machine Learning Techniques. TU Delft. Retrieved from uuid:9e95d11f-35ba-4a12-8b34-d137c0a4261d Peffers, K., Tuunanen, T., Rothenberger, M. A., & Chatterjee, S. (2007). A Design Science Research Methodology for Information Systems Research. Journal of Management Information Systems, 24(3), Pengfei, Z., Haigui, K., & Li, L. (2006). A dynamic berth allocation model based on stochastic consideration. Proceedings of the World Congress on Intelligent Control and Automation (WCICA), 2, Polman, M., Asperen, E. Van, Dekker, R., & Arons, H. de S. (2003). Modeling Ship Arrival in Ports. Proceedings of the 2003 Winter Simulation Conference, Rodrigue, J., Debrie, J., & Fremont, A. (2010). Functions and Actors of Inland Ports : European and North American Dynamics Functions and Actors of Inland Ports : European and North American Dynamics. Journal of Transport Geography, 18 (4), Roso, V., Woxenius, J., & Lumsden, K. (2009). The dry port concept: connecting container seaports with the hinterland. Journal of Transport Geography, 17(5), Saanen, Y. A., & Dekker, R. (2006). Intelligent stacking as way out of congested yards? Part 1. Port Technology International, 31, Sargent, R. G. (2010). Proceedings of the 2010 Winter Simulation Conference B. Johansson, S. Jain, J. Montoya-Torres, J. Hugan, and E. Yücesan, eds. Simulation, (2001), Stahlbock, R., & VoB, S. (2008). Operations research at container terminals: A literature update. OR Spectrum, 30(1),

109 Steenken, D., VoB, S., & Stahlbock, R. (2005). Container terminal operation and operations research - A classification and literature review. Container Terminals and Automated Transport Systems: Logistics Control Issues and Quantitative Decision Support, Van Asperen, E., Borgman, B., & Dekker, R. (2013). Evaluating impact of truck announcements on container stacking efficiency. Flexible Services and Manufacturing Journal, 25(4), van Schuylenburg, M., & Borsodi, L. (2010). Container Transferium Rotterdam : an Innovative logistic concept. In Port Infrastructure Seminar Veenstra, A., Zuidwijk, R., & Van Asperen, E. (2012). The extended gate concept for container terminals: Expanding the notion of dry ports. Maritime Economics and Logistics, 14(1), Verhoeven, P. (2010). A review of port authority functions: Towards a renaissance? Maritime Policy and Management, 37(3), Vis, I. F. A., & De Koster, R. (2003). Transshipment of containers at a container terminal: An overview. European Journal of Operational Research, 147(1), Wiegmans, B., Van Arem, B., & Behdani, B. (2017). Communication between deep sea container terminals and hinterland stakeholders: information needs and the relevance of information exchange. Maritime Economics & Logistics. Wiegmans, B. W., Hoest, A. Van Der, & Notteboom, T. E. (2008). Port and terminal selection by deepsea container operators. Maritime Policy and Management, 35(6), Wilmsmeier, G., Monios, J., & Lambert, B. (2011). The directional development of intermodal freight corridors in relation to inland terminals. Journal of Transport Geography, 19(6), Xu, Y., Chen, Q., & Quan, X. (2012). Robust berth scheduling with uncertain vessel delay and handling time. Annals of Operations Research, 192(1), Zhao, W., & Goodchild, A. V. (2010). The impact of truck arrival information on container terminal rehandling. Transportation Research Part E: Logistics and Transportation Review, 46(3),

110 Appendices 95

111 A. Background literature on terminal operations This appendix will provide some background literature on the transport chain and terminal operations. In A.1 the container transport chain is defined. A.2 provides an overview of terminal operations. Finally, in A.3 improvements to a container terminal proposed by literature are analyzed. A.1 The container transport chain A container is a large standardized box that is used to transport goods. It protects its cargo inside and standardizes its handling. The container transport chain consists of the different stages in container transport from the point it was loaded at its origin until it is opened again at its final destination(veenstra & Zuidwijk in Menger (2016, p 7). The container is the main technical improvement that contributed to the globalization of world trade (J. P. Rodrigue & Notteboom, 2009). It enabled large productivity gains in the cargo handling in ports and the creation of efficient global transport networks by major shipping lines (Fremont, 2007). Almost all intercontinental transportation of goods is done by ship and a big share of these goods are packaged in containers(fransoo & Lee, 2013). A.2 Terminal Operations Container terminals are used to transship container from one mode of transport to another(vis & De Koster, 2003, p1). This can be discarding a container from a ship and loading it into truck for the hinterland transport. Container terminals that mainly perform transhipment between the landside and seaside are called gateway terminals. These terminals mainly transship import and export containers. Most major shipping lines use certain terminals as hubs in their transport network. In these hubs transhipment from one ship to another ships is a big part of operation (Fremont, 2007, p433). They mainly handle transhipment containers. These terminals are commonly referred to as transshipment terminals. As seen in Figure 54 all container terminals be described as open systems consisting of three parts. On the quayside, also called seaside, ships are discarded and loaded. On the landside trucks and/or trains are operated. In the storage yard the containers are stored. The yard decouples the quayside from the landside (Steenken, Vo??, & Stahlbock, 2005, p6). Figure 54 - Overview container terminal (Kemme, 2012, p561, fig 1) Container terminals can use different equipment and yard layouts. There are two main configurations for the yard lay-out which are classified as the European and Asian layout by (Carlo, Vis, & Roodbergen 96

112 (2014, p414)). The two layout are shown in Figure 55. In the Asian layout blocks are in a parallel position to the quay. In between the blocks are truck lanes. This layout is mostly used in combination with non-automated equipment. The European layout has blocks perpendicular to the quay. The only entry and exit points are at the landside and seaside of the quay. The transport equipment on the seaside and the yard equipment is automated in most container terminals using this layout. The European layout has lower operational costs, more storage capacity and faster crane speeds than the Asian layout. The investment costs of the European layout are higher (Carlo, Vis, & Roodbergen, 2014, p414). Figure 55 - European and Asian yard lay-out (Carlo, Vis, & Roodbergen, 2014, p414) The yard capacity and productivity are dependent on the used yard equipment. The most used types are reach stackers, straddle carriers (SC s), rubber tired gantry cranes (RTG s) and rail mounted gantry cranes (RMG s) (Stahlbock & Vo??, 2008, p3). RMGs can function automatically. These types have different specifications regarding the height of containers they can stack and the amount of bays in a block they can span. The terminology in a yard is explained in Figure 56. RMG s enable terminals to use the European layout decreasing the space between the yard blocks further. Figure Figure 57 shows the storage capacity of different handling equipment. Using RMGs improves the yard capacity almost with 10% compared to using RTGs and 50% compared to SCs. 97

113 Figure 56 - Container terminal yard terminology (Zhao & Goodchild, p328, fig 2) Figure 57 - Yard equipment and their storage capacity (Stahlbock & Vo??, 2008, p4) A.3 Improving terminal performance Improving the performance of container terminals has been studied extensively and is described in a few well known review papers (Kim & Gunther, 2007; Stahlbock & VoB, 2008; Steenken, VoB, & Stahlbock, 2005; Vis & De Koster, 2003). The most important issues will be explained briefly. Section A.3.1 will elaborate on important problems on the seaside of the terminal. Section A.3.2 will discuss research problems in the yard. 98

114 A.3.1 Operational research on the seaside The most studies issues on seaside operations are the berth allocation problem (BAP), the quay crane assignment problem (QC) and the quay crane scheduling problem (QCSP). The BAP is about allocation of the exact space on the quay and the exact time slot that a ship will be serviced. The decision which and how many quay cranes to assign to operate a berthed ship is addressed in the quay crane assignment problem. The crane scheduling problem is about the exact order that container are discarded and loaded on a ship. As shown in Figure 58 these three problems are heavily dependent on each other (Bierwirth & Meisel, 2015, p615). An important limitation to most research regarding these problems is the assumption that the ship arrival time is known. As shown in Figure 58 the vessel arrival time is important input to solve the mentioned problems. Most research either assumes static or dynamic arrival times (Bierwirth & Meisel, 2015, p620). When assuming a static arrival time, all ships are already in the port when the berthing schedule is made (Imai, Nishimura, & Papadimitriou, 2001, p404). With a dynamic arrival time ships have different arrival times. It is explicitly assumed that these arrival times are known when the berth schedule is created (Imai, Nishimura, & Papadimitriou, 2001, p406). Pengfei, Haigui & Li (2006, p7298) is one of the few studies that assume uncertainty about the arrival time of the ships. Xu, Chen & Quan (2012, p139) researched how the robustness of berthing schedules can be improved by including buffer times in the schedule when arrival times are uncertain. Their robust berthing schedules perform better compared to normal berthing scheduling when the arrival time of ships is considered uncertain. Figure 58 - Dependecy BAP, QCAP and QCSP (Bierwirth & Meisel, 2015, p618, fig5) A.3.2 Stacking policies in the yard The performance of the yard equipment has a direct effect on the performance of the quayside. The productivity of the yard equipment decreases when the yard occupancy increases. (Kemme, 2012, p570; Saanen & Dekker, 2006, p82) Much research regarding yard performance is aimed at improving stacking policies. Decisions on the tacit level of the stacking policy are the use prestacking and housekeeping or final grounding (Dekker, Voogd, & Van Asperen, 2007, p562). Prestacking is putting a container at a temporary location near the quay when it is discarded from a ship. This container will be put to its final position in the yard, when few ships are being operated. This is called a housekeeping move. Final grounding is putting the container at its right position right after discarding from the ship. Final grounding is in most cases more efficient than prestacking and housekeeping (Saanen & Dekker, 2006, p92). 99

115 The second decision is where to put the container exactly. Two main types of strategies are used. In the traditional strategy an area of the yard is specified where the containers can be stored. This space is reserved for these containers. In the random stacking strategy a container will be stored on a random location. The random stacking strategy is more productive than a traditional strategy in RTG equipped yards of transhipment terminals with a high yard occupancy(saanen & Dekker, 2006,p86). 100

116 B. Actor Analysis In this appendix the main actors are described and analyzed. It is concluded that most stakeholders want a transhipment terminal that is able to handle much volume with low handling rates and handling time. Section B.1 describes the main actors. Section B.2 describes the main needs toward and KPIs of a transshipment terminal. This appendix concludes with a brief conclusion in section B.3. B.1 Description of main actors The shipper A shipper is an economic agent that produces and ships goods (Holguín-Veras et al., 2011). They need container transhipment terminals that enable transport container chains to be reliable with low transit times and low transport cost (Kent & Parker, 1999). The carrier Carriers sell transport to shippers from the place of origin to either a container terminal, or in the case of carrier haulage, a place more inland. They need reliable terminals with low handling rates and high handling speed. This supports low transit times and transport costs in the transport chain. Carriers want competitive chains to increase the volume they handle. The most important carrier in the case of Algeciras is Maersk Line supplying more than half of all the cargo that is handled here. Maersk Line is the biggest carrier in the world. They operate a global transport networks using hubs. Algeciras used to be their main hub for connecting African and Mediterranean ports to their network(fremont, 2007). The terminal operator The main interest of the terminal operator is to increase profitability by increasing the number of containers they handle. Therefore the need to ensure a high performance in the handling of container while keeping their prices at a reasonable level. In Algeciras there are two container terminals with two different Terminal Operators, Total Terminals & APM Terminals. APM Terminals and Maersk Line are both part of the A.P. Moller Maersk group. This way Maersk Line can secure the port operations of key ports on the long term. It also enables Maersk to have more control of the costs of the total transport chain (Fremont, 2007). Two thirds of handled container volume by APM Terminals are carried by Maersk Line. Container terminals located in the same port compete with each other based on handling costs, handling speed, reliability and hinterland connections (B. W. Wiegmans, Hoest, & Notteboom, 2008). Hinterland connections are in the case of Algeciras less important as it is mainly used as transhipment hub. The port authority The Port Authority is the landlord, regulator and operator of nautical services in the port. Typical operator function they fulfil are pilotage and towage (Verhoeven, 2010). Their main interest is to increase regional economic growth and the number of jobs by increasing the volume handled in the terminals of a port. For the port authority an import or export container is more valuable than a transhipment container as import and export container generate more jobs and are more bound to the port. Competitive ports are those who help to minimize the sum of sea, land and port costs (T. Notteboom, 2006). To be competitive as a transhipment hub this would mainly mean ensuring low sea transport costs and low handling costs in the container terminal and low port tariffs. The KPI s of the port are thus the volume that is handled by the port of Algeciras, the share of import & export containers that 101

117 is being handled and the transport costs of the chain going through Algeciras. The main resources of the port authority are lowering port tariffs, increasing the productivity of nautical services and by increasing the logistic infrastructure. Custom Authorities The customs authorities are responsible for controlling the import and export flow of goods and for collecting tariffs. The main interest of the customs authority is to maximize enforcement of regulation while effecting port operations as less as possible. Freight Forwarder Importers hire freight forwarder to coordinate the transport chain for their client (B. Wiegmans et al., 2017). They help overcome coordination problems and in this way reduce the associated coordination costs for their clients (Langen & Chouly, 2004). They main interest of the freight forwarder is to increase their profitability by forwarding more containers. There operations is mainly landside focussed. They benefit from more volume (import and export) containers that are being handled in a terminal. The logistic network of Andalucía (case specific). Logistic Network of Andalucía is a regional government company that aims at increasing the share of the world trade that is being transported through Andalucía by improving the logistic infrastructure in Andalusia. They are the initiator of the inland port that will be tightly coupled to the studied terminal in the port of Algeciras. Their main interest is to increase the volume of containers that is moving through the port to increase the economic growth in the region. Truck operator Truck operators pick-up and drop-off containers at the landside of the container terminal. Their main interest is to transfer as much containers as possible by ensuring low transit times and transit costs for their clients. Their operations are affected by the waiting times at the terminal and the information they receive from the terminal operator, custom authority and freight forwarder. Their most important KPI in the demarcated chain is the landside turnaround time for trucks. Beneficial Cargo Owners (BCO s) (Importer) Beneficial Cargo Owners (BCO s) are owners of containers that arrange their own transport on the landside. They face similar problems and have similar goals as truck operators. Because they ships bigger volumes compared to other shippers, they are more likely to negotiate deals with other stakeholders in the chain. They select their transport chain based on transit times, transport costs, reliability of the transport time, the frequency of transport offered and the amount of value added services in a chain. Stevedoring company The stevedoring company provides the stevedores (workforce) to terminals. Stevedoring companies only exist in countries like Spain with a strong port union. The main interest of the stevedoring company to provide as much work possible for their members. This can by either protecting as much as jobs in the port as to increase the number of jobs in the port. 102

118 Table 18 shows a structured overview of the interest, problems and improvements that stakeholders currently have with transhipment terminals in the transport container chain. The statements are based on the case study of the studied terminal. Table 44 - Interest of different actors Stakeholder Interest Desired Situation Carrier Terminal operator Port Authority Shipper Customs Authority The logistic network of Andalucía Maximising Profit Maximising Profit Regional economic development Maximising Profit Protection of European economy and safety Regional economic development Handling more volume Handling more volume Good economic growth and creation of jobs. High number of containers & ships handled. High Quality and cheap container transport. Maximize enforcement of regulation while effecting port operations as less as possible Handle more containers in Algeciras Existing or Expected situation Capacity of the transhipment terminal is to low Moves per hour in the Terminal to low. Growth of volume is threatened. Capacity of the transport chain is constrained. Customs is hurting the productivity in terminals. Growth is constrained by spatial development Causes Productivity in the terminal is to low High utilization rate of yard, lack of accurate information. Development of neighbouring ports which also want to fulfil a transhipment functions. The productivity of the transhipment terminal is to low. Containers needs be withdrawn from stack and be put back when they are being inspected. Lack of good logistical infrastructure to hinterland. Possible solutions Use other transhipment terminals, force terminal operator to increase productivity. Mitigate part of operations to an close inland port. Improve information exchange with direct transport partners. Increase number of import / export containers. Increase the competitiveness of the port of Algeciras. Increase productivity container terminal. Use other transport chains. Use innovative solutions to decrease the number of containers that need to be unstacked. Inland port with direct rail connection to hinterland 103

119 Freight Forwarder Truck operator BCO Stevedoring Company Maximising profit Maximise Profit Maximise profit Maximise profit Handle more volumes Move more containers High quality and cheap transport options to hinterland. Monopoly on port jobs problems for the studied terminal The share of import containers is small in the container split. Avoidable waiting times in operations Transport chain to hinterland lacks quality Job market might be liberalised Accessibility of the hinterland is too low. High waiting times at gate Lack of value added services. Lack of infrastructure to hinterland European Law Direct train connection between Algeciras and inland ports in Madrid. Truck appointment system Inland Port. More infrastructure to hinterland. Use striking 104

120 B.2 Needs and KPI s towards an transhipment terminal Different actors have different needs towards a transhipment terminal. These needs translate to different key performance indicators. In Table 45 the KPI s, goals and resources of the main actors towards an transhipment terminal are shown. Table 45 - Goals, KPI's and resources of different actors Stakeholder Need KPI s Resources Shipper Cheap, reliable and Handling Costs of Money fast transport chains terminal, Handling supported by the speed of terminal, container terminal. Reliability of terminal Carrier (Maersk) Container terminals Capacity of the Knowledge, Market should be able to terminal, Handling power, Money (for handle much volume Costs of terminal, investments), against reasonable Handling speed of Equipment rates with fast terminal, Reliability of handling time. terminal Terminal Operator Increase moves per volume handled, Knowledge, Money for hour while keeping handling costs per investment, operations costs as container, ship equipment. low as possible. turnaround time at Furthermore the terminal, Moves per handling of more hour STS crane. volume is preferred. Port Authority Increase regional Regulatory power, economic Nautical Services, development by handling of more containers. Containers handled in the port, The share of import and export containers handled in the port, The costs of the transport chain. decides port tariffs, lobbying power for infrastructural improvements. Custom Authority The logistic network of Andalucia Guarantee import and oxpert in accordance with laws More profits in Andalucía from the global container transport. Freight forwarder Higher volume of import and export containers. Truck operators Improve landside transport Beneficial Cargo Improve landside Owner transport. Increase availability of value added services. Number of containers inspected, Hours of delay caused Number of containers handled in Algeciras, Share of import / export container in the container split. Pickup time container landside Truck turnaround time in terminal Handling Costs of terminal, Handling speed of terminal, Reliability of terminal Regulatory authorities. Investment Capital. Support of government. Low transaction costs due to contracts and relationship with local stakeholders. Truck Equipment. Workforce money 105

121 B.3 Conclusion Based on the actor analysis it can be concluded that most important actors need a transhipment terminal that is able to handle much volume with a high handling speed and low handling rates. Which part of the terminal is most important depends of the perspective of the actor. 106

122 C. Information demarcation The design of physical components of a transport systems should be intertwined with the design of related information flows in order to maximize the potential benefits (Lee and Wolfe, 2003). Especially because information is one of the enablers in of efficient freight transport systems (Giannopoulos, 2004). In this section the current information exchange between the different stakeholders in the transpor chain around the terminal operator are identified using a workshop with a terminal expert from Navis. The analysed case is the port of Algeciras. Section C.1 describes the information exchange that happens between the stakeholders on the seaside. Section C.2 describes inforamtion exchange on the landisde. In section C.3 the informaton is classusterd using the chronlogical specification from Menger (2016). C.1 information exchange seaside In this section the most important information exchange on the seaside are described. It is demarcated to information exchanged between the carrier, terminal operator and port authority. The carrier is here split in a carrier that owns the vessel and a carrier that owns the containers. Most ships transport containers from multiple carriers. The information exchanged by or to a carrier depends on whether it is also the owner of the vessel. The information exchanged is shown in Figure 59. A month before berthing a long term schedule is send by the owner of the vessels to the terminal operator containing the calling date of a vessel (1). From now on this schedule is updated every week or at a change (2). A week in advance the carrier sends the estimated arrival time (ETA), BAPLIE and Movin to the terminal operator (3). The BAPLIE contains the status of containers in the vessel, for example whether is full of loaded. The Movin contains the stowage plan of the vessel, i.e. the location of the containers in a vessel. The Coprar is send at a minimum of 18 hours in advance to the terminal operator by the carrier that owns the containers. The Coprar contains information about which containers should be discarded and which should be loaded (4). The terminal operator sends the berth window to all carriers who have containers on the incoming vessel and to the port authority (5). Based on the berthing window the port authority assigns pilots and tugboats to the vessel. Around three hours in advance the captain of a ship, representing the carrier that owns the vessel, calls the pilots to ask for instructions (6). This instruction can be to wait at a waiting area or to meet at the pilot boarding point. This is repeated 1 hour in advance. During loading and discarding the terminal operator sends a report every half hour to the carrier (owner of the container) containing which containers are loaded and discarded(7). This report is called the coarri L-D. Around one hour before completion the carrier (owner of the vessel) will send the estimated time of completion to the port authority so it can assign pilots to the vessel (8). When the loading of the vessel is completed, the terminal sends an updated version of the BAPLI, the BAPLI departure, to the carrier (9). 107

123 Figure 59 - Most important information exchange seaside 108

124 Customs Authority Carrier Terminal Operator Truck Operator Receive Booking File Pickup request (2) Pickup container and transport to customs Wait for inspection Transport container To terminal Lookup Container Availability File Pickup request (5) Pickup container Receive request Process Request Released Container Set container to available (4) Receive Request Process Request Phase Reach Port Send Customs Documents Receive inspection notification Book transport to customs Receive documents Evaluate documents Inspection needed Yes Mark for inspection (1) Inspect Container Release container (3) Figure 60 - Most important information exchange on the landside 109

125 C.2 Information exchange landside The information on the landside is shown in Figure 60. On the landside less information is exchanged. Customs must decide whether a container needs inspection or not. If a container is marked for inspection, they will notify the carrier (1). It is the responsibility of the carrier to arrange a truck operator to move the container from the terminal to customs. The carrier files a request for pickup of the container to the terminal operator (2). This request only gives the truck operator access to terminal. It does not contain any information on when the truck driver will arrive at the terminal. The truck driver will pick up the container and bring it to customs for inspection. In Figure 60 it is assumed that the container is cleared by customs. In Algeciras the terminal operator will publish that the container is available for pickup(3). A truck operator can login to a system to check whether the container is available. The truck operator files a similar request when it going to pick up a container at the terminal on behalf of an importer or freight forwarder (4). This request does not contain any information on when truck operator will pick up the container. C.3 Classification of information exchange In Figure 61 the information that is currently being exchanged between the different stakeholders is clustered based on their properties. The first property is if the information exchange is used for seaside of landside operations. The second property is the type of information based on the chronological classification of Menger (2016). Several terminal experts 16 agreed that that improving future event information is more likely to have a good impact on the productivity of the terminal than improving the exchange of the other types. This research will therefore be scoped to future event information. Figure 61 - Classification of information currently exchanged 16 Member of the board SMDG, VP Navis Atom 110

126 D. Information exchange alternatives In this section multiple event based information exchange alternatives have been identified based on expert interviews and literature research. Section D.1 describes improvement of the accuracy of the ETA. Real-time sharing of the berth window is discussed in section D.2. D.3 elaborates on sharing the availability prediction of import container in the terminal. D.4 contains a discussion a truck appointment system in the terminal. D.1 Improving the accuracy of the ETA The ETA of a vessel is used by both the terminal operator as the truck operators to make a planning for their resources. Section D.1.1. will discuss the current accuracy of the ETA based on case data of the studied terminal. Section D1.2. elaborates on how the ETA is used in gang nomination. Section D.1.3 the effect of the ETA on the allocation STS cranes. D.1.1 Accuracy of the ETA Interview with terminal experts 17 revealed that the ETA is not as accurate as desired from the perspective of a terminal operator. Figure 62 shows the estimated time of arrivals compared to the actual time of arrivals for multiple ships. For every ships multiple estimations are available. In the figure three lines are shown that display how the estimate time of arrival of a ship changes over time. It is concluded that there is discrepancy between the estimated time of arrival and the actual time arrival. Most estimations differ between 0 and 5 hours from the actual time of arrival, but also differences of more than 5 hours are present. It should be noted that this conclusion is drawn on a very limited data set. Figure 62 - Accuracy of estimated time of arrival 17 Functional experts, Navis & Member of the board of SMDG 111

127 D.1.2 Nomination of gangs A container terminal uses the ETA received from the carrier to allocate its resources. A critical moment in the allocation of resources is the scheduling of workforce. In the case of Algeciras the terminals need to nominate gangs every day at noon for the next four shifts between that day and next day. To allocate the right resources, a good estimation of the time of arrival of inbound vessels during the next 26 hours should be available. Figure 63 shows the difference between the ETA and the ATA at gang nomination for 31 ships. It shows that 3 out 31 ships arrive around 4 hours late, missing half of their berthing window. One ship is 14 hours late compared to its reported ETA, missing its whole or a big part of its berthing window. Four ships arrive 7 or more hours early compared with their reported ETA. Figure 63 - Difference between ETA and ATA at gang nomination Based on the ETA of inbound ships a workforce planning is made. Discrepancies between the estimated time of arrival and actual time of arrival lead to overmanning and undermanning in the terminal, decreasing productivity and performance (Fancello et al., 2011,p2). Figure 64 shows the effect of an inaccurate ETA on the ship turnaround time. D.1.3 Reallocation of STS cranes A terminal operator is continuously reallocating resources as ships arrive earlier or later than expected. The ETA is important information for a terminal operator in its decision making process. A terminal will try to optimise the performance of the terminal with respect to the service level agreements of the client ships. Inaccurate ETAs hurt this optimization process. Cranes can stay idle because they are assigned to a ship that is unexpectedly delayed. Contrariwise ship that arrive earlier than expected cannot be operated immediately because their cranes are currently operating another ship. A more accurate ETA enables the terminal operator to make better decisions in reallocation of the STS cranes. 112

128 Figure 64 - Relationship improved accuracy ETA and real time sharing of the berth window on ship turnaround time D.2 Real time sharing of berth window Sharing of the berth window enables a carrier to adapt its speed based on the predicted state of the terminal at arrival. This can decrease waiting time and increase the utilization of terminals resources (Lang & Veenstra, 2010). Waiting at the ships waiting area is considered waste, as a ship then could have sailed at a lower speed (Lang & Veenstra, p478). Reducing the speed of the ship can save significantly on the bunker consumption of the ship. The bunker costs of a ships can contribute more than half to the total operating cost (Notteboom, 2006, p36). Reducing speed can save up to 20% of the total operating costs of the ship (Maloni, Paul, & Gligor, 2013, p151). Real time sharing of the berthing window can also help the terminal to allocate their resources more efficiently. When cranes are scheduled to be available without any assigned ship, a terminal operator can advise ships to increase their speed a little. Ship will arrive earlier at the port. The terminal can increase it productivity of the cranes as idle cranes can now be used to operate this ship. D.3 Share container availability prediction Trucks operators would like predictions on container availability from the terminal(b. Wiegmans et al., 2017). In several interviews 18 with terminal experts it was mentioned that it is likely that truck operators could better allocate their resources if they have better predictions on when a container becomes available. Containers terminals do have the information to make these predictions. However as described in section there is no financial dependecy from the trucking companies towards the terminal operators. The terminals therefore lack incentive to spend resources on sharing the container availability with truck companies. This incentive could arise when truck operators will pick up containers earlier when good predictions are available. A researcher from TNO stated that it is possible that the dwell time of import containers are reduced when better predictions are available, but that there is currently no research that supports this. If this would be the case, reducing dwell times for import contains will have a 18 Vice president Navis Atom, Port Researcher TNO 113

129 positive impact on the yard occupancy of the a container terminals and thus on its efficiency. This mechanism is shown in Figure 65. This would provide the incentive for the terminal operator to share the container availability predictions with its hinterland partners. D.4 Truck appointment system on terminal Introducing a truck appointment system will force truck operators to share their information on when they will pick up a container. This information can be used in two ways two decrease the number of unproductive moves. The first is to prepare the container during a housekeeping move when the yard equipment would be idle. This method does not decrease the number of unproductive moves, but decreases their effect on the performance of the yard (Van Asperen et al., 2013, p. 557). The second way is to determine a better location of the container when it being rehandled. This means that a container will not be put on top of a container that will be picked up before the rehandled container will. Using this method can decreases the number of rehandles caused by the retrieval of import containers by 50%. (Zhao & Goodchild, 2010, p343). The effect of fewer productive moves on the ship turnaround is shown in Figure 65. It can be seen that fewer unproductive moves lead to less utilization of the yard equipment, improving the productivity of the both the cranes as the berth. No literature could be found on the quantitative effects of a truck appointment system on the productivity of the terminal and the ship turnaround time. Figure 65 - Effect of sharing container availability and truck appointment system on ship turnaround time 114

130 E. Representation of most important objects This appendix briefly discusses the conceptual modelling of transporters and terminal equipment. Section E.1 describes the modelling of transporters. Section E.2 describes the modelling of the terminal equipment. E.1 Transporters Transporters are all objects that move containers between locations. It is not possible to model the horizontal movement of containers inside the container terminal in an useful way, because all the stacks are modelled as one big stack. Therefore no internal terminal trucks are modelled inside the system. During validation it has been concluded that this will not decrease the usability of the model as the horizontal moves are not likely to be the bottleneck in the systems. The transporters which are used for the movement between the container terminal and the inland port are modelled in a more detailed way. The transporters are either trucks, a train or a shuttle train. Transporters pick up the containers in either the main stack or the buffer in the container terminal. The transport time between both terminals is based on google maps data and validated by local experts from Navis. The capacity of the transporters is measured in TEU s. A truck transporter has a capacity of 2 TEU, meaning that they can transport either 2 twenty feet containers of 1 forty feet. E.2 Terminal equipment Two types of yard equipment are available in the model. Ship to shore cranes (STS cranes) and rubber tired gantry cranes (RTG). STS cranes are used for operating ships. RTGs are used stacking and unstacking containers in a yard. STS cranes are modelled as simple queuing system. Each berth has separate queue s for containers that need to be loaded and containers that need to be discarded. Multiple STS cranes can be assigned into handling one queue. The allocation algorithm is discussed in section The performance of the STS cranes in Algeciras is estimated to be 30 moves per hour when it is not constrained by the yard performance. It is assumed that all twenty feet containers are being handled with a twin move. The handling time of a twenty feet containers is half the handling time of a forty feet container. RTGs are also modelled as simple queuing systems. There are separate queues for containers that are specified for a specific berth or for the landside. The work schedule of the amount of RTG working in the yard is based on the schedule of the STS cranes. For each STS crane that is available in a shift, a terminal specific number of RTG s are available. The average productivity of a RTG is estimated to be 11 moves per hour. The performance of the RTG also depends on the handling time multiplier, which represent congestion in the yard. 115

131 F. Modelling the inaccuracy of the ETA Based on a dataset containing multiple ETAs communicated over time of around 40 ships calling the studied terminal in Algeciras, a model to represent the inaccuracy of the ETA is specified. The dataset contains multiple estimations per ship. These estimation are from different sources and are used in real world decision making. This dataset is shown in Figure 66. Each colour represents another ship. It is clear that some measurement errors are present in this dataset. Striking is a group of points where the difference between the ATA and ETA is roughly the same the number of hours that the ETA was received before the ATA. The hypothesis is that a systematic measurement error was made here. These points where corrected using a cluster algorithm (see Figure 67). The points now become white noise around the x-axis. When this hypothesis would be proven to be false and the points where actually correct, it would affect the outcome of this study as follows: The ETA is less accurate than shown in this study. Improving the accuracy of the ETA will improve the ship turnaround time and gang productivity more than shown in the results in this study. Real time sharing of the berth window will improve the ship turnaround time and gang productivity more than shown in the results of this study. Some additional points where removed from the dataset. These points are received after the ship already berthed. Also straight lines, consisting of set of point with equal distance on the x-axis of the same ships are considered measurement errors. These points are removed. Figure 66 - uncleaned dataset showing ETAs over time till ATA 116

132 require(tidyverse) data <- read_excel("~/r/***/infopiloteta.xlsx") data <- mutate(data, deviation_eta_ata = ATA - ETA) %>% mutate(received_before_ata = ATA - Received) data <- mutate(data, Source = replace(source, Source == 1, "PMS")) %>% mutate(source = replace(source, Source == 2, "COMS")) %>% mutate(source = replace(source, Source == 6, "Marine Traffic")) data$deviation_eta_ata <- as.double(data$deviation_eta_ata, units = 'hours') data$received_before_ata <- as.double(data$received_before_ata, units = 'hours') df <- filter(data, Received_before_ATA < 48) %>% filter(abs(deviation_eta_ata) < 48) %>% filter(received_before_ata > 0) %>% filter(vessel_name!= "MAERSK ARKANSAS") %>% filter(vessel_name!= "MAERSK BULAN") ## Filter ships with unrealistic paths cluster <- kmeans(df[,6:7], 3) ## Warning: K-means result can differ due to randomized starting points ggplot(df, aes(x = Received_before_ATA, y = deviation_eta_ata)) + geom_point(aes(color = cluster$cluster)) df <- mutate(df, deviation_eta_ata = ifelse(cluster$cluster == 1, deviation_eta_ata - Received_before_ATA, deviation_eta_ata)) Figure 67 - R Code for cleaning dataset Analysis of the ETAs provided by the carrier trough different systems of 40 ships calling the studied terminal in Algeciras shows that the estimations are quite inaccurate. A general distribution must be identified that can represent this inaccuracy. Figure 68 shows the estimated time of arrivals compared to the actual time of arrivals for multiple ships. For every ships multiple estimations are available. In the figure three lines are shown that display how the estimate time of arrival of a ship changes over time. It is concluded that there is discrepancy between the estimated time of arrival and the actual time arrival. Most estimations differ between 0 and 5 hours from the actual time of arrival, but also differences of more than 5 hours are present. It should be noted that this conclusion is drawn on a very limited data set.. 117

133 Figure 68 - Distribution of inaccuracy of ETA Figure 69 shows three kind of ways how they can change of time. They can stay the same, make small steps of make very big steps. The distribution that will model this inaccuracy should be able to cover these three behaviours. Figure 69 - Accuracy of ETA per ship (zoomed) The inaccuracy of the ETA is modelled as an hourly change in arrival time. This hourly change in arrival time can be seen as a biased random walk. The random walk represent the possibility of both small and big changes in the arrival time. The random walk is slightly biased, because the arrival time has a bigger possibility of being postponed than advanced. A Bernoulli distribution has been added to the random component to increase the probability that there is no delay nor advancement compared to the previous estimation. The model that is used to represent inaccuracy of the estimations of the time of arrival is shown here. y t = y t 1 + B(1, 0.5) N(0.1, 0.75) Figure 70 shows 40 sets of estimations of the time of arrival generated over time. By visually comparing Figure 70 with Figure 68 & Figure 69, it is concluded that a biased random walk is useful to model the inaccuracy of information. 118

134 Figure 70 - Generated inaccuracies ETAs 119

135 G. Generation of the pro forma berthing schedule Variance in the generated pro forma berthing schedules is desirable as the results should be robust for a wide range of possible berthing schedules. The goal of the ship arrival generator is to include enough stochasticity to generate a wide range of possible pro forma berthing schedules without generating to much unrealistic berthing schedules. An unrealistic pro forma berthing schedules is for example a schedule where too much ships arrive at the same time. Another unrealistic schedule is where too much mega vessels arrive at the same time. Normally a terminal operator would spread these ships in their pro forma berthing schedule 19. Model testing showed that the variance in the berthing window can be influenced by the used distribution of the interarrival time of the ships and by the distribution that determines the ratio loading to discarding containers on a ship. The container split and ship class split are here assumed constant. Literature shows three ways how the interarrival time can be modelled (Polman et al., 2003). These are as a Poisson process, equidistance or stock-controlled. The Poisson process is well known distribution where the interarrival time are exponential distributed. In equidistance the distance between the ships is constant. Stock controlled is the generation of ships based on the yard occupancy in the terminal. The effect of both a Poisson process as a form of equidistance is simulated and analysed. The equidistance strategy is adapted to include stochasticity by sampling the interarrival time from a normal distribution where the mean is the interarrival time when equidistance would be applied. The standard deviation is chosen is in such a way that an interarrival time of zero is possible with a probability of around 1%. Figure 71 shows that taking a equidistance based approach with a normal distribution has way less variance than a Poisson distribution for the interarrival time. It shows that big outliers are created by using the Poisson distribution. Also the average turnaround time is higher when the arrival process is a Poisson distribution compared to when using a normal distribution. This is in accordance to what is described by Polman et al. (2003). It is chosen to model the interarrival time with this equidistance approach, because a Poisson generates to much extreme scenarios causing to much variance in the KPI s. A equidistance based approach also reflects reality better as a terminal operator can make sure that the pro forma berthing window is balanced 9. Figure 71 - Poisson distribution vs Normal distribution 19 Terminal operation experts, Navis 120

136 A distribution must be chosen that determines the percentage of loading (or discarding) container on a ship when generating ships randomly. In reality this ratio can take any value between 100% and 0% loading (or discarding) moves. When a ship is mainly discarding containers it likely that on short notice a ship will arrive that mainly load containers. When generating ships at random to create different scenario s this causal relationship is not included. Therefore a distribution needs to be chosen that limits variance in the KPI s but captures the possibility of a difference in loading / discarding moves best. Figure 72 shows that the boxplot showing the effect of the distribution that is used to decide the percentage of loading moves on the total number of moves. It shows that using a normal distribution introduces a lot of variance on the KPI s. Using an uniform distribution decreases the variance a lot. It is chosen to vary the percentage of discarding containers between 45% and 55% to minimize variance. Figure 72 - Effect distribution for deciding percentage loading moves 121

137 H. Simulation model specification In this appendix the specification of the simulation model is explained in a detailed way. First an overview of the model is discussed in section H.1. Section H.2 discusses the ship and container entities. H.3 shows the structure of the different subparts of the model. H.1 Overview of model Figure 73 shows an overview of the simulation model that is used in the model. It can be seen that the model is be divided into six parts. The first is the ship generator which is located at the left of the model. The second is the sea route to Algeciras, which is the line between the ship generator and the red square. The red square represents the quay, which is the third part of the model. The fourth part is the yard which is the yellow block. In between the yellow and cyan rectangles the fifth part in form of the transport network between the inland port and the deep-sea terminal can be seen. The cyan rectangle is the inland port, which the last part of the model. Figure 73 - Overview of the simulation model in Simio H.2 Ship & Container entities Two main classes of entities are defined in this study: ships and containers. These classes of entities have been enhanced with additional state variables. Table 46 shows the state variables that are added to container class on top of the standard entity state variables. Table 47 shows the additional state variables of the ship class in addition to the standard entity state variables in Simio. Table 46 - Additional state variables container class Name Object Type Description Outbound seaside Boolean Set to true when container will leave the model on the seaside. Used for directing the container trough the model. Marked by customs Boolean Set to true when container needs to be inspected. Used for directing the model to inspection in either the inland port or deep-sea terminal. Maersk Boolean Set to true when the container belongs to Maersk. Maersk import and export container are threatened differently in some scenarios. Arrived at berth Real Stores the time that the vessel arrived at the berth. Used in calculating the terminal turnaround time. 122

138 Allocated outbound berth Node Reference Stores the destination berth of the container when leaving the main yard. TEU Integer Saves when the container represent 1 or 2 TEU. Dwell time Integer & Date Save the assigned dwell time to the model Time STS Handling time Real Saves the time the STS crane will delay this entity Allocated to ship Boolean Is set to true when a container is assigned to a container. Prevents the container to be assigned to other ships, while still in the main yard. Allocated to shuttle Boolean Is set to true when a container is assigned to be transported by the shuttle. Prevents the container to be assigned to another shuttle while in the main yard. Table 47 - Additional states of ship class Name Object type Description Total moves Integer Total number of moves that the ships will have at the terminal. Loading moves Integer Number of loading moves at the terminal Discarding moves Integer Number of discarding moves at terminal Discarding_20s Integer Number of 20s that will be discarded Discarding_40s Integer Number of 40s that will be discarded Maersk Boolean True when ships is from Maersk. Maersk ships transport more Maersk containers Current Speed Real Saves the current speed (knots) of the vessel. ETA Real & Datetime Estimated time of arrival at the terminal ETC Real Estimated time of completion. Is decided and use in when creating a berth window by the lookahead algorithm. ETA_early Real Time of estimated arrival when ships sails at maximum speed Advised speed Real Advised speed based on sharing of the berth window Scheduled in berth window Boolean Set to true when the ships is scheduled by the berth window and is reset when the algorithm is finished. Prevents the scheduling of the same ship twice. STS cranes in SLA Integer Number of STS cranes in the SLA Max STS Integer Maximum number of STS that can be assigned to the ship Assigned STS Cranes Integer Number of STS cranes assigned when the ship is berthed Start waiting time Datetime Saves the datetime when the ships entered the waiting area Waiting time Real Saves the time the ship had to wait in the waiting area Actual time of arrival Datetime Saves the datetime the ship arrived at the berth Turnaround time Real Saves the turnaround time for the ship 123

139 H.3 Subparts Model This section describes the implementation of the different parts in Simio. H.3.1 Ship Generator Figure 74 shows the model structure of the ship generator. Each class of ships has it own source. For every class a separate interarrival time can be set. Every source uses its own random number stream to reduce variance between different scenarios. All sources are connected to the searoute towards the terminal. Figure 74 - Structure ship generator After creation of the ship entities, add-on logic is used to assign multiple properties to the entity. The add-on logic only contains decide and assign blocks. Based on the class the following properties are assigned: Maersk or not? Number of cranes in the SLA Maximum number of cranes Loading moves at Algeciras Discarding moves at Algeciras o Number of 40 s o Number of 20 Current speed ETA according to pro forma berthing window H.3.2 Sea route to terminal Figure 75 shows the model structure of the sea route Algeciras. The searoute is represented by a path with a logical length of 481 sea miles. It connects the ship generator with the quay. In the top right corner a queue is modelled that acts a waiting area for the ships when no resources are available at the terminal. 124

140 Figure 75 - Sea route to terminal Three important processes are defined in this part of the model. The first is nomination of the gangs. The nomination of the gangs is triggered by a timer that fires once every 24 hours. It starts a main process that exists of multiple execute blocks in other to start subprocesses. Each subprocess is aimed at finding ships in certain parts of the model and determining when gangs are needed for that model. Figure 76 is an example of such a process. Using a search step ships are found that are on the searoute towards Algeciras. Based on the ETA of the ships a process is fired that nominates the number of gangs need in the next four shifts. An example of such a process is given in Figure 77. It starts with nominating the number of cranes in the SLA of the ships for the shift that the ship arrives, in this case the shift from 20:00 till 02:00. Next it decides after each shift if the ship will be still there based on its ETA, expected service time and the length of the shifts. When the ship is estimated to be there, cranes will be assigned for the shift. This process continues to either the ships is estimated to be finished, or the fourth shifts is reached. This process is repeated for all ships sailing toward the terminal, at the waiting area and at the berths. Figure 76 - Nominating gangs for inbound ships 125

141 Figure 77 - Nominate gangs for ships arriving between 20:00 and 02:00 The second process is delaying or advancing ships to represent wrong ETAs. The process is displayed in Figure 78. It shows a process that is fired hourly. First it is decided whether perfect information is assumed in this experiment. When this is not the case, ships sailing towards the terminal are searched. With a decide it is decided if a change in ETA takes places. When this is the case a new ETA will be assigned to the property of the ships and the movement will be adapted using the movement.rate property of the entity. Figure 78 - Change ETA inbound ships The third process is real time sharing of the berth window. The structure of the process is shown in Figure 18. There was no option to properly sort a list of ships based on their ETA inside Simio. Therefore the sorting function was implemented using recursion based on the search and execute step. H.3.3 Quay The quay is divided in six discrete berths. The implementation of the berth is shown in Figure 79. Each berth is identical. When the ship enters the berth a process is fired that serves multiple functions. One process will store the ship into a station and will use a scan step to see if the servicing of the ship is finished. The second step creates containers based on the property discarding_20 and discarding_40 of the ship. These containers will be transferred to the discarding queue. The third process searches the main yard for containers that can potentially be loaded into the ship. Potential containers are containers that are stored longer in the yard than there specified dwell time. The maximum number of containers that will be loaded into the ship is specified into the loading_moves property of the ship. Ships do not wait when not enough loading containers are available. 126

142 Figure 79 - Implementation of the berth A second important process in this part of the model is allocating a number of cranes to the ship. This is done accordingly to the conceptual model in Figure 17. The number of cranes assigned is split over equally over the server for discarding moves and the server for loading moves. This means that these moves are done in parallel. When an unequal number of cranes is assigned to a ships, the last crane is assigned to the server representing discarding moves. When all specified containers are discarding, the cranes doing discarding move will start doing loading moves. This works also vice versa. When the service is finished, the ship will leave the berth. After each loading or discarding move, the terminal recalculates the estimated time of completion of the ships. This is done using the number of work left and the performance of the terminal in the previous day. H.3.4 Container yard Figure 80 shows implementation of the yard in Simio. The yard is modelled as one station with one entry point and multiple exit points. The entry point is a server with a specified handling time for stacking a container. The capacity of this server is equal to the number of RTGs working in the yard. When the container is processed by the server there are two options. Either the container will be stored in the station, until it is retrieved by add-on logic (for example when a new ship arrives) or it will be transferred direct into an unstacking queue, where it will wait for a certain condition to be fulfilled. This is the case for containers that need to go to customs, the inland port or will be picked up by trucks from the landside. When a ship arrives at the berth, it will search for potential containers in the yard. When a container is found it is transferred to an unstacking server that is designated to a certain berth. Each berth has it own unstacking server to ensure that not all yard equipment is dedicated to serving one ship. The capacity of the dedicated server is based on the number of STS cranes assigned to the ship. All stacking and unstacking servers need to seize a RTG from a pool in order to operate a container. The capacity of this pool is dependent of the number of gangs that is nominated and a constant number of RTGs for landside operations. Add-on logic is used to keep track of the yard occupancy in the stack. This aggregated state is used to determine if unstacking moves take additional time due to a congested yard. Add-on logic is present that keep track of the number of empty containers in the main yard. When the deep-sea terminal is allowed to exchange empty container with the inland port, it will do so when the number of empty containers in the yard is below or above a certain threshold. 127

143 Figure 80 - Implementation of the yard H.3.5 Trucks, external train & shuttle Three modes of transport are modelled. The first is trucks. Trucks are used in two ways. The first is for transport between the deep-sea terminal and the inland port. The second is for transport between the deep-sea terminal and the external rail terminal. Two networks are specified in Simio to reflect these two options. Trucks are modelled using the standard transporter object from Simio. When a container needs transport it reserves the closest available truck. Trucks have a capacity of 2 TEU. Logic has been added that rejects all transport request from other containers when the truck is already transporting a container of 2 TEU. Figure 81 shows the implementation of the rail terminal and the train. The rail terminal is implemented as a set of two servers that are used for either loading or discarding the train. These servers have buffer capacity for when container are brought by trucks when the train is not at the terminal. The train is also a transporter object that dwells the train route. A maximum of two trains can be simultaneously on the railroad. The number of trains that visit the rail terminal a day can be controlled. Trains do only leave the rail terminal when they are full or when there are no more containers that are waiting for transport. Logic is present that avoids the train over transporting to many TEUs. Figure 81 - Implementation of Train and external rail terminal The shuttle train alternative is modelled very similar to the external train. The biggest difference is that the shuttle train departs from the terminal directly. When it arrives at the terminal is searches the main yard of containers it needs to transport to the inland port. These container are unstacked and transported towards the shuttle. The shuttle leaves when all container are loaded on the shuttle. 128

144 H.3.6 Inland port The implementation of the inland port is shown in Figure 82. The inland port is a combination of multiple aspects discussed in the previous sections. It contains drop-off and loading points for trucks, the external train and the shuttle train. The modelling of the yard is similar to the yard in the deepsea terminal. There is one loading server which represents the stacking of all containers. There are two unstacking servers. One server unstacks containers that will go to customs or will be moved towards the deep-sea terminal. The second server unstacks containers that are being picked up by truck operators. The inland port also contains a source that generates the arrival of containers through the landside. Figure 82 - Implementation of the inland port 129

145 I. Verification & Validation Validation is making sure that a computerised model produces results with the right accuracy needed for the purpose of the model in order to make it useful. Verification is testing whether the conceptual model is correctly implemented in the simulation model (Sargent, 2010, p5). Two ways of dynamic testing have been used for verification. These are model testing under extreme conditions and tracing. The extreme condition test are described in I.1. The tracing tests are described in I.2. Validation is done partly by testing the results of the model on historical data. This is discussed in section I.3. I.1 Extreme conditions testing The model has tested extensively under different conditions. These tested have been performed iteratively during model development. The test reported in Table 48 all have been executed after finishing up the model. Table 48 - Extreme Condition testing (observations) Group Input Observed behaviour Exchange strategy All containers are from Maersk All import and export containers move through the inland port. Exchange strategy No container are from Maersk None import export container move through the inland port. Exchange strategy Exchange strategy Exchange strategy Threshold for empty containers (both max and minimum) is set at 500. Transhipment containers with a dwell time higher than 0 will be exchanged with the inland port. Infinity transport capacity. Transhipment containers with a dwell time higher than 100 will be exchanged with the inland port. Start with yard occupancy of 0.9 Deep-sea terminal keeps exchanging empty containers with inland port to keep 500 empty containers in the terminal. All transhipment containers are being exchanged with the inland port. No exchange of transhipment containers. Productivity Yard High number of productive moves in the yard Productivity yard Start with yard occupancy of 0 Low number of productive moves in the yard Low number of productive moves in the yard Productivity yard 1 productive move per RTG per hour Information exchange Enable perfect information No unexpected speed up or slowdowns of ships seen Information exchange Disable perfect information Ships slow down or speed up at random Information exchange Enable berth window sharing No unexpected speed up or slowdowns of ships seen 130

146 Information exchange Disable berth window sharing Ships slow down or speed up at when the berth window is shared I.2 Tracing Two traces have been reported. The first trace is of a ship entity that calls the terminal to be serviced. This trace is shown in Table 49. The second trace is a transhipment container that is temporarily being stored at the inland port (see Table 50).The traces have been simplified in order to be understandable and sufficient to test the model. Table 49 - Tracing of a ship (simplified) Timestep Activity Create ship and assign properties 8.5 (repeated every hour until arriving at Assign delay if ship is delayed terminal) 12 Nominate gangs at the terminal for this ship 32 Terminal operator advices increase of speed based on the berthing window Arrival at terminal: Assign berth, Assign number of cranes, assign number of RTGs, Assign loading containers to ship Load / discard containers Recalculate estimated time of completion Save relevant statistics and destroy entity Table 50 - Tracing of a transhipment container stored at the inland port (simplified) Timestep Activity Container is created and properties are assigned Start unloading container from ship Container is picked up at quay for transport to inland port Start handling container by yard equipment inland port Container is stacked at inland port Start unstacking container at inland port Unstacking finished and loaded on truck towards container terminal Truck arrived at inland port and container is picked by a RTG Container is placed in stack Container is unstacked Operation STS crane started Container is placed on ship 131

147 I.3 Validation using historical data Using the berthing window from the studied terminal two results of the model where compared with the berthing window. The first was the number of the assigned cranes per shift. In the berth window on average cranes per shift were assigned. In the (base case) model on average cranes per shift were assigned. Table 51 shows that both samples have unequal variance. Therefore Welch s two sample T-test is used to test for statistical significance. Table 52 shows that statistically speaking the results of the simulation model are not different for the observed values of the berth window. Table 51 - Levene's test of equal variance for average assigned cranes Degrees of freedom F-Value P-value e-14 Table 52 - Welch two sample T-test for average assigned cranes Degrees of freedom T-value P-value The second comparison made with historical data is the expected ship turnaround time in the berth window compared with the average ship turnaround time from the model. The quality of expected ship turnaround from the berth window is limited as it only tells the expected turnaround time while the actual one might deviate from the expected one. Also the expected turnaround time is given in the number of shifts, while the measured turnaround time in the model is per hour. This means that the expected turnaround time in the berth window are only multiples of six hours. Nevertheless both values where compared. The average expected turnaround time in the berth window was hours. The average turnaround time in the simulation model was hours. This is around 30% lower than the expected turnaround time from the berth schedule. Table 53 show that both samples have unequal variance. Table 54 shows that both values are statistically speaking different. Table 53 - Levene's test for equal variance for turnaround time Degrees of freedom F-value P-value e-08 Table 54 - Welch's two sample T-Test for turnaround time Degrees of freedom T-value P-value e-06 This results was discussed with an terminal expert from Navis. The difference could be partly explained by other activities at the berth like bunkering and repairs that are not included in the simulation model. This cannot explain a difference of 30%. Another possible explanation is that there is difference between the expected berth time and the actual berth time in the terminal. This might cause differences, especially as the expected berth time is reported as a multiple of six hours. A third explanation would be that the model underestimates the turnaround time in the terminal. This is not a problem as the model is used to compare the ship turnaround time across different scenarios and not to predict the specific turnaround time. 132

148 J. Experimental Setup A runtime and the number of replications must be determined in order to order to setup the experiments. The needed warm-up period is discussed in J.1. In J.2 the runtime of the experiments is decided. In J.3 the number of replications is determined. J.1 Warm-up Period Four warm-up periods and their effect on the average turnaround are tested. The tested warm-up periods are 0, 10, 21 and 30 days. The warm-up period will be tested using 40 replications of the base case scenario and a runtime of 100 days. Table 55 shows the confidence interval of the mean given 4 different warm-up periods. Figure 83 visualizes these means using boxplots. The results show that the average turnaround time increases as the warm-up period increases. This means that when the warm-up period is to low, the steady state in the model is not reached yet and the lack of warm-up period is affecting the usability of the model. The differences in the confidence interval decreases as the warm-up period increases. A warm-up period of 21 is chosen as the difference with a warm-up period of 30 is limited, while it needs considerable less computational time. Table 55-95% confidence interval average turnaround time for different warm-up periods Warm-up period (days) 95% interval of the mean of the average turnaround time (hours) ± ± ± ± 0.27 Figure 83- Effect warm-up period on variance and mean average turnaround time J.2 Runtime Three scenarios are run to determine the runtime of the model. The tested runtimes are 60, 100 and 150 days. The runtime will be determined based on the base case, with a warm-up period of 21 days. Of each scenario 40 replications are run. 133

149 Table 56 shows the 95% confidence interval of the different means of the average turnaround time. In Figure 84 the results are visualised using boxplots. It can be seen that the variance of the results decrease when the runtime is longer. Furthermore it can be see that the average turnaround time slightly increases as the runtime is longer. An possible explanation is that a short runtime is still influenced by the (empty) starting conditions of the model. The difference in turnaround time is not statistically significant. A runtime of 100 days is chosen as it estimates the variance with a satisfying variance and reasonable computational time. The effect of the starting conditions of the model are less than a runtime of 60 days. Table 56-95% confidence interval average turnaround time for different runtimes Runtime (days) 95% interval of mean average turnaround time (hours) ± ± ± 0.21 Figure 84 - Effect runtime on variance and mean average turnaround time J.3 Number of replications The effect of three numbers of replications are tested on the mean and variance. The tested scenario is the base case with a runtime of 100 days and 21 warm-up period. The results are shown in Table 57 and Figure 85. It shows that both the variance as the mean is increasing when running 40 instead of 20 replications. This results is counter intuitive and can only explained by that by using 20 replications the outcome space is not represented well. The average turnaround time is statically different when comparing 20 replications with 40 replications. Comparing 40 with 60 replications the means only differs slightly. The difference is not statistically different. Also the half width of the confidence level interval is around the same. The decision how many replications should me made is a trade-off between the available computational power and the width of the estimated confidence intervals. In this case it is also dependent on the observation that 20 replications does not cover the whole solutions space, lacking the right accuracy of the estimations. 40 replications will be executed of each scenario as it produces reasonable confidence intervals with acceptable computational power needed. 134

150 Table 57 95% confidence interval for different number of replications Number of replications 95% confidence interval for average turnaround time ± ± ± 0.27 Figure 85 - effect number of replications on mean and variance average turnaround time 135

151 K. Output results In this appendix some output results are discussed in addition the results in chapter 7. The discussed results support claims made in chapter 7 further. K.1 discusses the effects of information exchange on the waiting time at the terminal and at the number of gangs that are nominated. In K.2 the statistical test for the customers facility experiment is executed. K.3 discusses the statistical test for the experiments about the use of buffers. Finally, K.4 shows the additional results for the growth potential of an inland port. K.1 Additional effects improved information exchange Figure 86 shows boxplots regarding the waiting time at the terminal for the different scenarios. It shows that an improved accuracy of the ETA has limited effect waiting time compared to the base case. The differences are not statistically significant. Real time sharing of the berthing window reduces the average waiting time with around 30%. This difference is statistically significant. Figure 86 - Effect improved information exchange on waiting time Figure 87 shows the effect of the information exchange alternatives on the number of gangs (workforce) is that is being nominated. It shows that the median of the average number of gangs that is being nominated decreases slightly the accuracy of the ETA is improved. However the difference is only statistically significant when the ETA is exact the ATA. When the berth window is shared and the accuracy of the ETA is improved with 25% or more, there are statistically speaking less gangs nominated. Because less gangs are being nominated the effect of improved information exchange on the ship turnaround is reduced. 136

152 Figure 87 - effect improved information exchange on workforce nominated K.2 T-test customs experiment This section describes the statistical testing of the customs experiment in section Table 58 shows that the variance of the scenario with and without customs facilities at the inland port is homogeneous. Table 59 shows that the differences in the mean of the average turnaround time is mot statistically different. Table 58 - Levene's test for homogeneity of variance customs experiment Degrees of freedom F-value P-value Table 59 - Two sample T-test for customs experiment Degrees of freedom T-value P-value K.3 T-test Buffers In this section the statistical significance of the results obtained from the experiments about using buffers in the deep-sea terminal is discussed. These results are discussed in section Table 60 shows that for the current volume the variance in the scenarios with and without buffers is not homogeneous. Table 61 shows in this case the results are also statistically different. Table 62 shows that when more volume is handled the variance in both scenario is homogeneous. Table 63 shows that also when more volume handled, the results are still different statistically speaking. Table 60 - Levene's test for homogeneity of variance of the use of bothers with current volume Degrees of freedom F-value P-value

153 Table 61 - T-test for difference in mean using buffers with current volume Degrees of freedom T-value p-value E-07 Table 62 - levene's test for equal variance of using buffers with high volume Degrees of freedom F-value P-value Table 63 - T-test for difference in mean using buffers with high volume Degrees of freedom T-value p-value E-03 K.4 Additional results growth potential inland port Figure 88 shows the extensive results of the information exchange strategies and the number of additional handled volume on the ship turnaround time. Compared to Figure 37 some additional insights are gained. When neglecting statistical significance, 2.5% of the transhipment have to be stored at the inland port in order to have the same turnaround time as the base case when handling 2.5% additional volume. It can also be concluded that the effect of 2.5% transhipment container is limited compared to only handling import and export containers at the port. Figure 89 shows that all researched scenarios have the same or a higher productivity than the basecase. Figure 88- effect different exchange strategies and handled volume on average turnaround time 138

154 Figure 89 - effect exchange strategy and volume on productivity gangs 139