Modeling autonomously controlled automobile terminal processes

Görges, Michael; Freitag, Michael Conference Paper Modeling autonomously controlled automobile terminal processes Provided in Cooperation with: Hamburg University of Technology (TUHH), Institute of Business Logistics and General Management Suggested Citation: Görges, Michael; Freitag, Michael (2019) : Modeling autonomously controlled automobile terminal processes, In: Jahn, Carlos Kersten, Wolfgang Ringle, Christian M. (Ed.): Digital Transformation in Maritime and City Logistics: Smart Solutions for Logistics. Proceedings of the Hamburg International Conference of Logistics (HICL), Vol. 28, ISBN 978-3-7502-4949-3, epubli GmbH, Berlin, pp. 186-214, https://doi.org/10.15480/882.2497 This Version is available at: https://hdl.handle.net/10419/209393 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-sa/4.0/ Proceedings of the Hamburg International Conference of Logistics (HICL) – 28 Michael Görges and Michael Freitag Modeling Autonomously Controlled Automobile Terminal Processes Published in: Digital Transformation in Maritime and City Logistics Carlos Jahn, Wolfgang Kersten and Christian M. Ringle (Eds.) September 2019,epubli CC-BY-SA4.0 Modeling Autonomously Controlled Automobile Terminal Processes Michael Görges1 and Michael Freitag2 1 – BLG LOGISTICS GROUP AG & Co. KG 2 – BIBA – Bremer Institut für Produktion und Logistik GmbH Purpose: Automobile terminals play an essential role in automotive supply chains. Due to short planning cycles and volatile planning information, the yard assignment determines terminals performance. Existing planning approaches are not able to cope with these dynamics. This contribution proposes a novel bio-analogue autonomous control method to face these dynamics, its effects and to improve the terminals performance. Methodology: Causes of internal and external terminals dynamics will be discussed and an autonomous control method will be derived. A generic 185arameterizable automobile terminal model and its implementation to a discrete event simulation will be introduced in this paper. This simulation is used to compare the new approach to classical yard assignment. Findings: This paper contributes to the theoretical understanding of causes and effects of dynamics in the context of automobile terminals. It will show that autonomous control outperforms classical approaches under highly dynamic conditions. Originality: The generic modelling approach is a novel description of automobile terminals. It allows investigations of a broad spectrum of use cases. Moreover, the bio-analogue autonomous control for automobile terminals is an innovative approach. Keywords: Automobile Logistics, Port Terminal, Autonomous Control, Discrete Event Simulation First received: 08.May.2019 Revised: 27.May.2019 Accepted: 11.June.2019 186 Michael Görges and Michael Freitag 1 Introduction During the recent years, the shipment volume of finished cars increased constantly, due to an emerging global interconnection between production and distribution networks. In this context, automobile terminals are central elements in international automotive supply chains. Automobile terminals allow the transshipment from the production plant to the target markets. Besides handling of finished cars from different transport modes (e.g., ship or truck), these terminals usually offer a broad spectrum of additional technical services in order to meet customers’ demands in the port of destination. In general, the main tasks of automobile terminals can be defined as handling, technical treatment and storage of finished vehicles (Mattfeld, 2006; Böse and Piotrowski, 2009). All related processes are triggered directly by the car manufacturers (OEM). Accordingly, automobile terminals can be interpreted as a classical decoupling point in the automotive supply chain, which allows to react flexibly to demand fluctuations (Dias, Calado and Mendonça, 2010). Hence, planning of processes automobile terminals is faced with forecast-driven and customer-order driven processes at the same time. This strongly affects the yard master planning, which aims at minimizing the distance between the point of car entrance, storage area and its exit point (Görges and Freitag, 2019). Classical master planning approaches solve this task by assigning predefined parking areas to the different vehicle types (e.g. sorted by manufacturer, model and destination). This leads to good planning results for situations with high forecast quality and less dynamics. However, due to its long term orientation, this type of yard Modeling Autonomously Controlled Automobile Terminal Processes 187 master planning is prone to forecast deviations, volatile parameter variations and unforeseen events, which may affect the terminals performance negatively (Cordeau et al., 2011; Mattfeld and Orth, 2006). Autonomous control of logistics processes address these shortcomings by transferring decision making capabilities from a centralized planning instance to the logistics object itself. Due to interactions and decision making of intelligent logistics objects, autonomous control aims at creating self-organizing systems behavior, which increases the systems performance (Windt and Hülsmann, 2007). This self-organization can be seen as emergent behavior of a complex dynamic system, which is not a characteristic of the systems elements but of the total system (Vaario and Ueda, 1998). For production logistics, different autonomous control strategies showed already their operational potential. In the context of automobile terminals, first implementations indicated promising results concerning the assignment of cars in import processes to technical service stations (Böse and Piotrowski, 2009). However, comprehensive autonomous control strategies covering all inbound and outbound material flows of an automobile terminal are still missing. Thus, this paper will focus a broader use case. It will derive an autonomous control strategy, which allows the integration all flows of cars (import, export and inter terminal) at an automobile terminal. In order to analyze the performance of the autonomous control strategy, this paper will present a generic modeling approach for investigating a broad range of related scenarios. Furthermore, it will introduce a discrete event simulation model implementation for analyzing these scenarios. This simulation model will be used to investigate the performance of the derived autonomous control method compared to a classical yard master plan. 188 Michael Görges and Michael Freitag 2 Autonomous Control of Automobile Terminals 2.1 Terminal Planning Material flows in automobile terminals can be characterized as a sequence of several generic sub processes (e.g. loading or storage operations). Basically, every process starts with unloading operations from different transport carriers (truck, rail, ship) followed by the storage of the vehicle. Subsequently, cars are loaded to outbound transport carriers or receive one or more technical services. Automobile terminals offer a broad spectrum of technical services with highly varying process times (Hoff-Hoffmeyer-Zlotnik et al., 2017). Figure 1 depicts this physical material flow of vehicles at an automobile terminal. Furthermore, it shows the related planning tasks in respect to their temporal occurrence (planning horizon). The overall objective of all planning tasks is the efficient operation of all physical vehicle movements from the source (i.e. unloading point at the terminal) to the sink (i.e. loading point)(Özkan, Nas and Güler, 2016). On a strategic level, planning focuses on long term decisions like the planning of infrastructure (e.g. additional berth or yard extensions). Forecasting of expected vehicle volumes and related long-term planning of resources belong to this strategic time horizon as well. Based on these forecasts, a long term orientated area master planning derives required parking areas (Mattfeld, 2006). A result of this planning step is a rough assignment of estimated vehicle volumes to parking areas. This first assignment is the starting point for the tasks on the tactical planning horizon. In this planning phase, forecasted vehicle volumes are used to plan berths and the utilizations of berths (Dias, Calado and Mendonça, 2010). Usually, forecasts be- Modeling Autonomously Controlled Automobile Terminal Processes 189 come more precise and get a higher level of detail with more specific information (e.g. model-destination split or volume related model-split). The results of the strategic planning is used in the tactical planning to generate and adjust the yard plan. The yard plan comprises the assignment of vehicle volumes to specific areas of the yard. In order to generate short routes between the loading and unloading locations, the yard planning often includes the localization of loading and unloading operations (berth allocation planning, storage space partitioning and storage area design) (Mattfeld, 2006; Mattfeld and Orth, 2006). The personnel requirement can be derived with the results of localization and vehicle assignment. In general, the operational planning is characterized by increasing level of relevant information (e.g. ETA of ships or the assignment of cars to ships). On this operational planning level, the results of tactical planning are refined in predefined turns or with a rolling time horizon (Mattfeld and Kopfer, 2003). This approach allows to react to changes and external disturbances (e.g. delay of ships or changes in ships transport quantities). 190 Michael Görges and Michael Freitag Figure 1: terminals planning task – based on (Görges and Freitag, 2019) These plan adjustments may lead to changes of routes length between inbound locations, storage areas and outbound locations and affect the overall terminals productivity and the personal requirements. The process control focuses on the execution of particular driving orders resulting from the previous planning tasks. It assigns driving orders to workers and monitors the progress of order processing Yard planning plays a key role in the described, cascaded planning process. It mainly determines driving distances between cars’ arrival and departure points and the related process productivity (i.e. cars per hour per worker). Incoming vehicles are sorted and assigned to parking lots according to the yard master plan. At the arrival of a vehicle, usually the information about its outgoing transport carrier is not available. Later, the customer (e.g. OEM) sends advices for particular cars, assigning them, for example, to a specific ship. Dias et al. (2010) describe this characteristic as parallel push rail ship truck rail ship truck Terminal material flow outbound processes storage and technical treatment Vehicle take-over vehicle stoarage technical services vehicle take-off physical material flow inbound processes process planning and control process control sequencing of driving orders assignment of driving orders carrier assignment carrier assignment Planning task operational planning tactical planning strategic planning yard assignment personal and resource planning planning of area and technical resources berth planning berth assigment personal and resource planning forecasting and volume planning yard master planning planning of infrastructure capacity planning yard planning unloading localization berth assigment personal and resource planning berth planning unloading localization Modeling Autonomously Controlled Automobile Terminal Processes 191 and pull processes occurring at the same time at an automobile terminal (Dias, Calado and Mendonça, 2010). These parallel push and pull processes allow terminals to react quickly to changing demands in the supply chain. However, this also leads to complex internal dynamics in the terminals processes and short planning time horizons. Classical yard planning addresses the orders’ neutral (forecast-driven) aspect. Volumes of vehicles are assigned to specific parking areas of the terminal based on forecasts. After customer orders are available, the operational planning (e.g., berth planning) aims at increasing the terminals productivity by reducing distances between storage area of the cars and the outgoing transport carrier (e.g., by assigning ships to quay positions). Figure 2 depicts both push and pull processes of automobile terminals and relates them to the planning tasks. In this context, terminals offer a higher degree of flexibility to the entire supply chain at expense of an increasing complexity of the terminals’ planning and its operative process execution. In this context, the yard planning is a key instrument to cope with forecasted vehicle volumes and to allocate it to parking areas. Accordingly, it determines routes of vehicles from the source to the sink on the terminal. Due to the order-natural nature of the arrival process, the yard planning cannot react to near-term changes (e.g., increasing or decreasing vehicle volumes). An increasing degree of flexibility and dynamical adjustment of yard assignments may increase the terminals’ performance (Görges and Freitag, 2019). 198 Michael Görges and Michael Freitag Figure 4: arrivals of OEM 1 & OEM2 (top); total arrivals (bottom) Table 2: inventory times OEM 1 - M1 OEM 1 - M2 OEM 2 - M3 OEM 2 - M4 avg. inventory time [d] 10 20 10 20 variance [d] 2 2 2 2 The departure of vehicles is modelled in two different variants. The first variant uses simple constant inventory times modelled by adding a normal distributed delay to the arrival time of each vehicle. Accordingly, cars leave the terminal after a predefined time. Table 2 summarizes the underlying departure rates. Modeling Autonomously Controlled Automobile Terminal Processes 199 The second variant models the departure of cars in a more realistic way. In this variant ships sailing to destination D1 and D2 are generated as a time series with a normal distributed shipment volume per vessel. Figure 5: inventory over time: avg. 600 cars per ship (top); avg. 2000 cars per ship (bottom) for bulked departures The ships arriving at the terminal have an average capacity based on a normal distribution. This leads to a bulked departure of cars over time. In this scenario, the average ships' capacity 𝑠 will be varied from 500 to 2000 with a standard deviation of 10% of the mean value. Figure 5 shows the estimated inventory over time for different mean ships capacities. It shows that the mean ships parameter 𝑠 has an impact on the dynamic of the inventory time series. Comparable inventory curves can be observed in real automobile terminals. This scenario comprises according to the sinusoidal inputs, the initial terminals inventory and the output rates (see Table 1 and Table 0 500 1000 1500 2000 2500 3000 50 100 150 200 250 300 350 inventory [cars] time [d] OEM 1 - M 1 OEM 2 - M 3 0 500 1000 1500 2000 2500 3000 3500 4000 50 100 150 200 250 300 350 inventory [cars] time [d] OEM 1 - M 1 OEM 2 - M 3 200 Michael Görges and Michael Freitag 2) approximately 127.000 vehicles running through this scenario in 365 days. In the 4x4 scaled scenario there are three sources and three sinks. The locations of sources and sinks will be addressed in detail in section 4.1 (Figure 6 summarizes their locations). The split of outgoing volumes of both OEMs is modelled as follows: At source 1 75% of OEM 1's volume arrive. At source 2 25% of OEM 1's and 25 % of OEM 2's volumes arrive and at source 3 75% of OEM 2's volume arrives. 75% of all ships sailing to destination 1 leave from sink 1, 20% from sink 2 and 5% from sink 3. For destination 2 75% leave from sink 3, 20% from sink 2 and 5% from sink 1. 4 Yard Assignment Methods 4.1 Conventional Yard Assignment Based on these information a simple planning and assignment of cars to parking areas has been done. Figure 6 shows these assignments. The main concern of this assignment is to generate short routes between sources, storage areas and sinks. For example most cars of OEM 1 will arrive at source 1 and leave at all sinks. Thus, the assignments are close to source 1. Usually, different models from one OEM may be mixed when the terminals utilization is high. Thus, Figure 6 shows the primary assignment of cars and a secondary assignment in brackets. The secondary assignment can only be used if no free row of the primary assignment is available. These assignments are considered as results of a classical planning process in the following evaluation. Modeling Autonomously Controlled Automobile Terminal Processes 201 Figure 6: classical yard assignment For the purpose of benchmarking a randomized assignment will also be used. In this case arriving cars are assigned to a randomly chosen row on the terminal. Only capacity restrictions of a row have to be met. 4.2 Pheromone Based Autonomous Control Approach The autonomous control method presented in this section allows cars to evaluate, to compare and to choose a parking row by a pheromone based approach, which is inspired by ant's natural foraging behavior. As depicted earlier, bounded rational strategies like this offer the possibility to consider many different decision parameters. The method at hand can be seen as a combined method, using bounded rational aspects and rational measures. Pheromone based approaches have shown their capability to react on dynamical changes and to stabilize the systems behavior under volatile conditions (Windt et al., 2010). Accordingly, this approach seems to be suitable OEM 1 M2 󰇛OEM 1 –M1󰇜 OEM 1 M1 󰇛OEM 1 –M2󰇜 OEM 1 M1 󰇛OEM 1 –M2󰇜 OEM 1 M2 󰇛OEM 1 –M2󰇜 OEM 1 M2 󰇛OEM 1 –M1󰇜 OEM 1 M2 󰇛OEM 1 –M1󰇜 OEM 1 M1 󰇛OEM 1 –M2󰇜 OEM 1 M2 󰇛OEM 1 –M1󰇜 OEM 2 M3 󰇛OEM 2 –M4󰇜 OEM 2 M3 󰇛OEM 2 –M4󰇜 OEM 2  M3 󰇛OEM 2 –M4󰇜 OEM2 M3 󰇛OEM 2 – M4󰇜 OEM 2 M4 󰇛OEM 2 –M3󰇜 OEM 2 M4 󰇛OEM 2 –M3󰇜 OEM 2 M4 󰇛OEM 2 –M3󰇜 OEM 2 M4 󰇛OEM 2 –M3󰇜 sink 3 sin k 2 sin k 1 source1 source2 source3 202 Michael Görges and Michael Freitag to the vehicles yard assignment at an automobile terminal. In general phenome based methods imitate communication principles of social insects (i.e. ants). While searching for food, ants leave evaporating pheromone trails, marking possible routes to food sources. Other ants are attracted by these trails and follow it. Ants following a trail increase the pheromone concentration. The pheromone concentration decreases in time due to the natural evaporation process. By using this interplay between marking trails with pheromones on the one hand and natural evaporation process on the other hand, ants are able to find the shortest routes to food sources. Autonomous control methods using this principle leave relevant information in the system (e.g. throughput times) as an artificial pheromone. Subsequent objects are able to read this pheromone information to make a local decision on this basis and to follow the trail with the highest concentration. The evaporation process is often modeled as a moving average over a predefined set of objects running through the system (Armbruster et al., 2006). This paper proposes a similar approach for assigning vehicles to parking rows. Vehicles belonging to a category 𝑘 calculate for every row 𝑖 a pheromone value 𝑃 and chose the row with the best 𝑃 value. Equation (2) describes this pheromone value. The total number of vehicle categories is defined as K. In this context criteria for vehicle categories are OEM, model types and the shipment destination (see also Table 1). 𝑃𝛾󰇛󰇜 󰇛󰇜    𝛾  𝛾󰇡1 󰇢𝛾 󰇛󰇜  󰇛󰇜 (2) The pheromone value 𝑃 consists of four terms. Each term focuses on a different target value and can be weighted by a factor 𝛾. Except from term 3 all remaining terms use the moving average concept to emulate the pheromone evaporation. All terms and the evaporation process will be described in the following. For each category 𝑘 a moving average of the last 𝛼 vehicles Modeling Autonomously Controlled Automobile Terminal Processes 203 is used to determine two key parameters. The first parameters are the most frequented sources and sinks of the specific vehicle category. These parameters are the basis for deriving distance related measures like 𝑊. The 𝑊 is defined as the distance between the most frequently used source, the storage area of the parking row 𝑖 and the most frequently used sink. The second parameter is the moving average of the inventory time (days at the terminal) 𝐺 of the vehicles belonging to category 𝑘. The first term of the pheromone value equation (2) focuses on balancing the estimated distance 𝑊 and the average inventory time 𝐺 of a category 𝑘. The basic intention of this term is to rate rows with longer estimated distance better for categories with higher inventory time and vice versa. Therefore, this term calculates the ranking position of the estimated distance factor 𝑊 divided by the amount of parking Areas 𝐹 and relates it with the ranking of inventory day of remaining categories. Most of terminal inbound and outbound processes operate in a FIFO mode. Thus, vehicles with same inventory times should stand closely together. The second term addresses the FIFO principle, by relating the inventory time of the latest vehicle in a storage area with the inventory time of the oldest vehicle of category 𝑘. The third term addresses the split of vehicles on the terminal. An obvious constraint coming from the basic yard planning is to minimize the geographical dispersion of vehicles belonging to the same category. The number of different separated storage areas per category should be as less as possible. Therefore, this term relates the volume of vehicles of category 𝑣 in the parking area of row 𝑖 to the overall volume of vehicles 𝑉 belonging to category 𝑘. 204 Michael Görges and Michael Freitag The fourth term focuses on the estimated distance for a vehicle stored on the parking area of row 𝑖. It tries to avoid an assignment, which lead to long driving distances. This term is defined as the ratio between the estimated distance 𝑊 based on the moving average and the maximal possible distance for category 𝑘 regarding all sources, storage areas and sinks. The pheromone value for each row can be derived with equation (1). By contrast to natural process, vehicles choose the row with the lowest value of 𝑃 as the highest concentration of pheromones. 5 Simulation Results 5.1 Impact of External Dynamics An discrete event simulation model has been set up according to section 4. This model will be used to investigate the impact of external dynamics on the conventional yard assignment and the autonomous control method for the constant and the bulked departure variant. The parameter 𝜇 (amplitude of the arrival function) will be varied as a source of external dynamics (e.g., stronger seasonal effects by varying order volumes of customers). Higher values of 𝜇 lead to stronger variations and a more dynamic situation. In this experiment 𝜇 is the same for every category k in one simulation run. Modeling Autonomously Controlled Automobile Terminal Processes 205 Figure 7: Simulation results for varying amplitudes Figure 7 shows the average driving distance of all cars in a simulation run for the conventional planning, the pheromone based autonomous control method and the random assignment. The values of 𝛾 have been set to (𝛾0.1 and 𝛾0.4). The role of this parameters will be discussed later in section 5.2. As expected, the random assignment performs worst. Due to the random assignment possible short routes between source, storage area and sink are neglected. This leads to long driving distances. Figure 7 shows that this method is not affected by an increasing amplitude. By contrast, Figure 7 depicts a strong dependency between the conventional planning and the amplitude of the arrival function. A higher amplitude causes stronger peak periods with higher amount of arriving cars. In this situation cars are assigned to the pattern shown in Figure 6. The higher the incoming volume in a peak period, the more often secondary assignments (peak reserve) are used and occupy parking areas of other models (primary assignment) with potentially shorter routes. This leads to longer routes un400 450 500 550 600 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 avg. distance [m] Amplitude ( k ) [cars/d] conv. planning PHE-Mehtod rand assignment 206 Michael Görges and Michael Freitag der dynamic arrival conditions. Compared to the conventional planned situation the autonomous control method behaves different. Although, the average driving distance increases with higher values of 𝜇, this effect is slightly lower compared to the conventional planning. The autonomous control method is able to cope with the external dynamics more robustly. Regarding the absolute values, the autonomous control method outperforms the conventional planning for every 𝜇. This effect is stronger for higher values of 𝜇. Despite higher external dynamics the autonomous control method is able to find suitable row assignments with shorter routes. As described in section 3, the implementation of bulked departures can be seen as a source of additional dynamics. Figure 8 depicts simulation results for the scenario with bulked departures. For Figure 8, the mean vessels' capacity has been increased in steps of 100 cars per vessel (starting from 500 car up to 2000 cars per vessel). Every simulation run had a fixed mean arrival 𝜆 (see Table 1) and an amplitude of 𝜇 =50 cars per day in order to provide comparability with Figure 7. As already discussed, bigger ship capacities lead to longer inventory times. These longer inventory times affect the overall performance negatively. This can be confirmed by Figure 8. It shows that bigger vessels' capacity increase difference between conventional planning and the new autonomous control method. Like in the first scenario, the autonomous control method outperforms the conventional assignment. For vessels' capacity of 500 vehicles, the average driving distance is about 3.8% higher for the conventional planning. By contrast, this gap is for vessels' capacity of 2000 cars 11.35% higher compared to the autonomously controlled situation. In total, Figure 7 and Figure 8 confirm the hypothesis that autonomous control improve the terminals' performance under increasing external dynamics conditions induced by volatile demand Modeling Autonomously Controlled Automobile Terminal Processes 207 fluctuations (Figure 7) and varying bulked departures (Figure 8). Comparing both types of dynamics, the impact of varying amplitudes seems to be stronger than the vessels' capacity. Both sources of dynamics lead to differences in the internal systems' behavior for the autonomous control method and the conventional planning. Figure 8: Simulation results for vessel capacities Figure 9 confirms the impact of increasing dynamics on the conventional yard assignment and on the autonomous control method. It presents scatter plots for the pheromone based method and for the conventional planning. Each plot depicts the driving distance against the terminals inventory for different points in time in a simulation run. The terminals inventory is an indicator for the externally induced dynamics. In both cases (autonomous control and conventional planning) the systems inventory level is defined by the arrival and the departure function (see also Figure 4 and Figure 5). There is no influence of the control methods on the inventory over time. Thus, this measure can be seen as an indicator of external dynamics. In addition, Figure 9 presents the average driving distance related to the terminals inventory at the same time. The driving distance 300 350 400 450 500 550 600 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 avg. distance [m] vessel ca p acit y [ cars/vessel ] conv. planning PHE-Mehtod rand assignment 214 Michael Görges and Michael Freitag Rekersbrink, H., Makuschewitz, T. and Scholz-Reiter, B., 2009. A distributed routing concept for vehicle routing problems. Logistics Research, 1(1), pp.45–52. Scholz-Reiter, B., Rekersbrink, H. and Görges, M., 2010. Dynamic flexible flow shop problems—Scheduling heuristics vs. autonomous control. CIRP Annals, 59(1), pp.465–468. Toshniwal, V., Duffie, N., Jagalski, T., Rekersbrink, H. and Scholz-Reiter, B., 2011. Assessment of fidelity of control-theoretic models of WIP regulation in networks of autonomous work systems. CIRP Annals, 60(1), pp.485–488. Vaario, J. and Ueda, K., 1998. An emergent modelling method for dynamic scheduling. Journal of Intelligent Manufacturing, (9), pp.129–140. Windt, K., Becker, T., Jeken, O. and Gelessus, A., 2010. A classification pattern for autonomous control methods in logistics. Logistics Research, 2(2), pp.109–120. Windt, K. and Hülsmann, M., 2007. Changing Paradigms in Logistics — Understanding the Shift from Conventional Control to Autonomous Cooperation and Control. In: M. Hülsmann and K. Windt, eds., Understanding Autonomous Cooperation and Control in Logistics. Berlin, Heidelberg: Springer Berlin Heidelberg, pp.1–16.