Simulation and Logistics

Size: px
Start display at page:

Download "Simulation and Logistics"

Transcription

1 Simulation and Logistics Rommert Dekker Professor of Operations Research Introduction Many Cases: Port, container stacking Elevator control Inventory control (optional) Conclusions 1

2 Logistics and transportation Transportation ships, planes, trains, elevators Systems need to be designed, planned and controlled. For a good evaluation of options in the design simulation is needed. For a good evaluation of policies in operation: simulation is also needed Logistics also concerns location of warehouses, cooperation between companies and control of inventories. For a good evaluation of operating policies one needs simulation. 2

3 Examples Dekker Port of Rotterdam Chemical plant design of jetty (how many mooring points) this afternoon Vopak can another jetty be incorporated? Vopak Tank terminal, bottleneck analysis Chemical plant redesign of supply chain: shift from truck to barge Container terminal ECT 3

4 Waiting times at a tank terminal Problem - too long waiting times for tank trucks at a tank terminal in Rotterdam Port Approach - measure times (per truck) for a month First alternative - an extra weighting bridge Our alternatives - work through during lunch and or give customers who do not have to be weighted priority Simulation over one month Standard operations Extra weighting bridge No lunch stoppage Priority to weighting bridge Avg throughput time 1h51 1h46 1h49 1h48 4

5 Waiting times at a tank terminal II Conclusion after studying problem in detail: Increasing bottleneck capacity does shift bottleneck for weighting bridge to loading points. Main issue is that some customers (in France) suddenly send many trucks to terminal to pick up goods. A solution can be provided in another direction: using tank containers and filling them the day / night before they come. 5

6 Examples Europe Combined Terminals ECT ECT used to be the place where people build huge simulation programs. For a long time they only used MUST, an add-in to Turbo Pascal developed at TUD. Why? They had very large and complex simulations which did not fit in commercial simulation packages. Examples Stacking (twice) AGV routing: how will we control the movement of AGVs Berth planning: which ship will be where on which jetty Scheduling jetty and stacking crane activities Inter Terminal Transport: which transport equipment to use? 6

7 7 Rotterdam s Maasvlakte terminals

8 8

9 9

10 1

11 Evaluation automation Advantages Automated material handling pays off in case of rather standardized operations Less damages, safer operations Less people needed, smaller dependence on crafted personnel, existing personnel can be paid better Disadvantages Complexity of software and operations Very expensive software people are needed Much less flexibility in operations Example: ECT operates only one stacking crane per stacking lane - this limits throughput in case of reshuffles 1 Trend: new automated terminals in Thamesport, Hamburg

12 vessel loaded AGV Quay Cranes AGV area empty AGV transferpoints seaside stack lanes, each with a single ASC 1 transferpoints landside

13 Some operational AGV problems AGVs hinder each other in their movements because of their size when making turns: this causes traffic problems. As a result AGVs may arrive too late at the quay crane Sending the AGV earlier on the yard increases the traffic and hence the congestion. AGVs always need a crane to start or finish their operation. This requires a coupling of tasks between stacking cranes and AGVs. This may even create deadlocks. Ex. One container at an AGV needs to move into a stacking lane operated by one crane and one needs to move out. In what order will this take place if the move out has priority? 1 All requires complex simulation models (much work by Ottjes)

14 Classification of stacking problems Design phase - selection of stacking equipment, manual or automatic - determination of stacking layout ( length, width, number of lanes, position of reefers) - determination of stacking height (2,3,4 or higher) Operational phase - for any incoming container: where do we put it - which handling equipment is going to pick up the container Tactical phase stack reorganization (rather costly) 1

15 Information available If at the moment of entry of a container, the exact moment of departure as well as departure modality (ship n+ name, rail(car) or barge) is known, then almost all reshuffles can be avoided. In practice only the expected departure mode (ship name, rail, barge or truck) is known. This info may change. Notice that the stack throughput capacity depends on the residence time of containers Ship-ship containers enter and leave stack at same side, yet ship-other modality enter and leave stack at different sides! 1

16 Stacking optimization Usually done with sophisticated and complex simulation programs which model (=simplify) the operations at a terminal Problem: how to validate the model (= trust outcomes) many assumptions are made which are not visible to outsiders Animation shows movements, but not plans (=strategy) Requires experts to interpret them Experience at Erasmus University Development takes much time (6 months or more) If 3 groups independently develop programs, outcomes may all differ (by more than 20%) 1

17 1 Typical results of simulation experiments A. B. C. D. E. F. G. H. Reshuffle occasions Jumbo Deepsea Shortsea/feeder Export Truck Rail Barge Import Reshuffles Jumbo Deepsea Shortsea/feeder Export Truck Rail Barge Import No position for new No position reshuffle (per 1000)

18 Stacking some insights Better information reduces stacking problems. Most problems on the individual collections of land containers by trucks If shippers or consignees can indicate (the day) when containers will be picked up, then many reshuffles can be avoided. Alternatively, much can be gained if instead of individual containers one of a group should be collected. 1

19 Other logistic cases Punctuality railways Simone- this afternoon Prison cells Cell net (analytic) / Cellsim (simulation) - assess balance between capacity and home sendings in penitentiary system Maintenance of highways concept evaluation Elevators operation improvement Medical simulation Performance of different heart valves Model for waiting times at border check points at Schiphol 1

20 General issues Models become complex, take a lot of time and are difficult to validate / manage. Yet savings can be very large! For new problems often new models have to be built. Packages have templates, but to what extent do they work? Changing existing models can be quite costly! Simulation versus analytical methods: brothers or opponents? 2 Use package or low level language? Depends who you are and how important problem is! - if problem is complex and important hire professionals - if problem is simple and fits in toolkit use package - if problem is simple and a cheap solution is wanted use Visual B or any other low language with which you are familiar.

21 Analytics vs simulation Analytic models give more insight, can be more accurate, can be faster development can take a lot of time analytic methods can better be incorporated in other models, and allow for optimisation Simulation is more flexible (more situations can be handled) development can take a lot of time animation and graphics convinces simulation packages can be expensive outcomes are subject to stochastic variation simulation usually only does evaluation: black box optmisation is more difficult, IPA estimators require complex theory 2 However, let us not fight but use methods combined!

22 Trends as tool to verify analytical calculations reintroduce analytics in simulation simulation optimisation methods (e.g. OPTQUEST) analytical meta models, explaining and validating simulation models (Kleijnen) simulation from strategic to operational control tool games (supply chain: beer game) virtual reality tools / simulators (port, aircraft, car) simulation over Web simulation within SAP with historic data simulation with intelligent agents 2

23 Simulation in science Solve a problem with simulation does not lead to a scientific publication Need to improve a method or come up with general observation Development of analytical method is much easier to publish Simulation journals reasonably low ranked compared to general OR journals 2

24 Elevator simulation Issue: too long waiting times for the elevators in Woudestijn s Hoogbouw. What can be done about it? Student project with Dekker as leader and 11 students. Collect data, develop and implement simulation models and draw conclusions. Problem: how to get data on travel demand Solution: own measurements Issue: which language to use. Solution: VB, C++ and Arena were tried out. 2

25 Simulation inventory control of slow moving parts methods: - existing reorder points - the (s-1,s) model with exponential times between demands - idem with Erlang-4 times between demands - bootstrapping method (sample randomly from actual demand for a period equal to leadtime) with µ + 3σ as cut-off for exceptional demand experiment items were selected related to equipment use two years of data (1998, 1999) to determine reorder point test method on data from last year (2000) Use service levels 99% for High, 95% for Medium and 90% for Low important items. Determine demand rate from previous yrs 2

26 Experiments - slow movers II (s-1,s) Poisson vs manual: 36% more costs, 2% more stockouts cat 1: manual ROP higher 25% of materials, avg 3x higher, costs of (s-1,s) 26% lower, but 64% more stockouts cat 3: (s-1,s) ROP higher for 52% of materials, avg 2x higher, costs 97% higher, but 51% less stockouts (s-1,s) Erlang 4 vs manual: 7% less costs, 58% more stockouts cat 1: manual ROP higher: 50% of materials, avg 3x higher, costs of Erlang-4 25% lower, 107% more stockouts cat 3: Erlang 4 ROP higher: 8% of materials, avg 2x higher, costs Erlang-4 90% higher, 42% less stockouts 2 bootstrapping vs manual: 7% more costs, 10% more stockouts

27 Conclusions Simulation important tool for logistics and transportation is more and more necessary for complex logistic operations, especially in case of automation. Simulation can be time consuming, complex and requires high-level experts. Yet we can not without them (e.g. Schiphol baggage system, train punctuality) 2 More research needs to be done for - model re-use - model specification - insights into model building - combining simulation with analytical methods (optimisation, meta models, etc.) - use simulation as operational tools