Simulation model of dynamic AGV dispatching.
Report 2001.LT.5493, Transport Technology, Logistic Engineering.
Automatic Guided Vehicle systems have found a wide range of application.
The design of an AGV system affects the performance of the system and the
design can be very complicated. A number of interrelated decisions must be
made in order to exploit the full benefits of the system. An important
aspect is vehicle dispatching, selecting the right vehicle for performing
the transportation task.
In this report a bid-based dynamic dispatching process of move requests in
a simulation model is presented. The Bid-Based Dynamic Dispatching Rule is
a dynamic task assignment rule. As soon as a transportation task is
received, all the AGVs in the system are allowed to 'bid' for this move
request even when they are still busy performing another task. After
receiving all the bids the AGV with the shortest travel distance will get
the move request in its 'joblist'.
The AGV system has a network layout. In a network AGV system, each vehicle
is allowed to orbit in any track and each station can be pickup or
delivery point. The distance between two nodes situated next to each
other is the same for every two nodes. The lanes are perpendicular to
each other and have the same length. The path layout of this model
consists of multiple lanes. Every arc consists of two lanes and every lane
has its own direction The location of the pickup and delivery points is at
the nodes. Every node can be a pickup or delivery point of a move request.
The performance indicators implemented in this model are related to the
execution of the move requests and the performance of the AGVs. The
performance indicators are:
Several input parameters can be used to evaluate the performance of the
system. The three major input parameters are:
- The total of generated move requests in relation with the total
performed move requests;
- The waiting time of a move request before an AGV is at its staring node;
- The percentage of move requests that are committed by the first time;
- The total time a move request is in the system.
Simulation runs have been made by using different seeds for the uniform
distribution the define the starting nodes and end nodes of the move
requests and different seeds for the exponential distribution that is used
to define the time between two generated move requests. This resulted in
the conclusion that the system is very sensitive for deadlocks. Therefore
experimental runs with different threshold values and amount of returns
could not be not be properly evaluated.
- the seeds of the system;
- the threshold value;
- the amount of returns to the dispatcher.
In this simulation model only the Bid-Based Dynamic Dispatching rule is
implemented. The results of this dispatching rule should be compared with
the results that can be obtained by other dispatching rules. These other
dispatching rules should be first implemented before interesting
evaluations of the performance of the dispatching rules can be made.
This simulation model can be expanded with the implementation of a
deadlock recovery procedure, more dispatching rules and more detailed
Reports on Logistic Engineering (in Dutch)
, TU Delft