Material Flow Simulation: Integration of new techniques into the simulator d³fact

d³fact is a material flow simulator which focuses its attention on an interactive and immersive 3D-visualization of the simulation, as well as a freely extendable simulation kernel based on Java. Newly developed methods in the areas of simulation methodology and computer graphics were prototypically implemented this way in recent years. The efficiency of these methods was also proven.

Aggregated visualization of multiple parallel simulations

Stochastic properties (e.g. random processing times) of simulation models create a need for the extraction of expressive data from several simulation runs. However, when visualizing a simulation, only the outcome of one simulation run is considered; the observed behavior of the model cannot be used to gather expressive results. To overcome this disadvantage, we execute several simulations synchronous on the nodes of a PC²-Computercluster and aggregate the visualizations of the simulations in a single display window. Therefore, a “visual mean value” is generated. An additional advantage is the possibility to collect the data of all simulations at runtime, allowing for the delivery of statistical processed data (e.g. usage rates, downtimes …), which otherwise would only be made available after sequentially performing all simulations.

3D-visualization of the automated motion planning
Aggregated visualization in 3D

Knowledge-based material flow control

In the scope of the project “Modellbasierte Methoden zur echtzeitnahen Adaption und Steuerung von Distributionssystemen”(model-based methods for close to real-time adaption and control of distribution systems), which is supported by the German Research Foundation, the „Business Computing, especially CIM“ workgroup researches innovative methods in the area of dynamic distribution logistics in cooperation with the TU Dortmund.

On the level of network adaptivity load-dependent modeling and measurement frameworks as well as simulation-based methods for decision support have been developed. On the intralogistic level an event orientated and adaptive control of the material flow by means of automated guided vehicles (AGVs) is being focus. Due to the demand fluctuations in the distribution network and uncertainties in the real process, a single and in advance calculated scheduling of transport vehicles can seldom be executed as expected. For this reason, a technique was developed, which analyzes control rules based on current system characteristics and therefore is able to react to the present situation. First the method was applied to the problem of path finding for AGVs within a distribution center. It was possible to achieve significant improvements over traditional path finding algorithms. To preserve the greatest possible flexibility, the method only plans incremental path sections. For each step, a control rule for the further path calculation is chosen. These control rules represent a necessary prioritization between two transport vehicles. The critical selection of the optimal control rule for the current situation is conducted by help of a knowledge base. A knowledge base, which is built before the operative deployment that inductively learns from situations met and solved across the simulations. These situations are called training examples. It is possible to consider events, which occur shortly after, and all other information of the system status in an offline calculated solution. The training examples are generated in such a broad fashion, that situations, which occur in practice, have already identified the system, so that the optimal control rule can be chosen. The advantage of this method is that the selection can be made quickly by relying on the knowledge base, while it is still possible to include complex calculations and long time horizons in the decision process.

First results show that the method is able to use a growing number of training examples – the correct classification rate is on the rise – and the adopted control rule can be selected in milliseconds depending on the situation.