MMeAS

DFG project MMeAS: Model-based methods for contemporary adaption and control of distribution systems

In distribution logistics, the increasing dynamism of markets poses a tremendous challenge, which can just be fought with an ascending dynamic of transaction processes and of structures. As part of the DFG project „model-based methods for contemporary adaption and control of distribution systems“ does the research group of business computing, especially CIM, research in cooperation with the TU of Dortmund on two levels on innovative methods in the field of dynamic distribution logistics. Load-oriented modeling and evaluation frameworks such as simulation-based methods for decision support are developed at the level of networkadaptivity.

On the intra logistical level, for example for a single distribution center, an eventoriented and adaptive control of material flow in terms of automated guided vehicles (AGV) is on focus. Such vehicles are used in this case in order to solve given transportation problems (storage and retrieval or provision of material) under time constraints very well.

Thanks to fluctuations in demand in distribution network and uncertainties in real process, a once pre-computed statistic scheduling of vehicles can rarely be run as expected. Because of this reason, a simulation-based method has been developed which breaks down on the basis of current system features control rules and so it is able to react on the momentary situation.

Knowledge-based material flow control

First, the method was applied to the problem of path finding of driverless transportation vehicles within a distribution center. There, it is especially important what transportation orders are currently processing and therefore it is possible to react appropriately to disturbances. Fundamental progresses in comparison to traditional path finding algorithms could be achieved. In order to preserve maximum flexibility, plans the procedure just incremental partial paths with already developed methods of computer science. In every step a control rule has to be chosen for further path computing. These control rules represent necessary prioritizations between two vehicles, which would otherwise interfere.

The critical selection of the in the current situation optimal control rules happens with help of a knowledge-based method. Therefore, a knowledge base, which can inductively learn of the in the simulation met and dissolved situations, the so called training examples, is built up before the operative employment. In the offline computed solution, there short time later occurring events and all other information of the system state can be considered. The training examples are wide ranging generated, that the systems already knows in the praxis occurring situations and then the optimal control rule can be elected.

The advantage of this method is the fact that the selection can be quickly realized by drawing on the knowledge base, nevertheless complex calculations and great time horizons flow in the decision. Implementation and results. The material flow simulator d³FACT insight, which has been developed of the research group of business computing, especially CIM and of the research group of algorithm and complexity, has been enlarged by one component for the implementation of the procedure. Methods of artificial intelligence have been used for representation of knowledge, namely neural networks and decision trees. By using a 2D visualization it is possible to represent transportation vehicles in a symbolic way and to display their path calculations.

First results show that this method can use an increasing number of training examples – the correct classification rate is rising – and within milliseconds does it depending on the situation chose the learned control rule.