Simulation based Realtime-Control of Automated-Guided vehicles - a decentral approach

[laro]

Simulation based Realtime-Control of Automated-Guided vehicles - a decentral approach
Prof. Dr.-Ing. habil. W. Dangelmaier & Dr. Christoph Laroque

The steadily rising worldwide container handling brings the operators of container terminal operations to improve the performance. Because of limited available space, this is done through a growing automation of the transportation and handling equipment. Efficient methods of planning and controlling the material flows are needed to exploit the potential of this automation, because they determine the utilization of automation solutions.
The automated guided vehicle (AGV) is an example of this progressive automation, especially for the horizontal transport of containers between ship and the intermediate store. The high dynamic and stochastic influences require certain real-time control procedures for these systems during operation in the terminals. The required response time is about one second for finding a new plan for the AGVs within the terminal. Analytical methods for decision making do not work in practical application. Here one can find systems, in which priority rules are often combined with other rules (production rules) in situation-specific applications. The problem here is the set of possible situations or states that are not fully taken into account. Through its local target reference (e.g., minimizing the waiting time of a ship's crane) and its fast execution systems are robust, but on the other hand they in return also "just" good local solutions. Also interaction rules, that consider upstream or downstream system elements, move the solution only towards a global optimum. The prospective impact of decisions are generally not considered.
Simulation is a proven approach to predict the future behavior of systems. For some time there are approaches of using simulation in an operational term, such as early warning systems simulation (see Hotz and Schulze, 2006) or the control simulation. The last concept means terms of scheduling procedures that describe the simulation by testing different rules (Multi-Pass Scheduling) and based on the simulation result it is expected that the best rule will be selected. Often it is spoken of real-time processes. However, they are simplistic assumptions and they are far away from the alluded response times above (see Kim, 2003). The procedures alluded above cannot simply be transfered on the control of AGV in container terminal. For simulation-control decisions, the simulation must break down the key system elements in a sufficient detail. This leads to a temporary problem. Existing alternatives of duration reduction does not lead to a sufficiently fast response time. A central control approach is not effective, because of the nature of AGV, because of the terms of certain distributed units and the resultant of high (time-) effort for the communication.
It is important to examine how a decentralized procedure is able - with incomplete information and uncertainty about prospective system states – to make good and tough management decisions. The procedure should develop ad hoc networks of event-oriented decisions and use this simulation to make decision alternatives with sufficient precision to test and evaluate. Choosing the simulation period as short as possible is always a good preview for the forecast horizon to keep the results not distorted. Hence the problem of assessing the quality of local decisions on the global system state appears.