Project

** Motivation ** The results of the predecessor project Isabella produced initial improvements in transport order processing and identified further potential for improvement. These included the integration of transport modes into the system and the application of artificial intelligence to save computing time for the control algorithm. These should be taken up in order to further improve the logistical performance of the control system and to be able to apply it terminal-wide to all handling processes. Since the introduction of the solution approaches goes hand in hand with radical changes in the work situation for the employees, a virtual training environment was to be created with the involvement of operational employees to facilitate the transition to the new system for the driving personnel and increase its acceptance among the workforce. **Approach** Discrete-event simulation was used to model the process flows of the automobile terminal and to parameterize the dynamic priorities. A control algorithm was connected to the simulation model, which uses a two-stage optimization procedure to individually assign at first driving jobs to driving and shuttle personnel and then solves a shuttle routing problem via a branch and bound procedure. Artificial intelligence methods were used to increase the performance of the control algorithm. For this purpose, a Graph Convolutional Neural Networks was trained via imitation learning to imitate the approach of the classical optimizer (assignment of driving jobs to drivers and shuttles). In addition, a concept for information exchange in ships was designed and prototypically evaluated for the implementation of the control system. This is based on a router-satellite system and creates a mesh WLAN. **Results** Via discrete-event simulation, the modes of transport and the dynamical priorities could be successfully integrated into the control system, and via simulation studies, it was shown that the processes on the terminal can be reliably controlled in such a way that the time-critical processes of loading and unloading modes of transport can be completed within the specified layover/standby times. A preliminary version of the neural network was created and validated to improve the performance of the control algorithm. For a final application, a higher number of test data from simulations is still pending in order to further improve the solution quality. The concept for data reception on the ship was successfully validated prototypically and showed good response time, transmission speed and range throughout the vessel. Several scenarios (including area and process knowledge) were developed for the virtual training environment. The positive results from the simulation and laboratory studies could be confirmed in several test runs, which took place with the involvement of operational employees and under occupational psychology support on various automobile terminals of BLG.

Contact persons:
M. Hoff-Hoffmeyer-Zlotnik  (Project manager)

A. Ait Alla 
N. Jathe 

Funded by:
BMDV

Duration:
01.07.2020 - 30.06.2023

See project's publications
List all projects