Parametrization of a Simulation Model for Flexible Small Consignments Using Parallel Bayesian Optimization

Abstract

The increasing volume of shipments in the courier, express, and parcel (CEP) sector and the transition from traditional rectangular cardboards to small consignments with flexible packaging pose challenges for manufacturers and operators of sorting technology. Existing sorting systems often fail to meet the requirements of these changes. While physics simulations are commonly used to develop new sorting systems, they are inadequate for modeling flexible consignments. A novel simulation approach for small consignments with flexible packaging was developed recently. Our study builds on that approach by parametrizing and validating the simulation. Initially, we analyzed the movement and transit times of four different types of small consignments on three test rigs with various conveying technologies, such as belt and roller conveyors, operated at different speeds. We then created a simulation model for one of the test rigs and incorporated existing models for two consignment types. We identified the most influential parameters on the simulation results through exhaustive parameter screening. Using Bayesian optimization with batch evaluations, we determined the optimal parameter values to minimize discrepancies between the simulated and real-world movement trajectories. Finally, we validated the simulation models against additional independent real-world experiments, finding that the simulation results fell well within the range of variation observed in these experiments.