Including stochastics in metamodel-based DEM model calibration

In calibration of model parameters for discrete element method (DEM) based models the focus lies on matching mean key performance indicator (KPI) values from laboratory experiments to those simulation results. However, due stochastic nature granular processes experimental results can show large variances. To include behaviour, interpolation-based and regression-based metamodels are trained with data. These used in standard approach newly introduced mean-variance predict KPIs variance. addition, effect enriching data is investigated up 50 repetitions simulations. Based a hopper case study, use KPI repeated at least 20 times recommended. While differences between mean-variance-based were minor considered metamodeling clearly showed improved accuracy stability over metamodels.