Case Study // Process Engineering

Fig. 2: Overview of the parameter and their corresponding ranges

CASE STUDY // PROCESS ENGINEERING
as the cylinder is completely filled with particles, it is moved terwards a calibration between simulation and experiment
upwards creating the typical form of solid bulk. Here, both can be achieved with these important parameters by means
CALIBRATION OF ANGLE-OF-REPOSE AND DRAW- the angle-of-repose and the draw-down angle can be deter- of an optimization algorithm.
mined as a mean value along the bulk material heap.
DOWN ANGLE USING A ROCKY-DEM SIMULATION The angle-of-repose represents the outer angle of the
bulk material and can be determined directly after such an
Rocky Discrete Element Method (DEM) and optiSLang were used to calibrate a model for the simulation of particle experiment. The draw-down angle describes the inner angle
behavior in order to optimize the geometry of a bulk material funnel. after the bulk heap has dispersed from the middle. The mean
value along the bulk material was then calculated for both
angles and the results were used in the subsequent analyses.

Motivation Procedure
The angle-of-repose and the draw-down angle are critical conditions require a variant study with an efficient inves- The process of solving this task is divided into two steps.
parameters for assessing the particle behavior, e.g. regard- tigation of the parameter space. In the procedure, meta- First, the conduction of a sensitivity analysis and, second, the
ing bulk material. Prior to a particle simulation, the un- model algorithms are used to not simply “fit” the data, but use of an optimizer to minimize the deviations between the
known properties and parameters of the numerical simula- also take into account the influence of solver noise and, simulation model and the experiment.
Fig. 3: General workflow of a calibration with, first, the identification and,
tion should be determined by reproducing an experiment thus, illustrate global trends. These requirements can be second, the optimization of these found parameters to fit the simulation
within a particle simulation. fulfilled by using the MOP approach. with the reference data
Calibration represents the starting point of a param- The following example applies the fixed-funnel method Calibration
eter study, especially if uncertainties regarding the de- to identify the parameters for describing the particles. The The aim of a calibration is to match the results of a simula-
termination of material constants or the arrangement of simulation process of this method is shown in Fig. 1. As soon tion to the measurements. This can only be obtained after Workflow
constraints have to be considered in numerical simulation. an identification of the important parameters. Once they Based on the obtained parameters, 100 simulations were
Dealing with particle simulations, this is a constant chal- are detected, a variation can be conducted to minimize the generated arranged in a Latin Hypercube as part of a Design-
lenge, because the individual properties depend not only deviation. In this study, the physical properties were used as of-Experiment. The parameters were transferred to Rocky
on the material of the particle, but also on its shape and the parameters and varied in the ranges shown in Fig. 2. It should DEM using the custom integration that is pre-installed in
environmental conditions. This results in a large amount of always be noted here that some parameters cannot be con- optiSLang. Then the model was updated and numerically
parameters, which make the identification of interactions sidered as independent. In this case, the static and dynamic solved. Fig. 4 (see next page) illustrates the workflow.
significantly more difficult. A sensitivity study was first per- friction is multiplied by a weighting factor that describes the After completion of the simulation, the results were
formed to generate the Metamodel of Optimal Prognosis ratio between them and is always less than one. This value transferred back to optiSLang and used for the evaluation.
(MOP) and to identify the important parameters. Such an also represents a physical constraint in the parameter study. Using these 100 simulations from the sensitivity study, the
approach is particularly important and challenging for the In addition, the particle density, the particle diameter and MOP could be created for the two result values draw-down
Discrete Element Method (DEM), because of the consider- the mass flow while entering the test tube were examined. angle and angle-of-repose. The MOP was then used for an
ation of many parameters and the processing of quantities Fig. 1: Illustration of the transient simulation to determine the angle-of- The general procedure for a calibration is shown in Fig. optimization to minimize the deviations between simulation
due to the noise of the numerical solution. These boundary repose and the draw-down angle of a bulk material 3. First, the important parameters should be identified. Af- and experiment. The aim of the optimization was defined as

8 RDO-JOURNAL // ISSUE 1/2018 9

 Case Study // Process Engineering

Fig. 4: optiSLang workflow of an automized sensitivity study for a Rocky DEM simulation

the sum of the squared differences between simulation and For the data set examined here, the rolling friction is of im-
experiment. Here, a Non-Linear Programming by Quadratic portance, because it only describes the two result variables.
Lagrangian (NLPQL) proved to be a suitable method. This pro- With the help of the subsequent optimization based on the
cedure yielded a sufficient calibration without using another MOP, the rolling friction could be identified. Here, the devia-
numerical simulation run. tion between simulation and experiment was minimal. Thus,
both values were accurately determined, the angle-of-repose
differs 0.21° from the experiment and the draw-down angle
Results with 0.1°. This small remaining deviation can be seen exem-
Based on the results of the sensitivity study, first, the range plarily for the noise of the Rocky simulation in this example.
of values of the result variables could be compared with
the experimental data. The results and the associated ex-
perimental data are comparatively shown in Fig. 5. The ex- Conclusion
perimental values are positioned in the middle of the result Using the created workflow, Rocky DEM simulations can
space surrounded by the simulation results. This indicates be coupled with other platforms via optiSLang and optimi-
that the parameter space is sufficiently chosen and a calibra- zations can also be carried out. The procedure is fully au-
tion is possible. At this point, however, the optimum in the tomated and, thus, less prone to errors. The workflow was
parameter space cannot yet be located. For this purpose, a used to determine the numerical input variables based on
reduction of the parameters and a subsequent optimization experimental data for the simulation of the angle-of-repose
is required. and the draw-down angle. The identification of the rolling
friction as an important parameter as well as a sufficient
plausibility check including a physical test would not have
been possible without an automated parameter study using
the MOP. Those parameters identified as not important can-
not be calibrated by the experiment. Such values have to be
considered in other experiments.
By means of optimization, the suitable rolling friction
could be determined. For this purpose, not only the values
of the MOP but also the values of the numerical model
were calculated. The deviation between them is just a few
percent, thus, this value of rolling friction could be used for
further simulations.

Author//
Bernd Büttner (Dynardo GmbH)
Fig. 5: Illustration of the experimental values (red), which are the calibration
target and the results of the performed simualtions The case study was originally presented at the CASCON 2017
under the title “Automated optimization workflow with op-
The reduction of the important parameters could be reached tiSLang for Rocky DEM simulations”.
by using the MOP. Beside the identification of important pa-
rameters, the MOP also enables the elimination of the strong
solver noise as a typical phenomenon in a DEM simulation.

10