Performance Prediction and Optimization for Industrial Sieves by Simulation: a two-tier Approach Christian Jaeger 1 , Reza Housseini 1 , Manfred Hertwig 1 , Thomas Hofer 2 , Rudolf Marcel Füchslin 1 1 Institute of Applied Mathematics and Physics Zurich University of Applied Sciences, Winterthur, Switzerland 2 Ammann Group, Langenthal, Switzerland jaeg@zhaw.ch We present a numerical study of sieving behavior on industrial sieves, composed of several vibrating screens, a process widely used in diverse industries. The modeling approach is twofold: on the one hand, particle flow is modeled in some detail by means of a discrete element model (DEM). This allows studying the influence of various parameters on the behavior of individual particles, particularly transport velocities and collision rates. Computational complexity however forbids the simulation of an entire sieve as a DEM. Instead, the overall sieving behavior is modeled separately by means of a more phenomenological model, the so called thick layer model (TLM), which is based on mass-balance equations, that translate into an ordinary differential equation. The TLM obtains its most crucial input parameters as results of the DEM. Comparison of simulation results with measurements shows, that this combined approach is capable of accurately describing the sieving process at a reasonable computational cost. 1 Introduction Sieving or screening is the standard operation for the classification of particles according to their size, used in many industries. Experimental investigation of this process dates back to at least the middle of the 20th century, e.g. [8, 10]. The intricate dependencies of the sieving quality on various parameters (e.g. oscil- lation amplitude, frequency and sieve inclination) that were found, incited an interest in model-based studies Existing models can be divided into those that are es- sentially based on the theory of stochastic processes and those relying on discrete element (DEM) simula- tions. Interest in the latter seems to receive new mo- mentum with the increased availability of computing power in recent years, e.g. [4, 11] and others, see be- low. A widely used model is from Standish [13].It is based on the simple assumption that screening is a first-order process. This means that the material on the screen falls through the mesh with a constant rate, giving rise to an exponentially falling function of time, respectively of distance travelled along the mesh. The work of Andrzejczak and Wodzinski [3] is similar in assuming that passage through the mesh happens as a first order process with constant rates to be determined experimentally. In addition, they try to model the effect of a thick bed by assuming that only those particles can fall through the mesh that are in a so-called discharge layer of some given thickness, directly above the screen. The model of Sultanbawa et al. [14] is another macroscopic approach which is based on constant rates and mass balances. Instead of fitting the rates, the focus is on a related parameter q, the ratio of concentrations of undersize particles in the inlet and overtails streams. The theory is applied to a cascade of sieves and a graphical technique is devel- oped to predict the concentration of fines with time. With the rapid progress in computer efficiency, micro- scopic modeling and simulation techniques have be- come popular in the last 10 years. The idea is to model individual particles and their interaction among each other and with the screen and to numerically follow their trajectories. The different sub-classes of such