The linearity of the mass flow gives an indication of whether or not the sampling was performed at steady state conditions。 After 40 s the sample container is removed and stored for later analysis。  The speed is then lowered to the new set point and the power draw is allowed to stabilize once again。 At this point a new sampling procedure may be initiated and the process is repeated until all eccentric speed set points have been completed。 The size distribution of

the product samples has been analysed using sieve analysis following the EN933-1 standard (Standardization, 1998)。

2。2。 DEM model configuration

 The DEM model has been developed and simulated in EDEM 2。7provided by DEM Solution Ltd。 The rock material is based on twodifferent 3D scanned rock shapes in the size range 6–8 mm。 Thescanned particles shapes are used as moulds for a packed cluster of fraction particles that are bonded with beams according to the BPM modelling approach (Potyondy and Cundall, 2004; Potyondy  et al。, 1996)。 The full procedure and methodology of the specific  approach of using meta-particles is described by Quist (Quist and  Evertsson, 2016)。 According to the framework by Potyondy and  Cundall, particles are allowed to overlap given that the overlap is small compared with the particle fraction size。 Bonds of finite stiffness are created at contact points between two particles and these bonds carry load and break when the calculated stress on the bond exceeds a strength criteria。 The bonds fail under tensile or shear loads but not due to compression。 The micro-properties of the meta-particles are presented in Table 1。 When fraction particles  are liberated and are no longer part of a cluster, the interaction  with surrounding particles and geometrical elements is controlled  by the Hertz-Mindlin no slip contact model。

 The main limitation and challenge of performing DEM simulations of comminution equipment is to balance; the number of meta-particles included in the simulation, the size of the machine section modelled, and operational time needed to be able to draw any useful conclusions。 In this work a 40 degree section of the crusher is modelled in order to increase the amount of material in one feeding location with the intention

of achieving a choke feeding condition。 The section limitation is realised by the boundary wall where the friction parameters are set to low values。

 The nutational motion of the mantle is created by two sinusoidal rotations defined from a pivot point where one of the motions is phase shifted p/2 rad。 In addition, a third counter rotational motion around the vertical axis is added in order to simulate the mantle’s rolling motion on the concave。 In the real crusher the mantle is allowed to freely rotate around the main shaft axis and the interaction between the mantle and concave can be compared with a planetary gear where the rocks act as gear teeth。 If this counter rotation is not included the simulated particles will experience a false horizontal force causing particles to report to the boundary wall。 During the initial iterations of these simulations it was found that the number of meta-particles was not enough to achieve a satisfactory choke fed condition。 One reason is due to the significantly higher eccentric speed of the mantle that excites the feed particles

with a force upwards causing an unstable bed。 In order to mitigate this issue a new strategy was tested where a set of large spherical particles were created on top of the meta-particles to apply a choke feeding pressure。 The success of this strategy needs to be evaluated further as the interaction between the additional choke feed parti cles and the meta-particles creates a bias in the simulation results。

2。3。 DEM post processing

 The particle size distribution of the surviving clusters of meta  particles has been estimated using the methodology described in Quist and Evertsson (2016)。 Images of the surviving bonded clusters are recorded, see Fig。 3, and the MATLAB image analysis tool box is utilized to first apply a Gaussian filter and secondly calculate the major and minor axis lengths of all identified clusters。 The par ticles on image boundaries are removed。 From the two dimen sional size of each cluster an ellipsoid volume is calculated in order to calculate an estimate of the particle mass of each cluster particle。 The particle size distribution is further on calculated by sorting each cluster based on the minor length size and the estimated mass to each corresponding size class。