Granular materials, including iron ore sinter, play a crucial role in steelmaking, where bulk behaviour affects process efficiency. The highly irregular and angular morphology of sinter particles governs their bulk behaviour. In order to understand the bulk behaviour of granular materials, like sinter, simulations are performed. Discrete Element Method is a widely used technique for simulating bulk behaviour. However, in this method the particle shapes are often simplified as spheres to minimize computational cost. This raises the question of the extent to which particle shape detail is needed to maintain realistic bulk behaviour. In this study, the scope of bulk behaviour is limited to flowability. To overcome the computational load of DEM simulations, geometric complexity was transferred from the virtual to the physical domain by fabricating 3D-printed sinter particle replicas at various levels of simplification.
For shape simplification, three common DEM shape modelling approaches, multi-sphere, polyhedral, and super-ellipsoid, were first evaluated, after which polyhedral models were chosen due to their superior ability to capture angular features. Then, the shape fidelity across various simplification levels (400,000-40 faces) was quantified using shape descriptors (sphericity, convexity, and roundness). Based on these quantifications in shape deviations, four shape models were selected: the original shape (400,000 faces, previously 3D-printed by Wouter Schuitemaker), the threshold resolution (400 faces), an intermediate model between the threshold and most simplified resolution (100 faces), and the most simplified model (40 faces). The latter three models were 3D-printed in bulk (1000 particles per model). Then these models were used in flowability experiments, where angle of repose (AoR), coefficient of static friction (μₛ), and Hausner ratio (HR) were measured.
From the results, it follows that geometric simplification increases the sphericity, convexity, and roundness, with a critical threshold near 400 faces, below which these shape descriptor values increase nonlinearly. The AoR, μₛ, and HR exhibit only modest variations across simplification levels, and confidence intervals overlap substantially, showing that stepwise differences are generally not statistically significant. Statistically detectable cumulative reductions occur only at very low face counts (~100 faces or fewer) and are primarily associated with increases in sphericity and convexity; roundness has a weaker influence.
Importantly, the AoR remains within the ‘cohesive’ flowability classification up to 100 faces, indicating negligible practical change in bulk flow behaviour. Only for the most simplified 40-face particles does AoR approach the upper bound of the ‘fair-flowing’ classification. The HR follows a similar trend, remaining within ‘poor flowability’ before reaching the upper bound of the ‘passable’ category for the 40-face particles, implying a marginal increase in flowability.
Overall, flowability appears relatively insensitive to particle shape simplification within the tested range. Statistically detectable changes might occur only at very low resolutions (~100 or fewer faces), with the 40-face particles showing a slight shift in flowability classification. Whether other aspects of bulk behaviour beyond flowability are sensitive to shape simplification remains to be determined.

The Discrete Element Method (DEM) in combination with an One Factor At a Time (OFAT) experimental simulation plan were employed to investigate the effect of a selection of factors on the mixing performance of a double shaft, batch-type paddle mixer. Most influential factors on the mixing performance of the paddle mixer are desired for optimization purposes. Selected factors are the three material characteristics (particle size, particle density, composition), three operational conditions (initial filling pattern, fill level, impeller rotational speed) and three geometric characteristics (paddle size, paddle angle and paddle number).

A 175L paddle mixer is converted into a simulation model and the material model is adopted from literature. The material model comprises of free-flowing, perfectly spherical glass beads. The Hertz-Mindlin contact model is used to simulate the granular material. Furthermore, the simulation strategy consists of the filling process and mixing process. In the former, all input settings are specified. The latter process uses the filling process results to mix the 2-component mixture for a total real-time of 30 seconds. To ensure robust, stable and fit-for-purpose DEM simulations, a 'worst-case scenario' simulation is built for the filling process. By variations in shear modulus and time step, the simulation time is being reduced without compromising the stability or realistic behavior of the material. The output is then used in both processes for all simulations in the experimental simulation plan.

The mixture quality is assessed by the mixing index Relative Standard Deviation (RSD), where both the mixing effectiveness (mixture quality after 30 seconds) and the mixing efficiency (the time it takes to reach a RSD lower than or equal to 0.2) are evaluated. The particle size and particle density have a significant effect on the final mixture quality after 30 seconds of mixing (KPI 1). Moreover, screenshots in x, y and z directions at predefined time steps are generated to serve as visual feedback to understand the mixing mechanisms and flow patterns qualitatively. And, the RSD only stipulates the global mixing performance of the system. To be able to observe local differences in mixture quality, heat maps are generated. Additional local grid systems derived from the global 14x14x9 are designed such that the mass concentration of one of the components can be visualized in x, y or z direction.

The particle size and particle density have a significant effect on the final mixture quality after 30 seconds of mixing (KPI 1). Moreover, the composition, initial filling pattern, fill level, impeller rotational speed and paddle size have a significant effect on how fast a steady-state mixture quality is reached (KPI 2). Finally, the paddle angle and paddle number have a significant effect on both KPIs and therefore holds most potential with respect to optimization purposes.