betaFeedback

WorksPhysics and Astronomy

Granular mixing and flow dynamics in horizontal stirred bed reactors

Abstract

Horizontal stirred bed reactors (HSBRs) are used in gas–phase polyolefin production, where efficient solids mixing and controlled residence time distributions are essential for product quality and stability. Despite their industrial relevance, the influence of operating conditions on granular flow and mixing in HSBRs is not well understood. Discrete Element Method (DEM) simulations are used to study the effects of rotation speed and fill level on particle motion, mixing, and axial transport in a lab–scale HSBR. An industrial–grade polypropylene powder is modelled using calibrated contact parameters. Mixing is quantified using the Lacey index in axial (z) and cross–sectional (xy) directions. Particle circulation is characterised via cycle–time analysis and a coarse–grained angular velocity field. Axial dispersion coefficients are obtained from particle trajectories using both Einstein–type and cycle–based approaches, and validated with a diffusion model predicting the axial Lacey index. Results show that axial mixing depends strongly on rotation speed and fill level: higher rotation speeds accelerate homogenization, while higher fill levels slow mixing. Cross–sectional mixing is mainly sensitive to rotation speed, with fill–level effects diminishing at higher speeds. Cycle time decreases with increasing rotation speed and fill level, indicating enhanced circulation. Axial dispersion increases with rotation speed but decreases with fill level, with consistent results across methods. These findings reveal trade–offs between axial mixing, circulation, and dispersion, highlighting the need to balance operating conditions and demonstrating the capability of DEM to support HSBR optimisation.

Introduction

Horizontal Stirred Bed Reactors (HSBRs) are widely used in industrial gas-phase polyolefin production, particularly for polypropylene, one of the two major commercial polyolefins alongside polyethylene, where efficient solids mixing and controllable residence-time distributions are essential for product quality and stable operation [1, 2, 3, 4]. HSBRs originated from the early Amoco horizontal reactor concept [1] and have evolved into advanced industrial systems (e.g., Innovene and Horizon) employing segmented agitation to sustain continuous production at large scale [2, 3, 5]. A key characteristic of HSBRs is that they combine plug-flow-like axial transport with strong internal mixing, leading to relatively narrow residence time distributions (RTDs) and improved control of polymer property distributions [6, 7, 8].

From an operational perspective, HSBR performance is governed by the coupled dynamics of (i) mixing within cross–sections, (ii) solids circulation induced by the agitator, and (iii) axial transport and back–mixing along the reactor length. These mechanisms directly affect RTD shape and, consequently, polymer property uniformity and process stability [6, 7, 8]. Among the controllable operating parameters, impeller rotation speed and fill level are particularly important because they determine the balance between shear–driven agitation, bulk circulation, and axial migration of particles [5, 7, 9].

Recent experimental studies using single–photon emission radioactive particle tracking (RPT) and X–ray imaging by van der Sande et al. [9, 10] have provided direct evidence that particle motion and solids distribution in HSBRs are highly sensitive to operating conditions, including rotation speed and fill level. These measurements highlight how changes in operating conditions can alter circulation patterns and the degree of solids redistribution, potentially leading to non–uniform flow structures or stagnant regions when circulation is insufficient. Such observations motivate quantitative, systematic analysis of mixing and axial transport as functions of operating parameters.

Modelling efforts—including RTD analyses and reactor–scale frameworks for gas–phase polymerization—have further demonstrated that axial back–mixing, circulation, and local granular flow behaviour influence reactor performance and product distributions [2, 5, 6, 7, 8]. However, establishing clear, parameter–resolved relationships between rotation speed and fill level on the one hand, and mixing efficiency, circulation, and axial transport on the other, remains challenging, partly because obtaining detailed internal flow information experimentally is difficult [9, 10].

Despite these advances, a detailed and systematic understanding of how rotation speed and fill level jointly influence mixing efficiency, particle circulation, and axial transport in HSBRs is still limited. This work addresses this gap by developing a Discrete Element Method (DEM) model that reproduces and explains the experimentally observed trends of van der Sande et al. [9]. Specifically, we investigate how rotation speed and fill level affect mixing efficiency, particle circulation, and axial transport in a lab–scale HSBR. By combining DEM simulations with multiple independent mixing and transport metrics, and by validating these against experimental observations, this work provides mechanistic insight into the coupled dynamics governing HSBR performance.

The paper is structured as follows. § 2 describes the methodologies used, including the DEM model, simulation procedure, and mixing and flowability analysis tools. Afterward, § 3 presents the obtained results, measuring the Lacey index, cycle time, and axial and radial dispersion coefficients. Furthermore, we study the effect of operating parameters on mixing quality and particle motion. Finally, § 4 summarizes the main points of this work.

Methods

Particle characterization

To model industrial conditions, we selected an industrial–grade polypropylene reactor powder as the granular material. The particle size distribution and aerated bulk density were characterized experimentally, as described in [11]. The particle size distribution (PSD), measured using laser diffraction, had a median diameter of approximately 904.6 μm904.6\ \mu\mathrm{m}, with the full distribution presented in Table 1. The aerated bulk density was found to be 368 kg/m3368\ \mathrm{kg}/\mathrm{m}^{3}, while the elastic modulus, shear modulus, and coefficient of restitution used in simulations were based on material databases, shown in Table 2. Detailed procedures and characterization data are available in ref [11].

Particle diameter (μm\mu\mathrm{m})Cumulative volume (%)
506.010
712.725
904.650
112375
134590

Table 1. Particle size distribution.

ParameterValue
Elastic modulus EE1.325×109 Pa1.325 \times10^{9}\ \mathrm{Pa}
Shear modulus GG4×108 Pa4 \times10^{8}\ \mathrm{Pa}
Coefficient of restitution ee0.5
Aerated (loose) bulk density ρ\rho368 kg/m3368\ \mathrm{kg}/\mathrm{m}^{3}

Table 2. Material properties of the polypropylene particles.

Simulation

The Discrete Element Method (DEM) was used to simulate powder mixing in the lab–scale horizontal stirred bed reactor (HSBR). DEM allows for the accurate modeling of granular material behavior by representing particles as discrete entities and simulating their interactions [12]. The simulations were performed using the MercuryDPM software package [13, 14].

A combined Hertz–Mindlin model [15] and rolling friction model [16] were utilized for the DEM simulations of the HSBR, as previously described in [11]. To maintain consistent units, the rolling stiffness was defined as equal to the tangential stiffness used in the Mindlin model [16].

To reduce computational cost while preserving the overall granular dynamics, the particle size in the HSBR simulations was scaled by a factor of 3, compared to a scaling factor of 5 used in [11]. This scaling approach accelerates simulations by reducing the total number of particles while maintaining similar bulk flow behaviour. As scaling influences contact interactions, the sliding (μs\mu_s) and rolling (μr\mu_r) friction coefficients were recalibrated under the present scaling conditions.

The calibration procedure described in [11] was followed, in which the friction coefficients are adjusted to ensure that the simulated angle of repose (AoR) matches experimental measurements. Based on this procedure, the values μs=0.8\mu_s = 0.8 and μr=0.4\mu_r = 0.4 were adopted for the simulations in this study.

Simulation procedure

The DEM simulations were conducted to evaluate the impact of fill level and rotation speed on particle dynamics and mixing efficiency in a lab–scale horizontal stirred bed reactor (HSBR). The simulated geometry and operating conditions were chosen to match the experimental setup reported by van der Sande et al. [9], enabling direct comparison between simulation results and experimental measurements. The cylindrical reactor had an inner diameter of 134 mm and a length of 150 mm, with an agitator shaft featuring seven blade positions, each fitted with two blades positioned 90° apart. The inner blades were 20 mm wide, while the end blades were 15 mm wide. A schematic of the simulated HSBR is presented in Figure 1.

Lab-scale HSBR geometry

Figure 1. Lab-scale HSBR geometry. The column is fixed, and the agitator rotates.

The interaction between particles and the reactor walls was modeled using the same contact properties as those applied for particle–particle interactions.

To investigate the influence of reactor parameters, simulations were conducted for combinations of four fill levels and six rotation speeds, as shown in Table 3. The density in the simulations was set to ensure that the simulated bulk density matched the experimentally determined bulk density in [11].

Fill Level (%)Rotation Speed (rpm)
4010
5020
6030
7040
50
60

Table 3. Fill levels and rotation speeds used in simulations.

Spherical particles were used in the DEM simulations, consistent with previous work [11]. This simplification was used to reduce computational complexity while preserving the essential particle dynamics. The shape effects of non–spherical particles were accounted for by adjusting the rolling friction coefficient (μr\mu_r) to higher values based on the relationship between particle shape and rolling resistance [17, 18].

The simulation time step (Δt\Delta t) was set to 20% of the Rayleigh time step (ΔtRayleigh\Delta t_{\mathrm{Rayleigh}}) to ensure numerical stability [19, 20]. The definition and calculation procedure of ΔtRayleigh\Delta t_{\mathrm{Rayleigh}} are provided in Reference [11].

To further reduce computational time, the particle stiffness was decreased by a factor of 10310^3, following the guidelines of [21] and [22], without any effect on numerical stability or particle flow.

Simulations were conducted for all combinations of fill levels and rotation speeds, yielding a detailed dataset for analyzing the effects of these parameters on HSBR performance.

Mixing and flowability analysis tools

Lacey index

The Lacey index, first introduced by Lacey in 1954 [23] and later refined in 1997 [24], is one of the most widely used statistical indicators for quantifying the degree of homogeneity in particulate and powder mixing processes [25]. In Lacey's formulation, the degree of mixing is defined as:

M=σ2−σ02σr2−σ02(1)M = \frac{\sigma^2 - \sigma_0^2}{\sigma_r^2 - \sigma_0^2} \tag*{(1)}

where σ2\sigma^2, σ02\sigma_0^2, and σr2\sigma_r^2 are the sample variances of the actual, fully segregated, and perfectly random states, respectively [25, 26]. This normalization ensures that MM is dimensionless and independent of sample size, enabling consistent comparisons across different systems and operating conditions.

The Lacey index provides a normalized measure of how closely a system approaches a random distribution relative to complete segregation. A value of M=0M = 0 represents a fully segregated state, while M=1M = 1 corresponds to a perfectly random (ideally mixed) system.

The variance of the actual mixture, σ2\sigma^2, is calculated from the composition of multiple spatial samples taken from the system:

σ2=1S−1∑i=1S(fa,i−fa)2(2)\sigma^2 = \frac{1}{S-1} \sum_{i=1}^{S}(f_{a,i} - f_a)^2 \tag*{(2)}

where SS is the total number of samples, fa,if_{a,i} is the fraction of particles of type aa in the ithi^{\mathrm{th}} sample, and faf_a is the global fraction of type-aa particles in the system.

Following Lacey [24], the variance of a completely segregated state is defined as:

σ02=fa(1−fa)(3)\sigma_0^2 = f_a(1-f_a) \tag*{(3)}

since in the initial state, each sample contains only one type of particle. The variance for a perfectly random mixture—i.e., the statistical variance due to random fluctuations—is:

σr2=fa(1−fa)n(4)\sigma_r^2 = \frac{f_a(1-f_a)}{n} \tag*{(4)}

where nn is the average number of particles per sample.

The Lacey index has become a standard metric in industrial mixing research due to its strong theoretical foundation and practical interpretability, and has been employed across a wide range of mixing systems in both batch and continuous processes. It has been applied in studies of rotating drums [26, 27, 28, 29, 30, 31, 32, 33, 34, 35], ribbon mixers [36, 37], blade mixers [38, 39, 40, 41, 42, 43], as well as fluidized beds [44, 45] and tote or continuous blenders [46].

In this study, the Lacey index was employed to quantify both axial and cross–sectional mixing behavior in the horizontal stirred bed reactor (HSBR).

A fully segregated initial condition was prepared by labelling particles according to their spatial position after the insertion phase, once all particles had been introduced into the reactor. To measure axial mixing, the reactor was divided into two equal halves: particles in the left half were labeled as red, and those in the right half as blue, as shown in Figure 2(a). The domain was then segmented into ten axial bins using a mass–based sampling strategy such that each bin contains approximately equal total particle mass. The local mass fraction of red particles, fa,if_{a,i}, was calculated in each bin over time and used in Equation (1) to compute the Lacey index M(t)M(t). In this configuration, type–a particles are defined as the red fraction, and since exactly half of the particles were initially assigned as red, the global composition was fixed at fa=0.5f_a = 0.5.

Initial particle labeling used for mixing analysis

Figure 2. Initial particle labeling used for mixing analysis: (a) axial (zz–direction) labeling; (b) cross–sectional (xyxy–plane) labeling for the HSBR at 40% fill level and 40 rpm rotation speed.

It is important to note that the labels red and blue are purely virtual identifiers based on initial particle position; both particle fractions have identical material properties. Therefore, given sufficient mixing time, these two labeled populations are expected to fully homogenize, providing a reliable basis for assessing mixing performance.

To measure cross-sectional mixing, particle labeling was defined radially, as illustrated in Figure 2(b). Particles located closer to the reactor center were labeled as blue, while those in the outer radial region, near the vessel wall, were labeled as red. The cross-section was partitioned into 10×1010 \times10 bins using a mass-based sampling strategy, and the red particle mass fraction was computed in each bin throughout the simulation to evaluate the Lacey index over time.

This sampling strategy partitions the domain into bins with approximately equal total particle mass, aiming to keep particle counts per bin reasonably balanced and to reduce statistical noise in variance estimation. The time evolution of M(t)M(t) effectively captures the transition from a fully segregated state (M=0M = 0) to an ideally mixed state (M≈1M \approx1). By analyzing M(t)M(t) under varying operational parameters such as impeller rotation speed and fill level, the efficiency of granular mixing can be quantitatively characterized. Owing to its normalization and theoretical rigor, the Lacey index continues to serve as a reliable and widely accepted measure of mixing quality in both experimental studies and DEM simulations of mechanically agitated systems, such as the HSBR examined here.

Cycle time

Cycle–time analysis has been employed as a practical tool to investigate particle circulation dynamics in mechanically agitated granular systems, particularly horizontal stirred bed reactors (HSBRs). Van der Sande et al. [9, 10] introduced this concept experimentally to characterise solids circulation in HSBRs using both radioactive particle tracking and X–ray imaging. Their studies established cycle time as a useful indicator of solids circulation and flowability within the reactor. Similar circulation concepts have been discussed in earlier granular mixing research, where the cycle time represents the period required for a particle to complete one full revolution around the agitator or vessel circumference [47]. A shorter cycle time means the solids recirculate more quickly, enabling more frequent particle exchange and typically enhancing radial and circumferential mixing. Conversely, longer cycle times are associated with slower circulation and the persistence of localized circulation loops or stagnant zones [9, 10, 47].

In continuous HSBRs, cycle time is also linked to macroscopic transport behaviour: rapid internal cycling promotes solids back–mixing and broadens the residence time distribution, whereas slower cycling tends to yield more plug–flow–like operation [6]. Experimental measurements using radioactive particle tracking technique [9] and numerical studies employing the Discrete Element Method (DEM) [47] have demonstrated that operating conditions such as rotational speed and fill level strongly influence circulation time and, consequently, flowability.

In this study, we extend this concept in silico by implementing a DEM-based definition of cycle time to quantify angular transport across different HSBR operating scenarios. The cycle time Δti,kc\Delta t^{c}_{i,k} is defined as the time required for particle ii to complete its kthk^{\mathrm{th}} revolution around the central shaft, i.e.,

Δti,kc=ti,k+1c−ti,kc(5)\Delta t^{c}_{i,k}=t^{c}_{i,k+1}-t^{c}_{i,k} \tag*{(5)}

where ti,kct^{c}_{i,k} is the time at which particle ii completes its kthk^{\mathrm{th}} cycle. In this study, a full revolution is identified when the particle has sequentially traversed all four quadrants in the xy–plane. This criterion is based on angular displacement rather than returning to the exact same geometric location or angle. As a result, the angular positions of the particle at successive ti,kct^{c}_{i,k} may differ, as shown in Figure 3.

Sample particle trajectory in the xy–plane

Figure 3. Sample particle trajectory in the xy–plane (blue), particle positions at ti,kct_{i,k}^{c} (red), and shaft center (yellow). Trajectories are linearly interpolated between simulation time steps.

To avoid counting the transient period during which particles are still being inserted into the mixer, only revolutions after the insertion phase are considered in the analysis.

The mean cycle time, averaged over all particles and their valid revolutions, is computed as:

Δtc‾=∑i=1Np∑k=1NicΔti,kc∑i=1NpNic(6)\overline{\Delta t^{c}} = \frac{\sum_{i=1}^{N_p}\sum_{k=1}^{N_i^c}\Delta t_{i,k}^{c}}{\sum_{i=1}^{N_p}N_i^c} \tag*{(6)}

Here, NpN_p is the total number of particles in the system, and NicN_i^c is the number of completed revolutions detected for particle ii.

Figure 3 illustrates the xy trajectory of a sample particle (blue line), with red circles marking the positions at each ti,kct_{i,k}^{c}. The yellow circle marks the center of the shaft. Note that the particle positions at ti,kct_{i,k}^{c} are not located at the same angle or distance from the center. This reflects the fact that the revolution condition is based on passing through all four angular quarters, not exact spatial repetition. Additionally, the blue trajectory segments are formed via linear interpolation between simulation output time steps.

Cycle Time Verification via Coarse–Graining Analysis. To verify the trajectory-based measurement of the mean cycle time, a continuum coarse–graining (CG) analysis was also implemented using the fstatistics tool of the MercuryCG framework [48]. The DEM particle data were converted into continuum fields by applying a Gaussian smoothing kernel with a bandwidth equal to one mean particle radius, yielding spatially coarse–grained fields for the solid volume fraction ϕ(x,y)\phi(x,y) and the velocity components vx(x,y)v_x(x,y) and vy(x,y)v_y(x,y) in the xy–plane.

The tangential velocity in the angular direction was computed from the coarse–grained fields as

vθ(x,y)=xvy−yvxx2+y2,r=x2+y2,(7)v_{\theta}(x,y) = \frac{xv_y-yv_x}{\sqrt{x^2+y^2}}, \qquad r = \sqrt{x^2+y^2}, \tag*{(7)}

and the corresponding angular velocity field followed directly as

ω(x,y)=vθ(x,y)r.(8)\omega(x,y) = \frac{v_{\theta}(x,y)}{r}. \tag*{(8)}

The connection between this coarse-grained angular velocity field and the mean cycle time can be derived from the angular momentum balance. The angular mass flux through the cross-section R2R^2 is

m˙=12π∫R2vθrρ dA,(9)\dot{m} = \frac{1}{2\pi} \int_{R^2} \frac{v_{\theta}}{r}\rho\,\mathrm{d}A, \tag*{(9)}

where ρ\rho denotes the coarse-grained mass density. The total mass in the same cross-section is

m=∫R2ρ dA.(10)m = \int_{R^2} \rho\,\mathrm{d}A. \tag*{(10)}

These two quantities define the mass-weighted mean angular velocity,

ωˉ=∫R2ωρ dA∫R2ρ dA.(11)\bar{\omega} = \frac{\int_{R^2} \omega\rho\,\mathrm{d}A}{\int_{R^2} \rho\,\mathrm{d}A}. \tag*{(11)}

Since the particle material density ρp\rho_p is constant, the coarse-grained mass density satisfies ρ=ρpϕ\rho= \rho_p\phi, and the mass-weighted average becomes equivalent to a volume-fraction-weighted average:

ωˉ=∫R2ω(x,y) ϕ(x,y) dA∫R2ϕ(x,y) dA≈⟨ϕω⟩⟨ϕ⟩.(12)\bar{\omega} = \frac{\int_{R^2} \omega(x,y)\,\phi(x,y)\,\mathrm{d}A}{\int_{R^2} \phi(x,y)\,\mathrm{d}A} \approx\frac{\langle\phi\omega\rangle}{\langle\phi\rangle}. \tag*{(12)}

Finally, the mean cycle time associated with the coarse-grained angular velocity field follows from the ratio of the total mass to the angular mass flux:

Δt‾CGc=mm˙=2π∫ρ dA∫ωρ dA=2πω‾,(13)\overline{\Delta t}^{c}_{\mathrm{CG}}=\frac{m}{\dot{m}}=\frac{2\pi\int\rho\,\mathrm{d}A}{\int\omega\rho\,\mathrm{d}A}=\frac{2\pi}{\overline{\omega}}, \tag*{(13)}

which represents the time required for a mass–weighted average particle to complete one full revolution around the reactor shaft.

The coarse–grained estimate Δt‾CGc\overline{\Delta t}^{c}_{\mathrm{CG}} provides an independent prediction of the mean cycle time that can be directly compared with the trajectory–based value obtained from particle revolutions. In the Results section, the agreement between these two measurements is used to confirm that the DEM-based definition of cycle time reliably captures the angular circulation dynamics of the HSBR.

Axial dispersion coefficient

In addition to the Lacey index and cycle–time analysis, the axial dispersion coefficient provides a complementary measure of how efficiently particles are transported along the reactor length. Rather than describing the overall degree of mixing, axial dispersion quantifies the stochastic spreading of particle trajectories around the mean axial drift. This behaviour is directly linked to the solids residence time distribution and the extent of axial back-mixing in HSBRs, both of which influence process stability and product uniformity.

Previous modelling and experimental studies have shown that axial transport strongly affects polymer properties, catalyst distribution, and thermal behaviour in gas–phase polyolefin reactors [2, 3, 6, 7]. In practice, an appropriate level of axial dispersion helps maintain smooth solids flow along the reactor, prevents local accumulation or maldistribution of catalyst, and supports the desired balance between plug-flow-like behaviour and controlled back-mixing [5, 8].

In the literature, two main approaches are typically distinguished for evaluating dispersion coefficients in particulate systems: a macro–scale approach and a micro–scale approach [49, 50]. In the macro approach, the dispersion coefficient is obtained by fitting the evolution of concentration profiles to a Fickian advection–diffusion model at the continuum scale. In the micro approach, the dispersion is computed directly from particle displacements, following Einsteins original treatment of Brownian motion [51, 52] and its subsequent extensions to granular flows [49, 50, 53]. Since DEM provides full particle trajectories, the micro (Einstein–type) formulation is adopted here.

Time–Based (Einstein) Formulation. Following this micro–scale approach, the instantaneous axial dispersion of particle ii over a sampling interval [tj,tj+Δt][t_j, t_j+\Delta t] is defined as

Di,jz(Δt)=(Δzi,j−Δz‾)22Δt,(14)D^z_{i,j}(\Delta t) = \frac{(\Delta z_{i,j}-\overline{\Delta z})^2}{2\Delta t}, \tag*{(14)}

where the axial displacement over one sampling interval is

Δzi,j=zi(tj+Δt)−zi(tj),\Delta z_{i,j}=z_i(t_j+\Delta t)-z_i(t_j),

where, assuming a statistically steady regime, Δz‾\overline{\Delta z} is the mean axial displacement, averaged over all particles and all sampling intervals,

Δz‾=1NtNp∑j=1Nt∑i=1NpΔzi,j.(15)\overline{\Delta z}=\frac{1}{N_tN_p}\sum_{j=1}^{N_t}\sum_{i=1}^{N_p}\Delta z_{i,j}. \tag*{(15)}

Here, NpN_p is the number of particles and NtN_t is the number of sampled time steps.

The axial dispersion coefficient of the system at time tjt_j is then obtained by averaging over all particles,

Djz=1Np∑i=1NpDi,jz,(16)D_j^z = \frac{1}{N_p} \sum_{i=1}^{N_p} D_{i,j}^z, \tag*{(16)}

and the time-averaged steady-state axial dispersion coefficient DzD^z of the whole system, as in Yang et al. [49], is obtained by averaging DjzD_j^z over all sampling intervals,

Dz=1Nt∑j=1NtDjz.(17)D^z = \frac{1}{N_t} \sum_{j=1}^{N_t} D_j^z. \tag*{(17)}

This time-based (Einstein) formulation is used as the primary definition of axial dispersion throughout this thesis.

Cycle-Based Formulation. An alternative dispersion formulation can be derived using particle displacements evaluated cycle-by-cycle, following the methodology of Ingram et al. [54]. For particle ii, the axial displacement during its kk-th cycle is

Δzi,k=zi(ti,k+1c)−zi(ti,kc),(18)\Delta z_{i,k} = z_i\left(t^c_{i,k+1}\right) - z_i\left(t^c_{i,k}\right), \tag*{(18)}

where ti,kct^c_{i,k} is the time at which the particle completes its kkth revolution around the mixer shaft.

The mean cycle-wise displacement is

Δz‾=∑i=1Np∑k=1NicΔzi,k∑i=1NpNic,(19)\overline{\Delta z} = \frac{\sum_{i=1}^{N_p}\sum_{k=1}^{N_i^c}\Delta z_{i,k}}{\sum_{i=1}^{N_p}N_i^c}, \tag*{(19)}

which approaches zero due to the geometric symmetry of the HSBR.

The cycle–based axial dispersion coefficient is then

Dcz=∑i=1Np∑k=1Nic(Δzi,k−Δz‾)22∑i=1NpNic.(20)D_c^z = \frac{\sum_{i=1}^{N_p}\sum_{k=1}^{N_i^c}\left(\Delta z_{i,k}-\overline{\Delta z}\right)^2}{ 2\sum_{i=1}^{N_p}N_i^c}. \tag*{(20)}

This definition treats each particle cycle as an independent sampling interval. Although both formulations are physically consistent, the time–based definition is used as the primary dispersion measure in the Results, while the cycle–based formulation serves as an independent verification.

Choice of Sampling Interval Δt\mathbf{\Delta t}. To determine an appropriate sampling interval, the dependence of the time–based dispersion coefficient DzD^z on Δt\Delta t was evaluated over a wide range of Δt\Delta t values. The resulting curves are presented in Appendix § 6, Figure A.15. For very small Δt\Delta t, the calculated dispersion coefficient decreases rapidly because particle motion remains strongly correlated, in agreement with the observations of Pallarès and Johnsson [55]. For intermediate Δt\Delta t, a quasi–plateau region is observed in which the dispersion coefficient becomes only weakly dependent on Δt\Delta t. For very large Δt\Delta t, the curves corresponding to different operating conditions gradually approach each other, indicating a loss of resolution in the dispersion measurement, as also reported for experimental HSBRs by van der Sande et al. [9].

In the present DEM simulations, a slight residual decrease of the dispersion coefficient with increasing Δt\Delta t is observed even within the quasi–plateau region. This behaviour is consistent with a sub–linear scaling of the mean-squared axial displacement, ⟨Δz2⟩∝Δtα\langle\Delta z^{2} \rangle\propto\Delta t^{\alpha}. The fitted exponent α<1\alpha< 1 (Appendix § 6, Figure A.16) indicates sub-diffusive particle motion.

Based on the existence of a broad quasi-plateau region, and to retain sufficient temporal resolution while preserving sensitivity to operating conditions, a sampling interval of Δt=50 s\Delta t = 50\,\mathrm{s} was selected for all time–based dispersion calculations in this study. This choice follows the same physical criteria used in the experimental HSBR study of van der Sande et al. [9], while accounting for the slightly slower decorrelation dynamics observed in the present DEM system with scaled particles.

Axial Dispersion Verification via a Diffusion Model. To independently verify the axial dispersion coefficient obtained from the particle–based formulations, a one-dimensional diffusion model was solved in the axial direction. The solids transport was approximated by a Fickian diffusion equation for the local volume (or number) fraction of type–a particles, c(z,t)c(z,t),

∂c∂t=Dz∂2c∂z2,(21)\frac{\partial c}{\partial t} = D^{z}\frac{\partial^{2}c}{\partial z^{2}}, \tag*{(21)}

where zz is the axial coordinate and DzD^{z} is the constant axial dispersion coefficient to be tested. The domain length matched the DEM reactor geometry, 0≤z≤Lz0 \le z \le L_{z}.

The initial condition was chosen to mirror the axial labeling used for the DEM-based Lacey index. The reactor was split into two equal axial halves: the left half fully occupied by type-aa particles and the right half by type-bb particles,

c(z,0)={1,0≤z<Lz/2,0,Lz/2≤z≤Lz,(22)c(z,0)=\begin{cases} 1, & 0 \le z < L_z/2, \\ 0, & L_z/2 \le z \le L_z, \end{cases} \tag*{(22)}

corresponding to a fully segregated system with global composition fa=0.5f_a = 0.5.

Zero-flux (Neumann) boundary conditions were imposed at both ends of the reactor:

∂c∂z∣z=0=0,∂c∂z∣z=Lz=0,(23)\left.\frac{\partial c}{\partial z}\right|_{z=0}=0,\qquad\left.\frac{\partial c}{\partial z}\right|_{z=L_z}=0, \tag*{(23)}

preventing any net axial flux of labeled particles. Equation (21) with the initial and boundary conditions above was solved numerically using the pdepe solver in MATLAB on a uniform axial grid.

To compare the diffusion-model predictions with the DEM results, the numerical solution c(z,t)c(z,t) was post-processed to compute a model prediction of the axial Lacey index. The axial coordinate was divided into the same SS axial bins used in the DEM analysis, and the local fraction of type-aa particles in bin ii was taken as

fa,i(t)≈c(zi,t),f_{a,i}(t) \approx c(z_i,t),

where ziz_i is the bin center. From these sampled concentrations, the Lacey index was computed using the definition introduced in § 2.3. Because the diffusion equation produces a smooth continuum concentration field without sampling noise, the perfectly random-state variance σr2\sigma_r^2 tends to zero, consistent with the continuum limit of the Lacey formulation.

This procedure provides a model prediction of the axial Lacey index Mz(t)M_z(t) for any chosen value of the dispersion coefficient DzD^z. In the Results section, the diffusion–based prediction of Mz(t)M_z(t) is directly compared with the corresponding Mz(t)M_z(t) obtained from the DEM particle data. The agreement between these curves is used to assess the consistency of the independently computed dispersion coefficients (time–based (Einstein) and cycle–based) and to verify the selected value of DzD^z.

Results and discussion

Mixing and flowability measurements

Lacey index

As outlined in § 2.3, the Lacey index was used to monitor the evolution of mixing over time in the horizontal stirred bed reactor (HSBR). Figure 4 shows how the Lacey index evolved during mixing at 40 rpm rotation speed with a 40% fill level of polypropylene powder.

Figure 4. Lacey index in the axial direction (left) and cross-sectional direction (right) for the HSBR at a fill level of 40% and rotation speed of 40 rpm.

The plot on the left side of Figure 4 shows the axial mixing behavior. The Lacey index gradually increases and approaches 0.99 after about 102 seconds (68 rotations), indicating that the system reaches a nearly homogeneous state in the axial direction within that time.

In contrast, the plot on the right shows much faster mixing in the cross–sectional (xy)(xy) plane. Here, the Lacey index reaches 0.99 in just 6 seconds (4 rotations), reflecting a rapid approach to uniformity.

This stark difference in mixing time reflects the dominant transport mechanisms in each direction. Cross–sectional mixing is primarily driven by convective motion from the impeller blades, supplemented by dispersion, leading to rapid homogenization. In contrast, axial mixing relies predominantly on slower dispersive transport. As a result, mixing in the transverse plane is significantly faster than in the axial direction.

To visually illustrate the progression of mixing, Figure 5 and Figure 6 present simulation snapshots of particle distributions at selected time points in the axial and cross-sectional directions, respectively. These snapshots complement the Lacey index results by providing a visual representation of how the initially segregated red and blue particles become increasingly intermixed over time.

Figure 5. Simulation snapshots of axial mixing evolution in the HSBR at a fill level of 40% and rotation speed of 40 rpm.

Figure 6. Simulation snapshots of cross-sectional mixing evolution in the HSBR at a fill level of 40% and rotation speed of 40 rpm.

In the axial direction (Figure 5), even after 30 seconds, large regions of red and blue particles remain clearly distinguishable, with only limited interpenetration between the two zones. By 60 seconds, some red particles have dispersed into the blue region and vice versa, and by 102 seconds, the particle distribution appears largely uniform. This gradual transition aligns with the slower rise and eventual plateau of the axial Lacey index, highlight ing dispersive transport as the primary mixing mechanism along the reactors length.

In the cross-sectional plane (Figure 6), mixing proceeds much more rapidly. At 1.5 seconds, partial intermixing is evident, but distinct red and blue clusters still exist. At 3 seconds, segregation is still visible, yet clear convective movement can be observed as particles are rotated and redistributed. By 6 seconds, the system achieves near-complete homogenization. These patterns align with the sharp rise in the radial Lacey index and reflect the combined influence of convective transport, driven by the rotating blades, and localized particle dispersion, which together promote rapid radial mixing.

Mixing Time. To quantify the mixing behavior observed in both the axial (zz) and cross-sectional (xyxy) directions, the time evolution of the Lacey index was fitted to an exponential model of the form [56, 57, 58]:

Mfit(t)=1−exp⁡(−t−t0tm)(24)M^{\mathrm{fit}}(t) = 1 - \exp\left(-\frac{t-t_0}{t_{\mathrm{m}}}\right) \tag*{(24)}

Here, tt denotes time, t0t_0 is the offset time representing the delay before mixing begins, and tmt_{\mathrm{m}} is the characteristic mixing time. Smaller values of tmt_{\mathrm{m}} indicate faster mixing, while larger values reflect slower homogenization.

The fitting was carried out using the NelderMead simplex algorithm [59] (implemented via MATLABs fminsearch) to minimize the sum of squared residuals between the simulated Lacey index and the fitted curve. Figure 7 shows the simulated Lacey index data along with the fitted exponential curves for both axial and cross-sectional directions. The good agreement between simulation and fit supports the use of tmt_{\mathrm{m}} as a representative mixing timescale.

Figure 7. Fitted exponential curves (solid lines) and simulated Lacey index data (markers) for the axial direction (left) and cross-sectional direction (right) at a fill level of 40% and rotation speed of 40 rpm.

The uncertainties in tmt_m and t0t_0 were obtained from the covariance matrix of the nonlinear least-squares fit, following the standard procedure described in Reference [60]. The error bars represent one standard deviation.

The fitted tmt_m values are used as a quantitative measure of mixing efficiency and will be analyzed in § 3.2 to examine the effects of fill level and rotation speed.

Cycle time

The mean cycle time was first determined directly from the particle trajectories at a fill level of 40% and rotation speed of 40 rpm, following the procedure described in § 2. For each particle, the times ti,kct^c_{i,k} associated with the completion of successive revolutions were identified, and the corresponding cycle times Δti,kc\Delta t^c_{i,k} were computed as the difference between consecutive entries, according to Equation (6). The distribution of all valid cycle times collected across all particles is shown in Figure 8. The resulting mean cycle time is Δt‾c=3.63 s\overline{\Delta t}^{c} = 3.63\,\mathrm{s}, which represents the average time required for particles to complete one full revolution around the shaft.

Cycle time distribution and mean value for the HSBR

Figure 8. Cycle time distribution (blue) and mean value (red) for the HSBR at a fill level of 40% and rotation speed of 40 rpm.

To verify this trajectory-based measurement, a continuum coarse–graining (CG) analysis was performed using the framework detailed in § 2. DEM particle data from the first 120 s of the simulation were coarse–grained on a 50×5050 \times50 grid in the xy–plane using a Gaussian kernel of width equal to one mean particle radius. This provided spatially resolved fields for the solid volume fraction ϕ(x,y)\phi(x,y) and velocity components vx(x,y)v_x(x,y) and vy(x,y)v_y(x,y), from which the angular velocity field ω(x,y)\omega(x,y) was computed.

The resulting angular velocity field is shown in Figure 9. Using the mass–weighted averaging procedure introduced in § 2, the coarse–grained analysis produced a mean cycle time of Δt‾CGc=3.22 s\overline{\Delta t}^{c}_{\mathrm{CG}} = 3.22\,\mathrm{s}. This result differs from the trajectory-based value by approximately 11%, indicating good agreement between the two independent methods. This close correspondence confirms that the DEM-based definition of cycle time reliably captures the angular circulation dynamics within the HSBR.

Figure 9. Coarse-grained angular velocity field ω(x,y)\omega(x,y) obtained using MercuryCG. The color map represents the local angular velocity magnitude for the HSBR operated at a fill level of 40% and rotation speed of 40 rpm.

The mean cycle time will be further analyzed in § 3.2 to assess how fill level and rotation speed influence the particle angular motion and circulation dynamics.

Axial dispersion coefficient

Using the time-based (Einstein) formulation in Eq. (17) with a sampling interval of Δt=50\Delta t = 50 s, the axial dispersion coefficient for the HSBR operating at a 40% fill level and 40 rpm rotation speed was found to be Dz=3.6×10−5D^{z} = 3.6 \times10^{-5} m2^{2}/s.

The cycle-based formulation (Eq. (20)) provided an independent estimate.

For the same operating condition, this method yielded an axial dispersion of Dcz=1.18×10−4 m2/cycleD_c^z = 1.18 \times10^{-4}\ \mathrm{m^2/cycle}. Using the previously determined mean cycle time of Δtc‾=3.63 s\overline{\Delta t^c} = 3.63\ \mathrm{s} to convert this value to SI units resulted in an axial dispersion coefficient of 3.25×10−5 m2/s3.25 \times10^{-5}\ \mathrm{m^2/s}, which is in close agreement with the time-based estimate.

To further verify these results, the diffusion-based Lacey index comparison introduced in § 2 was applied. The axial diffusion equation was solved using DzD^z, and the predicted axial Lacey index Mz(t)M_z(t) was compared with the DEM-based Lacey index. As shown in Figure 10, the two curves nearly overlap. A similar comparison using DczD_c^z also produced good agreement.

Figure 10. Comparison of the axial Lacey index computed from DEM particle positions with the axial Lacey indices predicted by the diffusion model using the time–based (Einstein) and cycle–based dispersion coefficients.

Finally, by adjusting DzD^z in the diffusion model to obtain the best overlap with the DEM-based Lacey index, a fitted value of DDEMz=3.95×10−5 m2/sD_{\mathrm{DEM}}^z = 3.95 \times10^{-5}\ \mathrm{m^2/s} was extracted.

Overall, the three independently obtained dispersion coefficients (3.6×10−53.6 \times10^{-5}, 3.25×10−53.25 \times10^{-5}, and 3.95×10−5 m2/s3.95 \times10^{-5}\ \mathrm{m^2/s}) are mutually consistent and lie within about 22%22\% of each other. This consistency demonstrates that all three approaches capture the same underlying axial spreading mechanism in the HSBR and verifies the robustness of the time-based dispersion measurements.

In the following section (§ 3.2), the effect of operating conditions on axial dispersion is examined. In particular, the time–based axial dispersion coefficient DzD^{z} is analysed as a function of fill level and impeller rotation speed to determine how these parameters influence particle spreading along the reactor.

Effect of operating parameters (fill level and speed)

In this section, we studied the effect of operating parameters (fill level and rotation speed) on the mixing efficiency and flow dynamics of dry polypropylene powder in angular, radial, and axial directions.

Mixing time

First, we studied the effect of rotation speed and fill level on the parameter tmzt_{\mathrm{m}}^{z}, which represents the time required to achieve uniform mixing for axial (z–direction) mixing. It was observed that tmzt_{\mathrm{m}}^{z} consistently increases with fill level at all rotation speeds, while it decreases with increasing rotation speed across all fill levels.

At lower speeds (10–20 rpm, Fr=0.015−0.06Fr = 0.015 - 0.06), tmzt_{\mathrm{m}}^{z} is highest for the 70% fill level, reflecting the slower axial mixing due to increased material resistance and compaction effects. For lower fill levels (40%, 50%, and 60%), axial mixing is more efficient, resulting in lower tmzt_{\mathrm{m}}^{z} values. The reduced material resistance at these fill levels facilitates axial transport.

As the rotation speed increases (30–60 rpm, Fr=0.135−0.54Fr = 0.135 - 0.54), tmzt_{\mathrm{m}}^{z} decreases for all fill levels, indicating faster axial mixing. The stronger blade-induced forces enhance axial transport by promoting more uniform redistribution of material along the reactors length. For higher fill levels (e.g., 70%), the compaction effects limit the efficiency of axial transport, resulting in consistently higher tmzt_{\mathrm{m}}^{z} values even at higher speeds.

These results indicate that axial mixing efficiency depends strongly on both rotation speed and fill level. Higher rotation speeds promote faster mixing along the axial direction, as reflected by decreasing tmzt_{\mathrm{m}}^{z} values. However, as the fill level increases, axial mixing becomes slower due to greater material resistance and compaction, resulting in higher tmzt_{\mathrm{m}}^{z} values across all speeds.

Figure 11. Effect of the fill level and rotation speed on the parameter tmzt_m^z for axial mixing.

Second, we studied the effect of rotation speed and fill level on the parameter tmxyt_m^{xy}, for cross-sectional (xy-direction) mixing. We observed that tmxyt_m^{xy} decreases consistently with increasing rotation speed for all fill levels. However, the effect of fill level differs significantly compared to the axial mixing case.

At the lowest speed (10–20 rpm, Fr=0.015−0.06Fr = 0.015 - 0.06), tmxyt_m^{xy} is much higher for the 70% fill level compared to the other fills, indicating slower mixing. This behavior is attributed to material compaction at higher fill levels, which restricts the ability of the blades to effectively redistribute material radially within the cross-section. For lower fill levels (40%, 50%, and 60%), the parameter tmxyt_m^{xy} is approximately equal, reflecting similar mixing efficiency at these speeds. The reduced material resistance at these fill levels allows the blades to circulate the material more uniformly across the xy-plane.

By increasing speeds (20–30 rpm, Fr=0.06−0.135Fr = 0.06 - 0.135), tmxyt_m^{xy} for the 70% fill level decreases and approaches the values of the lower fill levels, reflecting ing improved mixing efficiency as blade-induced forces overcome compaction effects. At these speeds, slight differences emerge among the lower fill levels.

At higher speeds (40–60 rpm, Fr=0.24−0.54Fr = 0.24 - 0.54), tmxyt_{\mathrm{m}}^{\mathrm{xy}} for all fill levels becomes nearly equal, suggesting that the effect of fill level diminishes at these speeds. For the 70% and 40% fill levels, tmxyt_{\mathrm{m}}^{\mathrm{xy}} values are slightly higher compared to 50% and 60%. This indicates that the blades effectively redistribute material across the cross-sectional regardless of fill level at higher speeds, although compaction effects still slightly hinder mixing at the 70% fill level.

These results show that rotation speed has more effect on cross-sectional mixing compared to fill level, with higher speeds consistently improving mixing efficiency (lower tmxyt_{\mathrm{m}}^{\mathrm{xy}} values). The differences in tmxyt_{\mathrm{m}}^{\mathrm{xy}} at lower speeds reflect the combined effects of material distribution around the shaft and compaction at higher fill levels, which are minimized as rotation speed increases.

Figure 12. Effect of the fill level and rotation speed on the parameter tmxyt_{\mathrm{m}}^{\mathrm{xy}} for cross-sectional mixing.

Cycle Time

An overview of the effect of impeller speed and reactor filling on the cycle time of large particles (d>dv,75d > d_{v,75}) (where dv,75d_{v,75} is the particle size at 75% cumulative volume; see Table 1) is illustrated in Figure 13. It can be observed that cycle time decreases as rotation speed and fill level increase. The most considerable point is the influence of the filling level on the angular motion. Low filling (40%) causes improper angular motion through the reactor, and long particle cycle periods. This causes inhomogeneous heat removal. Therefore, to have a low cycle time, which is demanded for uniform heat removal, we should increase the fill level and impeller speed.

Figure 13. Comparison of cycle time obtained from DEM simulations using two particle scale factors with experimental measurements [9].

To validate our findings, we compared the simulated cycle times with the experimental measurements reported by van der Sande et al. [9]. The simulations reproduce the experimentally observed trends across rotation speeds and fill levels, indicating that the DEM framework captures the essential mechanisms governing particle circulation in HSBRs. Furthermore, reducing the scale factor (i.e., approaching the actual particle size) improves agreement with the experimental results, particularly in terms of magnitude.

Axial Dispersion Coefficient

Figure 14 shows the effect of reactor filling and rotation speed on the axial dispersion coefficient. It can be observed that the axial dispersion coefficient increases linearly with increasing impeller speed and decreasing fill level.

Figure 14. Comparison of axial dispersion coefficient obtained from DEM simulations with two scaling factors and experimental measurements [9].

The DEM–based dispersion coefficients were compared with the experimental data reported by van der Sande et al. [9], showing consistent agreement in trends across operating conditions. As observed for cycle time, reducing the particle size improves agreement with the experimental results, particularly in terms of magnitude.

Conclusions and outlook

This study investigated the influence of rotation speed and fill level on key aspects of particle dynamics and mixing efficiency within a lab–scale horizontal stirred bed reactor (HSBR). Using Discrete Element Method (DEM) simulations validated against experimental data by van der Sande et al. [9], we analysed the key mechanisms governing reactor performance under different operating conditions.

Axial mixing was found to depend strongly on both rotation speed and fill level. Higher rotation speeds promote faster mixing along the axial direction. However, as the fill level increases, axial mixing becomes slower due to greater material resistance and compaction. Cross–sectional mixing, on the other hand, was found to be more sensitive to rotation speed than to fill level. Higher speeds consistently enhanced mixing efficiency, while variations at lower speeds highlighted the influence of material distribution around the shaft and compaction at higher fill levels.

Cycle time, the average time required for a particle to complete one revolution around the shaft, decreased with increasing rotation speed and fill level. At lower fill levels, prolonged cycle times led to inefficient heat removal due to insufficient particle circulation. In contrast, higher fill levels led to shorter cycle times, thereby improving heat transfer. The strong qualitative agreement between our simulation results and previously published experimental data confirms the reliability of the DEM study, with quantitative accuracy improving as the particle size scaling factor was reduced.

Axial dispersion coefficients increased with higher rotation speeds but decreased with increasing fill levels. Enhanced particle mobility at higher speeds and greater displacement in less compacted beds contributed to this trend. The alignment of these results with existing experimental findings illustrates the ability of the simulations to capture complex flow dynamics within the HSBR.

Overall, the results reveal a clear operational trade–off. Increasing the rotation speed consistently improves mixing efficiency, reduces the cycle time, and enhances axial dispersion. Increasing the fill level, in contrast, shortens the cycle time and promotes solids circulation, but simultaneously reduces axial mixing efficiency and axial dispersion. These findings highlight the need to balance operating conditions to achieve optimal reactor performance. The agreement between simulations and experimental measurements further supports the validity of the DEM approach. Consequently, the DEM framework developed in this work provides a powerful tool for analysing these interactions and guiding industrial reactor optimization.

Future work should extend this analysis to full–scale HSBRs and diverse particulate systems. Incorporating more realistic particle shapes and further reducing particle size scaling in simulations can improve the accuracy and applicability of the results. Experimental validation involving multi–particle systems would also bridge the gap between simulation and practical implementation, further enhancing the utility of HSBRs in industrial polymerization processes.

Acknowledgment

This research is part of the Industrial Dense Granular Flows project, which received funding from the Dutch Research Council (NWO) in the framework of the ENW PPP Fund for the top sectors and from the Ministry of Economic Affairs under the PPS-Toeslagregeling.

The authors sincerely thank Professor Stefan Luding and Professor Anthony Thornton for valuable scientific discussions and partial supervision.

Influence of the sampling interval on the axial dispersion coefficient

To determine a suitable sampling interval Δt\Delta t for the time–based (Einstein) axial dispersion coefficient, the dispersion was evaluated over a wide range of Δt\Delta t values. Figure A.15 shows the resulting dependence of DzD^z on Δt\Delta t.

At very small Δt\Delta t, the calculated dispersion coefficient is strongly affected by short–time correlations. As Δt\Delta t increases, DzD^z decreases rapidly and then enters a quasi–plateau region in which its dependence on Δt\Delta t becomes much weaker.

Figure A.15. Axial dispersion coefficient DzD^{z} as a function of the sampling interval Δt\Delta t for different HSBR operating conditions (see legend). The analysis is performed for the selected particle size class defined by d>dv,75d > d_{v,75}.

To further analyse this behaviour, the mean squared axial displacement (MSD) was evaluated as a function of Δt\Delta t, as shown in Figure A.16. The MSD follows a power–law relation

⟨(Δz)2⟩∝Δtα,(25)\langle(\Delta z)^2\rangle\propto\Delta t^{\alpha}, \tag*{(25)}

with α<1\alpha< 1, indicating sub–diffusive axial transport. As a consequence, the apparent dispersion coefficient derived from ⟨(Δz)2⟩/(2Δt)\langle(\Delta z)^2\rangle/(2\Delta t) does not reach a perfectly constant plateau but decreases slowly with increasing Δt\Delta t.

This sub–diffusive behaviour reflects long–time correlations in the axial displacement, which may be promoted by dense frictional contacts and the structured, blade–driven circulation in the HSBR. In addition, the particle upscaling used in the DEM model can reduce collisional diffusion and prolong decorrelation relative to the experimental system.

Figure A.16. Mean squared axial displacement as a function of the sampling interval Δt\Delta t for different HSBR operating conditions (see legend). Symbols denote simulation data and solid lines represent fits in the range 25 s<Δt<100 s25\,\mathrm{s} < \Delta t < 100\,\mathrm{s}. The slope of each fitted line corresponds to the exponent α\alpha, indicating sub-diffusive behaviour. The analysis is performed for the selected particle size class defined by d>dv,75d > d_{\mathrm{v},75}.

Based on this analysis, a sampling interval of Δt=50 s\Delta t = 50\,\mathrm{s} was selected for the evaluation of the axial dispersion coefficient in this study. This value lies within the quasi-plateau region, avoids short-time correlation effects, and retains sufficient sensitivity to operating conditions.

References

  1. [1]J. W. Shepard, J. L. Jezl, E. F. Peters, R. D. Hall, Divided horizontal reactor for the vapor phase polymerization of monomers at different hydrogen levels, US Patent 3,957,448 (May 18 1976).
  2. [2]M. Caracotsios, Theoretical modelling of amoco’s gas phase horizon-tal stirred bed reactor for the manufacturing of polypropylene resins, Chemical engineering science 47 (9-11) (1992) 2591–2596. doi:10.1016/0009-2509(92)87098–B.DOI
  3. [3]J. B. Soares, T. F. McKenna, Polyolefin reaction engineering, John Wiley & Sons, 2013. doi:10.1002/9783527646944.DOI
  4. [4]S. Pourandi, P. C. van der Sande, I. Ostanin, T. Weinhart, Calibration of a DEM contact model for wet industrial granular materials, Powder Technology 476 (2026) 122404. doi:10.1016/j.powtec.2026.122404.DOI
  5. [5]A. Gorbach, S. Naik, W. Ray, Dynamics and stability analysis of solid catalyzed gas-phase polymerization of olefins in continuous stirred bed reactors, Chemical engineering science 55 (20) (2000) 4461–4479. doi:10.1016/S0009-2509(00)00080–4.DOI
  6. [6]J. J. Zacca, J. A. Debling, W. Ray, Reactor residence time distribu-tion effects on the multistage polymerization of olefinsi. basic principles and illustrative examples, polypropylene, Chemical Engineering Science 51 (21) (1996) 4859–4886. doi:10.1016/0009-2509(96)00258–8.DOI
  7. [7]C. J. Dittrich, S. M. Mutsers, On the residence time distribution in reac-tors with non-uniform velocity profiles: The horizontal stirred bed reac-tor for polypropylene production, Chemical engineering science 62 (21) (2007) 5777–5793. doi:10.1016/j.ces.2007.06.031.DOI
  8. [8]Z. Tian, X.-P. Gu, L.-F. Feng, J.-P. Corriou, G.-H. Hu, Modeling and simulation of polypropylene particle size distribution in industrial hori- zontal stirred bed reactors, Journal of applied polymer science 125 (4) (2012) 2668–2679. doi:10.1002/app.36473.
  9. [9]P. C. van der Sande, E. C. Wagner, J. de Mooij, G. M. Meesters, J. R. van Ommen, Particle dynamics in horizontal stirred bed reactors characterized by single-photon emission radioactive particle tracking, Chemical Engineering Journal 482 (2024) 149100. doi:10.1016/j.cej.2024.149100.DOI
  10. [10]P. C. van der Sande, C. de Vries, E. C. Wagner, A. C. Vögtlander, G. M. Meesters, J. R. van Ommen, Flow behavior of polypropylene reactor powder in horizontal stirred bed reactors characterized by x-ray imaging, Chemical Engineering Journal 500 (2024) 156891. doi:10.1016/j.cej.2024.156891.DOI
  11. [11]S. Pourandi, P. C. van der Sande, T. Weinhart, I. Ostanin, A mathematical model for the dynamic angle of repose of a granular material in the rotating drum, Powder Technology 446 (2024) 120176. doi:10.1016/J.POWTEC.2024.120176.DOI
  12. [12]P. A. Cundall, O. D. Strack, A discrete numerical model for granular assemblies, geotechnique 29 (1) (1979) 47–65. doi:10.1680/geot.1979.29.1.47.DOI
  13. [13]T. Weinhart, L. Orefice, M. Post, M. P. van Schrojenstein Lantman, I. F. Denissen, D. R. Tunuguntla, J. Tsang, H. Cheng, M. Y. Shaheen, H. Shi, et al., Fast, flexible particle simulationsan introduction to mercurydpm, Computer physics communications 249 (2020) 107129. doi:10.1016/j.cpc.2019.107129.
  14. [14]A. R. Thornton, T. Plath, I. Ostanin, H. Götz, J.-W. Bisschop, M. Hassan, R. Roeplal, X. Wang, S. Pourandi, T. Weinhart, Recent advances in mercurydpm, Mathematics in Computer Science 17 (2) (2023) 13. doi:10.1007/s11786-023-00562-x.DOI
  15. [15]A. Di Renzo, F. P. Di Maio, Comparison of contact-force models for the simulation of collisions in DEM-based granular flow codes, Chemical engineering science 59 (3) (2004) 525–541. doi:10.1016/j.ces.2003.09.037.
  16. [16]S. Luding, Cohesive, frictional powders: contact models for tension, Granular matter 10 (4) (2008) 235–246. doi:10.1007/s10035-008-0099-x.DOI
  17. [17]C. Wensrich, A. Katterfeld, Rolling friction as a technique for modelling particle shape in dem, Powder Technology 217 (2012) 409–417. doi:10.1016/j.powtec.2011.10.057.DOI
  18. [18]T. Roessler, A. Katterfeld, Scaling of the angle of repose test and its influence on the calibration of dem parameters using upscaled particles, Powder technology 330 (2018) 58–66. doi:10.1016/j.powtec.2018.01.044.DOI
  19. [19]S. C. Thakur, J. P. Morrissey, J. Sun, J. Chen, J. Y. Ooi, Micromechanical analysis of cohesive granular materials using the discrete element method with an adhesive elasto-plastic contact model, Granular Matter 16 (2014) 383–400. doi:10.1007/s10035-014-0506-4.
  20. [20]Y. J. Huang, O. J. Nydal, B. Yao, Time step criterions for nonlinear dense packed granular materials in time-driven method simulations, Powder technology 253 (2014) 80–88. doi:10.1016/j.powtec.2013.10.010.DOI
  21. [21]Z. Yan, S. Wilkinson, E. Stitt, M. Marigo, Discrete element modelling (DEM) input parameters: understanding their impact on model predictions using statistical analysis, Computational Particle Mechanics 2 (2015) 283–299. doi:10.1007/s40571-015-0056-5.DOI
  22. [22]S. Lommen, D. Schott, G. Lodewijks, DEM speedup: Stiffness effects on behavior of bulk material, Particuology 12 (2014) 107–112. doi:10.1016/j.partic.2013.03.006.DOI
  23. [23]P. M. C. Lacey, Developments in the theory of particle mixing, Journal of applied chemistry 4 (5) (1954) 257–268. doi:10.1002/jctb.5010040504.DOI
  24. [24]P. Lacey, The mixing of solid particles, Chemical Engineering Research and Design 75 (1997) S49–S55. doi:10.1016/S0263-8762(97)80004-4.DOI
  25. [25]B. Füvesi, C. Goniva, S. Luding, V. Magnanimo, Mixing indices in up-scaled simulations, Powder Technology 456 (2025) 120775. doi:10.1016/j.powtec.2025.120775.DOI
  26. [26]R. J. Brandao, R. M. Lima, R. L. Santos, C. R. Duarte, M. A. Barrozo, Experimental study and dem analysis of granular segregation in a rotating drum, Powder Technology 364 (2020) 1–12. doi: 10.1016/j.powtec.2020.01.036.DOI
  27. [27]M. Jiang, Y. Zhao, G. Liu, J. Zheng, Enhancing mixing of particles by baffles in a rotating drum mixer, Particuology 9 (3) (2011) 270–278. doi:10.1016/j.partic.2010.06.008.DOI
  28. [28]P. Liu, R. Yang, A. Yu, Dem study of the transverse mixing of wet particles in rotating drums, Chemical Engineering Science 86 (2013) 99–107. doi:10.1016/j.ces.2012.06.015.DOI
  29. [29]Z. Zhang, N. Gui, L. Ge, Z. Li, Numerical study of particle mixing in a tilted three-dimensional tumbler and a new particle-size mixing index, Advanced Powder Technology 30 (10) (2019) 2338–2351. doi: 10.1016/j.apt.2019.07.017.DOI
  30. [30]P. Widhate, H. Zhu, Q. Zeng, K. Dong, Mixing of particles in a rotating drum with inclined axis of rotation, Processes 8 (12) (2020) 1688. doi: 10.3390/pr8121688.DOI
  31. [31]A. V. Patil, J. Hofsteenge, J. M. Bujalski, S. T. Johansen, Dpm model segregation validation and scaling effect in a rotary drum, Computational Particle Mechanics 9 (4) (2022) 693–707. doi:10.1007/s40571-021-00438-6.DOI
  32. [32]R. van Sleeuwen, S. Pantaleev, M. Ebrahimi, et al., Efficient dem modeling of solid flavor particle mixing in a rotary drum, Powder Technology 437 (2024) 119559. doi:10.1016/j.powtec.2024.119559.DOI
  33. [33]X. Tang, S. Wang, X. Jin, Y. Shen, Super-quadric cfd-dem modelling of chip-like particle-liquid flow in a rotary drum, Powder Technology 435 (2024) 119363. doi:10.1016/j.powtec.2024.119363.DOI
  34. [34]L. Xu, X. Wu, M. Liu, S. Bao, B. Liu, Z. Lu, Studying on the upper limit of the coarse-grained dem model for simulating mixing processes in a rolling drum with complex wall boundary, Advanced Powder Technology 35 (1) (2024) 104307. doi:10.1016/j.apt.2023.104307.DOI
  35. [35]J. Song, N. Naeem, H. Fan, G. Li, M. Li, Y. Guo, J. Lin, J. Tie, Flow and mixing of elongated particles in rotating drums of different sizes, Frontiers in Chemical Engineering 7 (2025) 1636010. doi:10.3389/fceng.2025.1636010.DOI
  36. [36]G. Chandratilleke, X. Jin, Y. Shen, Dem study of effects of particle size and density on mixing behaviour in a ribbon mixer, Powder Technology 392 (2021) 93–107. doi:10.1016/j.powtec.2021.06.058.DOI
  37. [37]X. Jin, G. R. Chandratilleke, S. Wang, Y. Shen, Dem investigation of mixing indices in a ribbon mixer, Particuology 60 (2022) 37–47. doi:10.1016/j.partic.2021.03.005.DOI
  38. [38]G. R. Chandratilleke, A. Yu, R. Stewart, J. Bridgwater, Effects of blade rake angle and gap on particle mixing in a cylindrical mixer, Powder Technology 193 (3) (2009) 303–311. doi:10.1016/j.powtec.2009.03.007.DOI
  39. [39]G. R. Chandratilleke, A. Yu, J. Bridgwater, K. Shinohara, A particle- scale index in the quantification of mixing of particles, AIChE journal 58 (4) (2012) 1099–1118. doi:10.1002/aic.12654.
  40. [40]G. R. Chandratilleke, A. Yu, J. Bridgwater, A dem study of the mixing of particles induced by a flat blade, Chemical Engineering Science 79 (2012) 54–74. doi:10.1016/j.ces.2012.05.010.DOI
  41. [41]M. Halidan, G. Chandratilleke, S. Chan, A. Yu, J. Bridgwater, Prediction of the mixing behaviour of binary mixtures of particles in a bladed mixer, Chemical Engineering Science 120 (2014) 37–48. doi:10.1016/j.ces.2014.08.048.DOI
  42. [42]R. Chandratilleke, A. Yu, J. Bridgwater, K. Shinohara, Flow and mixing of cohesive particles in a vertical bladed mixer, Industrial & Engineering Chemistry Research 53 (10) (2014) 4119–4130. doi:10.1021/ie403877v.DOI
  43. [43]D. K. Chibwe, G. M. Evans, E. Doroodchi, B. J. Monaghan, D. J. Pinson, S. J. Chew, Particle near-neighbour separation index for quantification of segregation of granular material, Powder Technology 360 (2020) 481–492. doi:10.1016/j.powtec.2019.10.079.DOI
  44. [44]K. Korkerd, Z. Zhou, R. Zou, P. Piumsomboon, B. Chalermsinsuwan, Effect of immersed tubes configurations on mixing and heat transfer of mixed biomass and silica sand in a bubbling fluidized bed using cfd-dem and statistical experimental design analysis, Powder Technology 437 (2024) 119542. doi:10.1016/j.powtec.2024.119542.DOI
  45. [45]W. Godlieb, N. Deen, J. Kuipers, Characterizing solids mixing in dem simulations, in: 6th International Conference on Multiphase Flow, ICMF 2007, 2007.
  46. [46]R. Kamesh, S. Vaddagani, C. Sumana, K. Y. Rani, S. R. Gopireddy, N. A. Urbanetz, Six-directional sampling method and mean mixing indices for solids blending performance analysis of dem simulations, Powder Technology 398 (2022) 117051. doi:10.1016/j.powtec.2021.117051.DOI
  47. [47]H. Norouzi, R. Zarghami, N. Mostoufi, Insights into the granular flow in rotating drums, Chemical engineering research and design 102 (2015) 12–25. doi:10.1016/j.cherd.2015.06.010.DOI
  48. [48]T. Weinhart, A. R. Thornton, S. Luding, O. Bokhove, From discrete particles to continuum fields near a boundary, Granular Matter 14 (2) (2012) 289–294. doi:10.1007/s10035-012-0317-4.DOI
  49. [49]S. Yang, Y. Sun, J. Wang, A. Cahyadi, J. W. Chew, Influence of operating parameters and flow regime on solid dispersion behavior in a gas–solid spout-fluid bed, Chemical Engineering Science 142 (2016) 112–125. doi:10.1016/j.ces.2015.11.038.DOI
  50. [50]S. Yang, L. Zhang, K. Luo, J. W. Chew, Dem investigation of the axial dispersion behavior of a binary mixture in the rotating drum, Powder Technology 330 (2018) 93–104. doi:10.1016/j.powtec.2018.02.021.DOI
  51. [51]A. Einstein, Über die von der molekularkinetischen theorie der wärme geforderte bewegung von in ruhenden flüssigkeiten sus- pendierten teilchen, Annalen der physik 4 (1905). doi:10.1002/andp.19053220806.DOI
  52. [52]A. Einstein, Investigations on the Theory of the Brownian Movement, Courier Corporation, 1956.
  53. [53]K. Luo, S. Yang, K. Zhang, M. Fang, J. Fan, Particle dispersion and circulation patterns in a 3d spouted bed with or without draft tube, Industrial & Engineering Chemistry Research 52 (28) (2013) 9620–9631. doi:10.1021/ie303555g.DOI
  54. [54]A. Ingram, J. Seville, D. Parker, X. Fan, R. Forster, Axial and radial dispersion in rolling mode rotating drums, Powder technology 158 (1-3) (2005) 76–91. doi:10.1016/j.powtec.2005.04.030.DOI
  55. [55]D. Pallares, F. Johnsson, A novel technique for particle tracking in cold 2-dimensional fluidized bedssimulating fuel dispersion, Chemical Engineering Science 61 (8) (2006) 2710–2720. doi:10.1016/j.ces.2005.11.030.DOI
  56. [56]P. Liu, R. Yang, A. Yu, The effect of liquids on radial segregation of granular mixtures in rotating drums, Granular Matter 15 (4) (2013) 427–436. doi:10.1007/s10035-013-0392-1.DOI
  57. [57]M. Halidan, G. Chandratilleke, K. Dong, A. Yu, Mixing performance of ribbon mixers: Effects of operational parameters, Powder Technology 325 (2018) 92–106. doi:10.1016/j.powtec.2017.11.009.DOI
  58. [58]A. Herman, J. Gan, A. Yu, The effect of rotation speed and mixer size on granular flow and mixing in bladed mixers, in: EPJ Web of Conferences, Vol. 249, EDP Sciences, 2021, p. 03036. doi:10.1051/ epjconf/202124903036.
  59. [59]J. C. Lagarias, J. A. Reeds, M. H. Wright, P. E. Wright, Convergence properties of the nelder–mead simplex method in low dimensions, SIAM Journal on optimization 9 (1) (1998) 112–147. doi: 10.1137/S1052623496303470.DOI
  60. [60]W. H. Press, Numerical recipes 3rd edition: The art of scientific computing, Cambridge university press, 2007.

Paper details

vlnc:W_c3gx6m48
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents