Assessing the representativeness of durability tests for wood pellets by DEM Simulation – Comparing conditions in a durability test with transfer chutes

. Dust generation when handling wood pellets is related to the durability of the product, in other words the wear rate of particles subject to forces. During transport, storage and handling wood pellets undergo different forces when interacting with different pieces of equipment. This paper assesses the representativeness of the tumbling can test in relation to transfer chutes, by comparing forces acting on wood pellets in durability tests and in transfer chutes using DEM. The study also incorporates effects such as shape and size variations. The results showed that the tumbling can test underestimates compressive and tangential forces. Since the tested material is subject to milder conditions than in reality, it can be concluded that this test is not representative for the conditions in the supply chain of wood pellets.


Introduction
Over the years a standard for wood pellets (EN15210 or ISO/NP17831-1) [1] has been developed to assess the durability of materials amongst others applied to wood pellets, such as a tumbling can. However, it is unlikely that this is representative for the handling in the whole supply chain because real operational conditions can greatly differ in terms of forces from tests on lab-scale as is suggested by previous work on the topic comparing durability tests to filling and reclaiming of silos [2]. Moreover, in the biomass industry the problem of dust generation and generation of fines (crumbling off of smaller parts of the pellets) remains, despite a standard being in place.
This paper assesses the representativeness of the tumbling can test in relation to transfer chutes, by comparing forces acting on wood pellets in durability tests and in five transfer chutes using DEM. In addition the effect of the particle composition of a wood pellet is investigated.

DEM simulations
Discrete Element Method [3] is a numerical technique to analyse the movement and interaction of multiple small bodies in a system. In this paper EDEM 2.6.1 [4] has been used as the DEM simulation software.

Materials: Pellet Models
Wood pellets are cylindrical shaped particles of varying length. They are created by compressing smaller wood particles together in a pellet mill. Here 12 mm wood pellets are considered.
For the simulations 3 different pellet shapes were used. The shape variation is shown in Figure 1 The equivalent sphere diameter (D eq ) of shape c) was calculated using : For shape a) and b) 3 different lengths were defined. The size distribution follows a total mass ratio of 4:3:3 for the 22mm, 30mm and 18mm pellet lengths respectively. An overview of particle shapes and size distribution is given in Table 1. For shapes a) 3 spheres and b) 35 spheres, two different contact models were applied: Hertz-Mindlin no-slip and Hertz-Mindlin with bonding. This means that the spheres are flexibly bonded together instead of a rigid connection. This gives a total of 7 different pellet models tested.

Calibration
In order to ensure realistic behaviour of the particles inside the tumbling can and the transfer chutes, the material model will be calibrated with 3 laboratory tests; the bulk density test, the lifting can test and the ledge test.
The particle density was measured by putting a sample of 100g in a measuring cylinder filled with water. The volume was determined by the immediate increase of the water level, which indicates that minimum water is absorbed and the particle density is unaffected.
The ledge test is a box open at the top ( Figure 2a). Two faces are made of wood and the rest is made of transparent acrylic. One wooden face is latched and hinged at the upper end. The inner height is 26cm, the door is 20cm and the viewing acrylic walls are 22 cm wide. First, the box is filled with pellets up to an approximate height of 14-15cm from the base. Then the door is unlatched and the material flows out until it reaches a certain angle of repose which was noted down. This was repeated 20 times.
The lifting can test is a cylinder open at both ends of 20cm height and 26cm diameter, i.e. more than 20 times the diameter of the sample pellet. The can is put on a base and then filled by pouring in pellets from the top until it is filled up to a height of 14-15cm. It is then lifted vertically and material escapes out from the bottom and spreads on the base resulting in formation of a pile ( Figure 2b). The slope of this pile is measured from 5 different directions. This is repeated thrice, leading to 15 observations.
After determining the mean and standard deviation of the observations, the upper and lower limit values of the 95% confidence interval were obtained for the sample.
These tests will be simulated with DEM as well. For that, different combinations of parameters were tested and the results compared to the experimental results.
After a number of iterations a calibrated model was achieved.
Since the value of the coefficient of static friction (wood-wood) in general lies between 0.2-0.5 and the rolling friction coefficient lies between 0-0.05 [5][6], the possible combinations were tested within this range.
Once calibrated, the model will then be used in simulations with the tumbling can test and the transfer chutes.

Tumbling can test
The tumbling can or rotating drum ( Figure 3) is a test device with which the durability of wood pellets is determined under mechanical handling conditions. Other durability testers, such as the Holmen durability tester and the Ligno tester focus on the pneumatic handling of pellets. Compared to the tumbling can, both testers operate in a smaller time frame, where the pellets are exposed to higher destructive forces [7]. Nevertheless, according to Temmerman et al. [8] more repeatable and reproducible results are achievable with the tumbling can compared to the Ligno tester. The Hood & Spoon model (Figure 5a) is an upgraded version of the above model. It has an added curved (spoon) exit face to ensure smooth exit of particles to the output conveyor belt besides the smooth entry. The Tralix (Transfer Helix) is depicted in Figure 5b and provides a helical path for particles to flow in a more streamlined and non-accumulative manner.
The modelled input and output belt speed was 4 m/s. The vertical distance between belts is 4.7m, except in helix where it is 11.2m. The mass flow is 1044 tph. The simulations were run twice with the simple pellet model (Figure 1a) and each took between 25 and 30 hours.

Calibrated material
The angle of repose had a mean value of 41.1 degrees, standard deviation of 2.4 degrees (95% confidence interval being 39.9-42.3 degrees) for the ledge test and a mean of 24.4 degrees, standard deviation of 2.4 degrees (95% confidence interval being 23.1-25.7 degrees) for the lifting can test.
The angle of repose results obtained via simulations were matched to lie in the confidence interval of the experimental results and then ranked in order of closeness to the mean value, with the value closer to the mean having a higher rank. Then the data that qualified for both cases was checked by summing the ranks achieved in the 2 cases. The lower the sum, the closer the results are to the experiments.
From the analysis the coefficient of static and rolling friction was ascertained as 0.4 and 0.04 for 12 mm wood pellets. The full listing of DEM parameters is given in Table 2.

Tumbling can test
With the calibrated model, simulations of the tumbling can were performed for the various pellet models introduced in Section 2.1.1. The particle trajectories in the tumbling can are illustrated in Figure 6. The forces acting on the particles during rotation of the can are analysed and the compressive force is depicted in Figure 7. The advanced shape particle is at the higher end in all cases, i.e. when all are non-bonded models (1 st three series) and when all are bonded models (next three series). This is probably due to the fact that the advanced shape model has a more realistic shape, lesser vacant voids and since it is made of more spheres, it has a higher area of contact with the surroundings. The other shapes and sizes undergo slightly smaller compressive forces.
Also the pellets built with bonded contact models experience slightly higher compressive forces compared to their respective non-bonded counterparts. However, a significant effect of the pellet model on the average compressive forces could not be shown. From a comparison of tangential and compressive forces the average tangential force is lower by a factor of 6-7 (not shown here). It can be inferred that the tangential forces do not play a critical role compared to the compressive forces for the durability of particles. The angular velocity analysis reveals how size variation leads to bigger size particles providing more gaps for the smaller particles to rotate and roll in, whereas the advanced shape provides indented surface leading to more friction and difficulty in rolling.

Transfer chutes
The average particle compressive force for the five different configurations of the transfer chutes is depicted in Figure 8. The initial behaviour (up to 1.5s) is common as the particles are in the feeding conveyer and are about to enter the chutes. After that the difference can be seen among the different chutes. The hood & spoon, the hood only and the sloped models have almost identical performance. After 2.5s the particle forces in the transfer chutes reach a steady state except for the rockbox and the Tralix. For the rockbox, this is because the accumulation and initial heap formation takes more time. After the flow in the rockbox is established it can be seen that the average compressive force is almost twice the value of the other configurations. The Tralix is gentler as it carries the bulk down gradually in uncrowded flow instead of particles falling down and impacting.
The pattern for tangential forces (not shown here) is similar except that the tangential forces are 7-8 times smaller than the compressive forces. The values of the hood and spoon configuration are closer to the rockbox.

Conclusions
Based on the DEM simulation results in this paper the tumbling can cannot be considered as representative for the conditions in transfer chutes. From a comparison of forces acting on the particles it follows that the average compressive forces and tangential forces in the tumbling can are underestimated by a factor of 10-20. This is compared to a mass flow of 1044 tph and velocities of 4m/s which in industrial practice are often higher.
Several shapes and particle contact models of wood pellets were tested in the tumbling can. The advanced shape consists of 35 spheres and is most realistic in design due to greater contact area and sharper edges that greatly affect the particle dynamics. However, in the tumbling can simulation a significant difference in compressive and tangential forces could not be shown. In addition, a disadvantage of the advanced shape is the increase in computational times.
Five designs of transfer chutes under similar simulation conditions were compared. The Tralix has the lowest average compressive forces and the traditional rockbox is the worst in that aspect.
The durability testing mechanism should be changed to ensure a better representation of the actual supply chain conditions. The improved tests must subject the material to higher compressive and tangential forces.