**3. Results and Discussion**

### *3.1. General Fuel Charactersitic of the Torrefied Biomass Pellets*

The seven different biomass materials all underwent slightly different torrefaction conditions with a TFtemp range between 260 ◦C and 308 ◦C while the residence time of the biomass in the reactor varied between 9 and 30 min (see Table 2). In order to compare the relative change of the material caused by torrefaction, the change in volatile content was chosen as TFdgr, which, in this case, ranged between 6 and 9% DM. Regarding the HHV, which is increased during torrefaction, values found by other torrefaction studies were in some cases similar to what was obtained in this study and in other cases not. Gucho et al. [26] for example found torrefied beech wood both at 260 ◦C and at 280 ◦C for 30 min and the mean HHV between the two is the same as the value obtained here for 30 min at 270 ◦C. Phanphanich and Mani [27] on the other hand torrefied pine chips at 300 ◦C for 30 min and obtained a HHV of 25.4 MJ/kg, much higher than the 21.9 MJ/kg observed here. However, both studies torrefied on lab scale in batch reactors and it is known from studies on biomass pyrolysis that particles size and sample size, let alone the reactor type, will have an impact on the product [28]. Reports on increasing HHV with increasing intensity of torrefaction when using pilot scale plants are available and support the findings here [29,30].


**Table 2.** Torrefaction conditions, fuel characteristics after torrefaction and pellet characteristics of the seven torrefied raw-materials.

> \* not recorded.

Specific requirements for chemical and physical-mechanical characteristic of the pellets made from thermally treated are currently finalized in a technical specification (TS) on international level as ISO/PRF TS 17225-8 [22]. Once finalized, ISO/PRF TS 17225-8 will have an outline much like the ISO 17225-2 [31] for pellets made from wood or ISO 17225-6 [32] for pellets made from non-woody biomass. However, as the market of thermally treated biomass is still developing a TS will serve as a placeholder until a full standard can be agreed upon. The main categorization in this TS is between woody biomass of energy content (HHV) below or above 21 MJ/kg DM. Again, there is a differentiation in woody and non-woody biomass and three respective quality classes, however, for non-woody biomass, the 21 MJ/kg DM threshold does not exist. In order to produce a fuel that is above or below 21 MJ/kg DM, the torrefaction plant operators will have to decide how they run their reactor accordingly. In case of this study, which took place before the conception of this TS, almost all woody fuels are above 21 MJ/kg DM (except for the beech pellets) which is why the requirements for bulk density, durability, moisture and ash will be discussed based on the thresholds that apply to that category. None of the here investigated woody biomass pellets could meet the criteria for the highest quality class. Even the second highest class, which requires bulk density to be >650 kg/m3, durability to be >96%, moisture to be <8% water content, and ash to be <3%, could only be met by spruce pellets (see Table 2). However, bulk density, durability, and moisture of the pellets are all parameters that are a result of pelletizing process while the HHV is a result of the torrefaction process. Creating standardized high quality fuel pellet will only be a problem when ash content of the input material is already high as it will only be increased during torrefaction. The straw pellets produced here were able to meet many of the requirements for the highest quality class of pellets made from thermally treated non-woody biomass such as bulk density (>600 kg/m3), durability (>97.5%, see optimization of durability in [9]) and ash content (<5%). However, the S content was posing a problem as it was above the threshold of 0.05% for the highest quality class. In addition, the moisture was slightly above the threshold of 10%, which is no major challenge and can be adjusted either during pelletizing or through subsequent drying. The S content, however, is much like the ash content a parameter that needs to be controlled in the raw material as the torrefaction will not help reduce it. In general, a non-woody biomass pellet to fall into the thermally treated biomass standard not only needs to be thermally treated but should also have a minimum HHV of 18 MJ/kg DM for the highest quality class.

### *3.2. Hardness and Modefied Hardgrove Index of the Torrefied Biomass Pellets*

Beyond the standard characteristics, which are prescribed in the respective ISO standards, the quality of the pellets may also be described by other relevant properties e.g., resistance to fracture or grindability. In this study, hardness was chosen as a parameter that may bridge the gap between the information obtained from the mechanical durability test (ISO 17831-1) to the actual grindability using a laboratory mill. In addition, the hardness test may also help understand the unintended grinding which happens when pellets are being fed, e.g., using a screw feeder. This, however, was not focus of this study.

The average hardness of the different pellets types was very different with 96 N for forest residue pellets and 496 N for spruce pellets (Table 3). This variation is partly reflected by the differences in torrefaction conditions with forest residue being exposed to the highest and spruce being exposed to the lowest TFtemp. This may be explained by the removal of hydrogen bonding sites, depolymerisation and the destruction of the fibrous structure (less interlocking bonds) with increased degree of torrefaction. The high standard deviation, which is observed when recording the pellet hardness across 20 pellets, as proposed here, may reflect the inhomogeneity of the torrefied material leading to different structural bonds in the pellets. In order to produce torrefied pellets with high durability and a respective hardness, application of steam and water prior to densification may be crucial to help plasticize the material and reduce the softening temperature of the remaining lignin [9].

In literature only one study could be found reporting the hardness of torrefied biomass pellets measured via the Kahl Hardness Tester [30], while Li et al. [33] compared the Meyer hardness of torrefied wood pellets which in addition to the force also includes the indentation depth and the surface area the force is applied to. Li et al. [33] found that the higher the severity of the torrefaction the lower the hardness of the pellets was. This is in line with the findings of this study.


**Table 3.** Mean values of the modified Hardgrove index and hardness of the torrefied biomass pellets as well as further statistical information on how the mean values were obtained.

Regarding a comparison with standardized Hardgrove coals by applying the modified HGI method, forest residue and straw had the highest value (HGI = 55) which reflects a low energy input, better grindability and potentially a higher throughput while spruce and beech had the lowest HGI values (HGI = 25 and 21 respectively) being not so favourable for grinding. Similar to the method used here with 50 cm<sup>3</sup> samples, Shang et al. [34] investigated the HGI of torrefied wheat straw at different TFtemp with a residence time much higher (2 h) than any of the TFtime used here. The HGI of the pellets made from straw torrefied at 270 ◦C for 30 min analysed here (HGI = 55) is therefore slightly below the value observed by Shang et al. [34] for straw torrefied at 250 ◦C for 2 h (HGI ≈ 60). Bridgeman et al. [16] measured the modified HGI of torrefied willow and found values between 24 and 51 mainly depending on the difference in TFtime (10 vs. 60 min) while the TFtemp was in both cases at 290 ◦C. The pellets from torrefied willow analysed here had a HGI of 34 which is between the two HGIs measured by Bridgeman et al. [16] probably due to a higher TFtemp (308 ◦C) at similar TFtime (9 min). Interestingly, the results obtained by Bridgeman et al. [16] and Shang et al. [34] compare well to the results obtained here even though the torrefaction in these two studies was done in lab-scale reactors as compared to the pilot-scale torrefaction which might affect homogeneity of the product.

No evident correlation could be observed between mHGI and the hardness. This is to be expected as the hardness test gives an information on the resistance of the pellet to fracture while the mHGI is measured on the pre-ground pellet (<1.18 mm) and therefore reflects the characteristics of those particles. The hardness test is more similar to the durability test which is supported by a correlation between the two parameters of R<sup>2</sup> = 0.72. Durability and HGI values are also not correlated.

### *3.3. Performance of the Mills and Specific Grinding Energy*

The range in time required to grind 2.5 kg of pellets using a 1 mm sieve varied between the different mills (Figure 1a). The smallest variation was observed for pellets ground with the IM (2.3–2.7 min) while for the CM the range was between 2.1 and 4.4 min. The type of pellet requiring the longest (tmax) or shortest time (tmin) to be ground was never the same between the mills (IM: tmax = poplar, tmin = beech; CM: tmax = straw, tmin = willow; HM: tmax = forest residue, tmin = pine) highlighting that different grinding forces are at work. For IM and HM, the grinding time was independent of mHGI, hardness or durability of the pellets (all R<sup>2</sup> < 0.01). However, the grinding time in the CM was showing signs of dependency on hardness (R<sup>2</sup> = 0.50) and a strong correlation (R<sup>2</sup> = 0.79) with durability. The fact that IM takes the least time for crushing the pellets below 1 mm is largely due to its capacity. As by construction the impact mill is designed for the highest throughput and operating at higher rotor speed than the other two mills. Given the procedure chosen in this study (2.5 kg input material) the IM was not operating at full capacity.

**Figure 1.** Boxplot for each mill type showing: (**a**) The pellet specific grinding time (tG); (**b**) The pellets specific throughput. Boxplot shows min and max values (error bars) as well as the 1st and 3rd quartile and the median.

Regarding the throughput of the mills (Figure 1b), which by definition is dependent on the grinding time, smallest variation was observed for IM and HM while CM had a wide range (34.1–71.3 kg/h). Since low grinding time leads to a high throughput, the pattern of pellet specific throughput per mill is the reverse pattern of pellet specific grinding time mentioned above.

How the different torrefied biomass pellets resulted in different SGEs can be seen in Figure 2. The mean SGE observed for CM and HM were about the same at 5.15–5.24 Wh/kg while the mean SGE of the IM was substantially higher with 8.23 Wh/kg. Again, this reflects the similar technical specifications of both CM and HM in terms of throughput and rotor speed, which is much different from the IM. More interesting, however, is the way the different mills performed with the respective pellets. The commonality between all three mills was that forest residue pellets required the least and straw the most energy for grinding. In addition, all three mills were able to grind forest residue, willow and pine well below the average SGE consumption while in addition to straw poplar and spruce and beech (except for the CM) required above average SGE. This pattern is in line with several physical-mechanical characteristics of the pellets. Forest residue, willow and pine have much lower hardness (96–125 N) and durability (87.7%–90.3%) and higher TFTEMP (300–308 ◦C) than the other pellet types (hardness: 206–496 N; durability: 94.5%–97.9%; TFTEMP: 260–280 ◦C).

**Figure 2.** Specific grinding energy obtained by applying the evolved grindability method to the respective torrefied biomass pellets using three different mill types: impact mill (IM), cutting mill (CM) and hammer mill (HM).

Current literature does not allow a good classification of the results obtained here as in other studies other parameters were the centre of focus when reporting grinding energy. However, a closer look at other studies recording grinding energy may still give further insights into the results obtained

here. Phanphanich and Mani [27], for example, showed a dramatic decrease in grinding energy between untreated wood chips (237 Wh/kg; at water content of 7%) and torrefied wood chips (as low as 23 Wh/kg) using a laboratory cutting mill (1.5 kW, mesh size 1.5 mm) and found a strong linear correlation with torrefaction temperature. Temmerman et al. found similar values for grinding energy of wood chips (162 Wh/kg, at water content of 5%–10%) as Phanphanich and Mani [27] using a laboratory hammer mill (1.1 kW, mesh size 2 mm), while regular wood pellets consumed energies much closer to the ones measured here for torrefied wood pellets (on average 11.4 Wh/kg; water content of 5%–10%). This may be explained by the fact that grinding pellets often is a disintegration of the pellets back into the particles they were made of. Hence, the difference in grinding energy between pellets from untreated material and torrefied material will be much smaller than between, e.g., wood chips and pellets. However, whether the grindability of untreated and torrefied material or non-torrefied and torrefied pellets are compared, increasing the degree of torrefaction should result in lower SGE. Shang et al. [35] reported an exponential decrease of the grinding energy for pine pellets exposed to increasingly higher torrefaction temperatures when using a coffee grinder as laboratory mill (0.55 kW; similar grinding forces as the IM). However, a comparison between the grindability of torrefied pellets and pellets from torrefied material may only partly apply (see discussion on bonding mechanisms during densification in Section 3.2).

The fact that pellets from torrefied material require a much lower grinding energy is one of the economic advantages of torrefaction as it will save energy during the final grinding for injecting the biomass powder into the boiler. However, the possible difference that may occur between different laboratory mill types, as shown here, may only partly be relevant for power plant operators since existing technology may not be replaced for new technology due to high investment costs. However, the results shown here could help identify the most appropriate mill type for a standardized methodology to assess grindability of regular and thermally-treated biomass pellets. The SGEs observed in the IM, for example, are much more distinct for the different pellet types than the values observed with the HM where poplar, spruce, beech and straw only differ minimally.

### *3.4. Particle Size Distributions after Grinding*

In addition to the SGE, the cumulative PSD curves (Figure 3) were compared to assess the impact of the mills on the material and to check whether a particular mill type might prove more suitable for use in a standardized methodology in relation to, e.g., co-combustion requirements.

**Figure 3.** Cumulative particle size distribution obtained when using: (**a**) Impact mill; (**b**) Cutting mill; and (**c**) Hammer mill on each of the torrefied biomass pellets.

As can be seen in Figure 3, each mill yields a uniquely shaped distribution curve, which is similar for all materials ground with that mill. In the current literature, a comparison of PSD curves between different lab-size mills and their performance with torrefied pellets is missing. Some studies, however, may be helpful to further understand the results shown here. Phanphanich and Mani [27], for example, produced cumulative PSDs of two materials (pine chips and logging residue) with different degrees of

torrefaction which were ground in a lab-size cutting mill. The resultant PSD curve showed a linear increase—much like the one in this study for the untreated materials (Figure 3b)—where it can be seen that as the severity of the torrefaction increased the incline of the curve was more pronounced at <0.25 mm before it started declining (at >0.25 mm). However, it is difficult comparing the grindability of wood chips with the grindability of pellets. Thrän et al. [9] showed that pellets from torrefied spruce were mostly disintegrated back to the particle sizes before densification. The disintegration of non-densified material is of course very different and may depend even more on the mill type. However, when looking at the three different mill types and their respective PSD curves it becomes clear that the pellets were not just disintegrated but fractioned differently. Especially the PSD curves of the IM (Figure 3a) had a unique characteristic with a slow incline up until 0.2 mm followed by a strong incline between 0.2 and 0.5 mm and phasing out thereafter. For both CM and HM the incline is steady up until 0.7 mm before the curves phase out (with the exception of forest residue in the HM). Bridgman et al. [16] showed cumulative PSD curves for two energy crops (willow and miscanthus) which where torrefied and ground in a roller-ball mill designed for the HGI test and they had very different characteristics than the curves of the three mills used here. Interestingly, in their study, all degrees of torrefaction of miscanthus resulted in curves similar to the ones from the HGI coals while the raw material yielded an entirely different curve. However, in the case of willow, only the PSD curve of the most severely torrefied sample showed the same pattern as all the standard coal curves. Despite the differences in mills, the PSD curves obtained by their study do not compare well with the result obtained here as they show the grindability of pre-ground material (as required by the HGI procedure) while here whole pellets were used. Only during the investigation of the mHGI of the seven pellet (when using a ball mill after pre-grinding the pellets) their findings (PSD curve of torrefied willow) could be confirmed (see Figure S4).

Regarding the grinding performance and suitability of the PSD for co-combustion application Thrän et al. [9] investigated the co-milling of torrefied pellets with coal in a bowl mill (400 kg/h at medium load, vertical rollers) at different mixture rates. When the share of torrefied pellets in the mixture was high (58.5%) 90% of the ground material had a particle size below 0.6 mm while at a lower share of torrefied pellets (32%) the main particle size fraction dropped below 0.2 mm. In comparison to the mills used here, only the IM can achieve similar results with most of the seven feedstocks resulting in 90% of the ground mass to be below 0.6 mm.

To assess the suitability of the mills to test grindability in a standardized way for an application in co-combustion process the distribution widths of the PSD curves were compared. The average distribution width (Figure 4) across all input materials was narrowest for the IM (0.41 mm), followed by the HM (0.51 mm) and the CM (0.62 mm). However, the IM had the largest variation in feedstock-based distribution width ranging from 0.33 mm all the way to 0.52 mm. Both HM (0.41–0.56 mm) and CM (0.57–0.69 mm) had more even distribution widths across the different feedstock. Looking back at the distribution curves again (Figure 3), it becomes obvious that the variation in the curves from the IM happened in the particle range of >0.3 mm, while the variation observed in the other two mills was more or less consistent along the particle size scale. The distribution curves of HM, on the other hand, had one strong outlier (forest residue) compared to the other mills. However, in all mills the distribution width was below average when grinding forest residue material, which is also in line with the low SGE required to grind this type of pellet in all mills. Along the same line, straw resulted in the largest distribution width in all mills (curve for straw in HM is missing from the data) as suggested by the SGE. However, a modest linear correlation of the SGEs and the respective distribution width could be observed for the CM (R<sup>2</sup> = 0.65) and the IM (R<sup>2</sup> = 0.47) while the HM had only weak correlations (R<sup>2</sup> = 0.27). Thus, for CM and IM it can be concluded that pellets from torrefied material, which generally require lower SGEs, also will be ground with narrower distribution widths. The fact that the IM not only has the narrowest distribution width but also the lowest values for x90 (0.53 mm, average across all seven pellet types; see Table S1) makes it a more ideal lab-mill for producing particles favourable in co-combustion settings.

**Table 4.** Correlation

 coefficient (R2) and equation

**Figure 4.** Width of each particle size distribution obtained from the seven torrefied biomass pellets when milled with the respective mill type: impact mill (IM), cutting mill (CM) and hammer mill (HM). The mean value across all distribution width values of each mill type is indicated.

### *3.5. Specific Grinding Energy versus Other Methods*

To assess whether the SGE is a parameter that can stand alone or is better complemented with another parameter the SGE values were correlated to the three relevant grindability parameters analysed here. The SGE was found to be in a positive linear relation with the durability of the pellets and the pellet hardness (Table 4). However, only the correlation with the durability is of strong enough significance (R<sup>2</sup> = 0.86 for both IM and HM) that one could argue that SGE is partly reproducing that information. All correlations of SGE with the modified HGI were weaker (R<sup>2</sup> ≤ 0.77) than the correlation between durability and SGE. The correlations between mHGI and SGE were all negative, as a lower value mHGI will imply higher energy requirement for grinding. The correlations of SGE with hardness were all of insignificant strength.


for predicting specific grinding energy (SGE) with


Looking at the close correlation between SGE and durability it might be advisable to combine SGE values with a PSD parameter such as the distribution width to help detangle it from durability and give a unique insight into the co-combustion characteristics of the torrefied biomass pellets.

### *3.6. Adjustments of the Grindablity Method for Reliable Performance*

For a standardized method, it is essential for the equipment used to perform reliably and a procedure that ensures high repeatability and reproducibility of the results obtained. For the proposed GM and the values it will yield, it is necessary to not only define the type of mill but also to adjust some parameters, e.g., the feeding rate. This will require a round robin test of the methodology across different laboratories all using the same equipment.

In terms of choosing the right mill technology for a standardized method, the transferability of the results to plant operator scenarios is the main focus. As plant operators will try to use their existing equipment (in most cases ball, roller or impact mills) and adapt it to the new biomass, a CM seems to be an unlikely choice. Another disadvantageous feature of a CM, also in terms of repeatability and reproducibility of laboratory results, are the knives, which tend to become dull and thereby bias the results (e.g., the sharpness of the knives used here could not be quantified). A HM on the other hand

may pose the challenge of proper adjustment of the hammers so that results will be repeatable and reproducible, which, however, is less of a challenge than what the CM poses. IM seem to be the easiest to handle and adjust and yielded the most promising PSD curves and distinct SGE values amongs<sup>t</sup> the three mills. Results from a mill like Thrän et al. [9] used in there study may be even more valuable to plant operators ye<sup>t</sup> it is not applicable for a lab standard given its capacity (double that of the IM) and respective investment cost. As the IM used here is still relatively large for lab application, it might be advisable to also do an investigation on the performance of a small-scale disc-grinding system as was used by Shang et al [35].
