*2.2. Thermal Treatment*

Figure 2 shows a schematic diagram of the thermal treatment experimental setup. The cherry wood samples were successively treated at 220, 260, 300, 350, 450 and 550 ◦C for 0.5 h for each temperature under flowing nitrogen in a quartz tube furnace. The samples were placed in a holder (Figure 2) and loaded into the middle of the tube at a fixed location and continuously purged with N2 at a 50 mL/min flow rate for 0.5 h at room temperature to remove oxygen. Once the furnace reached the target temperature measured with a thermocouple, the quartz tube with the samples was placed in the furnace. When the samples reached the selected temperature, which was measured by a thermocouple placed close to the sample holder, they were held for 0.5 h. The treatment time was the same for all treatments since the treatment temperature is the more critical variable for torrefaction [15]. It was selected based on previously reported woody biomass torrefaction [22] and low temperature pyrolysis studies [12]. Torrefaction time is normally less than 1 h because the biomass thermal decomposition rate is initially fast, slowing after about 1 h [15]. A half hour was selected as the optimal time for torrefaction based on the improvement of grindability, the mass and energy yields [22]. A half hour treatment time was also used in a study of forest biomass grindability and fuel characteristics at torrefaction temperatures from 225 to 300 ◦C [20]. After the 0.5 h treatment, the tube was removed from the furnace and quickly cooled to room temperature under flowing N2. The treated samples were stored in a desiccator.

**Figure 2.** Schematic diagram of the thermal treatment experimental setup.

### *2.3. Characterization and Analysis of the Chars Generated from Thermal Reatments*

After each treatment, the samples were characterized. Untreated wood (1, untreated 8–11) was used as a reference. For the twelve samples (2–7), one sample was removed after each temperature treatment in the sequence; therefore, each sample at a given temperature was cumulatively exposed to all the thermal treatments at lower temperatures. Two samples (8–9) were used to characterize physical property changes. The remaining two samples (10–11) were used for study of microstructural changes. These four samples (8–11) were taken through all the temperature treatments and examined after each step.

### 2.3.1. Cell Wall Component Decomposition Study of Untreated and Treated Woods by TGA and Differential Scanning Calorimetry (DSC)

For cell wall compositional changes, samples of approximately 10 mg were tested in the TGA (Perkin Elmer Pyris 1 TGA, PerkinElmer Inc., Shelton, CT, USA) using a non-isothermal method. Before the TGA test, the untreated and treated wood samples (1–7) were manually ground with a double-cut flat file and sieved through a 60 mesh screen (less than 250 μm). Samples were heated at a 2 ◦C/min from 25 to 800 ◦C under nitrogen with a total flow rate of approximately 125 mL/min. A slow heating rate was selected to clearly identify model components that thermally decompose over different temperature ranges. This rate was much lower than others reported (20 ◦C/min) [28,34].

Model compounds representing typical cell wall compositions were also analyzed by TGA to aid in identifying cell wall component changes in the treated wood samples. Cellulose (fibrous, long and medium), xylan from birch and beech (used as a model for hemicelluloses), and lignin (alkali) were selected as the model components. Both birch and beech are hardwoods and the use of xylans was to better represent hemicellulose because hemicellulose may be different between different types of biomass. All model compounds were purchased from Sigma-Aldrich (St. Louis, MO, USA).

Percent weight for each cell wall chemical component was calculated by weight loss in designated temperature ranges that correlate to the cell wall components from TGA analysis [34]. The designated temperature ranges were selected based on differential thermogravimetry (DTG) curves vs. temperature of the model cell wall components. Weight loss in each temperature range of the component was calculated by area in the range of DTG curve for the samples because the sample was heated at a slow constant rate and temperatures were linear with time. In this study, the area was adjusted based on the distribution of the model lignin in the designated temperature ranges because the lignin slowly decomposes over all temperature ranges (results shown in Section 3.1). Percent weight (dry basis) for the component was calculated by the weight loss in the temperature range of the component over the total weight loss in all designated temperature ranges (200–800 ◦C). The value of this method has been demonstrated with other biomass materials such as untreated and thermally treated switchgrass and cherry bark in our laboratory.

For differential scanning calorimetry (DSC) (Perkin Elmer Pyris Diamond DSC, PerkinElmer Inc., Shelton, CT, USA), the samples were tested at the same conditions as was used for TGA but over a temperature range of 25 to 600 ◦C. The DSC tests were used to further understand the thermal behaviors of biomass torrefaction and low temperature pyrolysis. All TGA and DSC experiments were performed in duplicate.

### 2.3.2. Physical Properties (Color, Weight, Dimensions, Bulk Density and Grindbility) of Untreated and Treated Woods

For physical changes, sample color was recorded using a digital camera and the dimensions measured using a digital caliper. After each treatment, the sample was weighed and weight loss percentage was calculated by the weight of treated sample over the untreated sample (mass yield = 100 − weight loss percentage). Dimensions of the samples are presented as height along the axial (longitudinal) direction, width along the radial direction and depth along the tangential direction (Figure 3) [39]. Bulk densities of the samples were calculated based on the measured weights and dimensions, however, due to the variations in sample shapes, the calculated bulk densities are approximate. The results of the bulk density measurement were used to estimate heating value in a volumetric basis. The physical property tests were performed in duplicate using samples (7–8). One sample was marked with a small "X" using a scalpel to differentiate it from the other.

Grindability testing of untreated and treated particles (1–7) was conducted using a hammer mill (Kinematica AG Polymix PX-MFC 90D, Luzernerstrasse, Switzerland) with a 0.8 mm bottom sieve at a rotational speed of about 1500 rpm. After milling, the samples were sieved using 500, 260 and 106 μm screen to four size fractions: >500 μm, 500–212 μm, 212–106 μm and <106 μm. Each fraction was weighted and its weight percentage was calculated over total sample weight to obtain particle size distribution for the sample. The fine particle size weight percentages of treated samples were compared with the untreated sample to evaluate the grindability changes resulting from treatment. The grindability test was performed one time for each treatment. However, multiple preliminary tests were conducted to determine the best grindability test method.

**Figure 3.** (**a**) Illustration of a section of wood showing the three planes discussed [39]; (**b**) Schematic of a wood sample used for these tests with the dimensions and orientations indicated.

### 2.3.3. Proximate Analysis and High Heating Value of Untreated and Treated Woods

For chemical property changes, proximate analysis (moisture content (MC), volatile matter (VM), fixed carbon (FC) and ash) was conducted by TGA (TA DSC-TGA Q600). The ASTM D-3172 (ASTM International, West Conshohocken, PA, USA) procedure for coal and coke was used. The samples used for the proximate analysis were from the grindablity tests. Approximately, 10 mg of ground sample with a size of <212 μm was placed in an alumina crucible for testing. MC and VM were tested at 105 and 950 ◦C, respectively, in Ar. Ashing was conducted at 750 ◦C in 22% O2 with Ar balance. FC was calculated on a dry basis by subtracting ash and VM percentages. The proximate analyses were performed in duplicate.

High heating value (HHV) was calculated using proximate analyses and cell wall composition (structure analysis) data. This is simple method for the estimation of caloric value and is acceptable for engineering estimations of net heating value of a biomass fuel [40]. In this paper, the correlation proposed based on structural analysis (lignin content) (Equation (1)) [41] and the proximate analysis (Equation (2)) [42] were selected. Those correlations developed based on raw biomass are also applicable to the estimation of the HHV of torrefied biomass [15].

For wood,

$$\text{HHV} = 0.0893 \text{ (L)} + 16.9742 \tag{1}$$

where HHV (MJ/kg) is the energy content on a dry-ash-free and extractive-free basis. L (wt %) is lignin content of biomass on a dry basis.

$$\text{HHV} = 0.3536 \text{FC} + 0.1559 \text{VM} - 0.0078 \text{Ash} \tag{2}$$

where HHV (MJ/kg) is the energy content on a dry basis. FC (wt %), VM (wt %) and ash (wt %) are fixed carbon, volatile matter and ash on a dry basis, respectively.

### 2.3.4. Microstructual Transformations Study Using SEM

For the SEM analyses (FEI Company (Thermo Fisher Scientific, Waltham, Massachusetts, USA) Quanta 600 field emission scanning electron microscope, low vacuum mode, secondary electron detector), three plane sections on the wood sample (tangential, radial, and cross or transverse) and twelve analysis spots for each section were examined using six magnifications at each spot resulting in a total of 216 images. The tangential plane section of the wood sample (10) (with the rough edge, indicated by the arrow in Figure 1) was prepared by manually breaking it to preserve its microstructure. The sample (11) was used for the microstructure study of the radial and transverse plane sections of the wood sample (indicated by the arrow in Figure 1). To allow the same analysis sites to be followed through all the treatment steps, small reference marks were made on the samples with a scalpel. Low magnification images were collected and a reference map was constructed allowing the reference marks and obvious features to be utilized for the relocation of the analysis sites following each treatment step. Following the changes observed at the same analysis spots through the thermal treatment steps allows correlation of the observed changes with other measured characteristics. These results contribute to the understanding of the property changes of the biomass during treatment.
