Characteristics of High-Temperature Torrefied Wood Pellets for Use in a Blast Furnace Injection System

Abstract

As the iron and steel industry needs to cut its CO₂ emissions drastically, much effort has been put into establishing new—less greenhouse-gas-intensive—production lines fueled by hydrogen and electricity. Blast furnaces, as a central element of hot iron production, are expected to lose importance, at least in European production strategies. Yet, blast furnaces could play a significant role in the transitional phase, as they allow for the implementation of another CO₂-reducing fuel, carbonized wood reducing agents, as a substitute for coal in auxiliary injection systems, which are currently widely used. Wood carbonization yields vastly differing fuel types depending on the severity of the treatment process, mainly its peak temperature. The goal of this study is to define the lowest treatment temperature, i.e., torrefaction temperature, which results in a biogenic reducing agent readily employable in existing coal injection systems, focusing on their conveying properties. Samples of different treatment temperatures ranging from 285 to 340 °C were produced and compared to injection coal regarding their chemical and mechanical properties. The critical conveyability in a standard dense-phase pneumatic conveying system was demonstrated with a sample of pilot-scale high-temperature torrefaction.

1. Introduction

Currently, it is well recognized that reducing CO₂ emissions without negative effects on global development is one of the imminent challenges to science, industry, and nations. The iron and steel sector is understood to account for roughly 8% of global anthropogenic CO₂ emissions [1]. At the same time, the steel industry is a major force of social development [2], being predicted to grow by about 30% until 2050 [3]. Thus, a reduction in specific emissions of 23% would be needed simply to avoid an increase in emissions from this sector on a global scale. In the EU, the goal was set to decrease industry-related emissions by about 85% in the very same year [4]. For the iron and steel sector, this goal is unachievable without a radical transformation of the production routes themselves. In response to this, the first direct reduced iron (DRI) European plants with renewable hydrogen and electric arc furnaces (EAFs) have been projected and built to substitute the inherently CO₂-intensive blast furnace route [5,6]. To date, no reliable timeframe can be predicted within which European hydrogen and electricity production can be ramped up to meet the demands of European steel makers [7]. Large-scale imports of hydrogen [8] and even DRI [9], mainly as hot briquetted iron (HBI) [10], are expected.

While this envisaged production route is able to almost eliminate direct steel-related CO₂ emissions, and the blast furnace route can only be partially optimized, the latter is currently widely used. In 2020, roughly 60% of Europe’s primary steel was produced with blast furnaces [11]; if a fuel substitution proves technically feasible, this will allow instant CO₂ emissions reductions instead of the lengthy and costly construction of new facilities beforehand. In this respect, the presented approach of substituting injection coal with a pulverized torrefied biomass, i.e., bioreductant, is an often-discussed option [12,13]. Modern blast furnaces use coke as a primary fuel, which is fed in at the top and burned with a hot blast in front of the tuyeres in the lower-blast-furnace region. Additionally, auxiliary fuels are primarily injected into this region to produce reducing gas. Most commonly, pulverized coal injection (PCI) is used. Almost 50% of the world’s blast furnaces are believed to employ coal injection technology [14]. There are particular strengths to substituting injection coal with upgraded biomass. It is a short-to-mid-term transitional solution [15] with a possibility to reduce blast furnace CO₂ emissions by about 18–40% [16] without significantly changing up-to-date infrastructure, thereby enabling this fast implementation which is essential in adhering to the 1.5 °C goal [17]. Amidst the establishment of a hydrogen market and a general electrification of processes, the use of an alternative fuel can help overcome problems of acquisition and, also, relieve pressure on named markets.

The use of thermally treated wood—in this case, charcoal—has a long tradition in iron production, causing severe deforestation in many regions, until the introduction of coke as the main blast furnace fuel [18]. Today, the specific fuel requirement is far lower than the 2.6 tons of charcoal per ton of iron that was common at the time [19] due to the envisaged usage in injection systems with supposed average injection rates of 0.15 tons per ton of iron. Nevertheless, the overall wood demand would be undoubtedly high as annual European iron production has increased from 200 kt in 1750, based on rough estimates [20], to over 200 Mt in EU27 and CIS countries [21]. The high efficiency of wood carbonization, i.e., solid yields higher than 20 m-%_db, as in traditional kilns [22], is mandatory, as is the realization that the sustainable production of carbonized wood is minimal or has yet to be established. Wood production in short-rotation plantations might be a viable solution; however, it is unlikely to provide sufficient amounts of wood for the use of bioreductants to be more than a niche strategy. Thus, supply-sided volatilities are to be expected, and maintaining the ability to temporarily use conventional fuel—PCI coal—is important. Both aspects of fast implementation and resilience motivate tailoring the bioreductant substitute to be suitable for coal conveying systems.

The essential characteristic of this envisaged bioreductant is the ability to provide a flow of energy and reduction potential to an extent similar to coal, with the same conveying and injection equipment. This general requirement can be divided into specific attributes concerning chemical characteristics and particle and bulk characteristics, especially where the development of volatile content, particle sphericity, and fluidizability [23] with increased torrefaction temperatures is the main differentiator. The specific chemical composition, in terms of carbon content in relation to oxygen content, is of top priority for its suitability as a reducing agent. In addition, the volatile content affects the bioreductant’s burnout in the raceway [24,25] and the efficiency of its usage [26]. Forty mass percent is generally understood to be the upper limit in this regard [27]. Generally, the ash and sulfur content are of importance as well; however, biomass and charcoal are superior to coal in this respect. Only the higher alkaline content must be addressed in the operation of the blast furnace [25,28]. Equally important are the particle and bulk characteristics. PCI systems usually employ a dense phase conveying system, which requires the feedstock to be fluidized before feeding it to the pipe system. This fluidizability is fundamental and can be achieved or improved with a proper particle size distribution, high enough bulk density, and particle shape [29,30]. Not every torrefaction temperature, however, allows for the tailoring of all these aspects [31]. Fluidization and conveying are performed with a preferentially low conveying gas stream, as excess injection velocities impair the combustion in the raceway [32]. Therefore, it is crucial that the achieved mass proportion of the conveyed solid—the gas loading ratio—is high. These aspects of proper particle size distributions and particle sphericities eventually determine the proper conveyability and affect the achievable specific mass flow of bioreductant and, together with its chemical composition, the specific input of reduction potential via the PCI system.

The aim of this study is to determine the lowest torrefaction temperature that provides all necessary bioreductant characteristics compared to a PCI coal. From over 100 material tests, the results of four different lab-scale torrefaction samples, which were produced, comminuted, analyzed, and tested for their bulk characteristics in comparison with PCI coal, are discussed. The lowest-temperature sample, which exhibited the required characteristics, was reproduced at pilot scale for a conveying test to validate its conveyability and assess its suitability as a PCI substitute.

2. Materials and Methods

2.1. Sample Preparation

A sample of the actual PCI coal of the studied case, labeled “PC” and already comminuted in an industrial vertical roller mill, was acquired from voestalpine Stahl Donawitz. This pulverized coal is perceived as exhibiting all necessary properties regarding combustion performance and conveyability to be used in a dense-phase pneumatic conveying system for blast furnace injection.

For the production of the bioreductant samples, eight tons of poplar pellets were produced and analyzed: poplar wood chips, including natural bark, were acquired from the company pan forst (Donaudorf, Austria), milled to 4 mm sieve size, and dried at Cycleenergy Gaishorn (Gaishorn, Austria). Eventually, pellet production was performed by Holzforschung Austria (Stetten, Austria) on an Amandus Kahl flat die press mill (35 mm press channel length, 6 mm diameter) without additional binders.

For the laboratory tests, a small portion of the pellets was taken to perform the torrefaction tests and material characterization. Four different samples—MF285, MF310, MF325, and MF340, where each sample’s peak temperature (sample surface temperature) is signified by the name’s number—were produced in a lab-scale pyrolysis reactor in a muffle furnace (Metronik Forno MK-FM35AT) with roughly 1 kg of educt each, loosely packed in steel mesh baskets. The retort used for torrefaction was flushed with N₂ (5–10 times the retort’s empty volume per hour) to ensure an oxygen-depleted atmosphere until the samples were cooled below 100 °C. With a heating rate below 10 K/min, the samples were heated to the peak temperature, held at this temperature for twenty minutes, and finally cooled very slowly due to the high heat capacity of the furnace’s lining. To generate bioreductant samples with grain sizes that are suitable for usage in a PCI system, all carbonization products were, subsequently, pulverized with a rotor mill (Fritsch Pulverisette 14 classic line; 20k rpm and a sieve size of 0.2 mm) for further testing. The chosen milling setup and parameters were specified by a prior test campaign.

The pilot-scale-sized sample “NF” was torrefied in a semi-industrial-scale rotary drum kiln at NextFuel (Frohnleiten, Austria) [33] at approximately 330 °C and—in an attempt to resemble the laboratory milling equipment to the largest possible extent—comminuted with a bench-scale impact classifier mill at Neuman & Esser Process Technology (Übach-Palenberg, Germany). The laboratory mill used for these tests was an impact mill type with a horizontal rotor orientation, central feed, and material outlet through a sieve ring around the perimeter. In contrast to this, the bench-scale mill is fed from the perimeter and only the rejects of the classifier are fed from the center. This choice should ensure that the milling is performed with impact forces, no imposed particle orientation towards these forces, and little particle residence time within the mill. The integrated classifier, not included in the lab mill, is typical for PCI mills [25].

2.2. Chemical Analysis

Proximate analysis was performed with all fuel types. The water content was measured according to ISO 18134-3 [34] by placing two samples per fuel type in a ventilated drying oven overnight and analyzing the weight difference of the samples after drying. For the measurement of the ash content, the already dried samples were heated to 550 °C and weighed again according to ISO 18122 [35]. The volatile matter was measured according to ISO 18123 [36]. The fixed carbon w_cfix,db was calculated by difference according to Equation (1). The parameter fuel ratio is often used to evaluate the expected burnout of the reducing agent in the raceway [37,38], and it can be calculated with the results of the proximate analysis.

$${\mathrm{w}}_{\mathrm{c}\mathrm{f}\mathrm{i}\mathrm{x},\mathrm{d}\mathrm{b}}=1-{\mathrm{w}}_{\mathrm{a}\mathrm{s}\mathrm{h},\mathrm{d}\mathrm{b}}-{\mathrm{w}}_{\mathrm{v}\mathrm{o}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{l}\mathrm{e}\mathrm{s},\mathrm{d}\mathrm{b}}$$

(1)

The chemical main components (C, H, N) of all fuel types were analyzed with a vario MACRO cube (Elementar Analysensysteme GmbH, Langenselbold, Germany). Oxygen contents were calculated by difference according to Equation (2).

$${\mathrm{w}}_{\mathrm{O},\mathrm{d}\mathrm{b}}=1-{\mathrm{w}}_{\mathrm{a}\mathrm{s}\mathrm{h},\mathrm{d}\mathrm{b}}-{\mathrm{w}}_{\mathrm{C},\mathrm{d}\mathrm{b}}-{\mathrm{w}}_{\mathrm{H},\mathrm{d}\mathrm{b}}-{\mathrm{w}}_{\mathrm{N},\mathrm{d}\mathrm{b}}$$

(2)

Alternative reducing agents for the blast furnace process can be assessed in terms of their reduction potential (ReP) [39], which can be used for calculating the mass ratio of coke that can be replaced by the injected reductant. According to Equation (3), the ReP is calculated by the weight proportions of carbon, hydrogen, and oxygen (with w_i being the weight fraction of component i), either on a db (dry basis) or daf (dry and ash-free) basis. Whereas the ReP calculation is based on a stoichiometric calculation of the blast furnace’s reduction process, coke replacement ratios (CRRs) are commonly estimated by the reductants’ chemical composition based on empirical data. All samples were evaluated by CRR formulae found in [25,40,41]. Empirical correlations might have the advantage of overcoming the strict limitation to stoichiometry purported by ReP but often refer to experiences with coal only. To avoid too narrow a focus, both approaches are discussed.

$$\mathrm{R}\mathrm{e}\mathrm{P}=250·{\mathrm{w}}_{\mathrm{C}}+500·{\mathrm{w}}_{\mathrm{H}}-125·{\mathrm{w}}_{\mathrm{O}}$$

(3)

All portions torrefied in the muffle furnace were weighed prior to and after torrefaction to measure the solid yield, and together with the mentioned analyses, the specific yields according to Equations (4) and (5) were calculated.

$${\mathrm{y}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{d}\mathrm{b}}={\displaystyle \frac{{\mathrm{m}}_{\mathrm{o}\mathrm{u}\mathrm{t}}}{{\mathrm{m}}_{\mathrm{i}\mathrm{n}}·(1-{\mathrm{w}}_{\mathrm{m}\mathrm{o}\mathrm{i}\mathrm{s}\mathrm{t}\mathrm{u}\mathrm{r}\mathrm{e},\mathrm{p}\mathrm{e}\mathrm{l}\mathrm{l}\mathrm{e}\mathrm{t}\mathrm{s}})}}$$

(4)

$${\mathrm{y}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{d}\mathrm{a}\mathrm{f}}={\mathrm{y}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{d}\mathrm{b}}·{\displaystyle \frac{1-{\mathrm{w}}_{\mathrm{a}\mathrm{s}\mathrm{h},\mathrm{s}\mathrm{a}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{e}}}{1-{\mathrm{w}}_{\mathrm{a}\mathrm{s}\mathrm{h},\mathrm{p}\mathrm{e}\mathrm{l}\mathrm{l}\mathrm{e}\mathrm{t}\mathrm{s}}}}$$

(5)

In [42], a mathematical correlation between ReP_daf and the solid yield_daf was found, which is noted in Equation (6). As both product parameters are mainly governed by carbonization intensity, a mutual dependency was inferred as a direct correlation. The torrefaction products considered were compared to this correlation in order to assess their validity.

$${\mathrm{R}\mathrm{e}\mathrm{P}}_{\mathrm{d}\mathrm{a}\mathrm{f}}=267.76·{\mathrm{e}}^{-1.0913·{\mathrm{y}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{d}\mathrm{a}\mathrm{f}}}$$

(6)

2.3. Particle Analysis

Particle size and sphericity distributions were measured using QICPIC/L02 equipped with the dry dispersion system RODOS/L-Q (Sympatec). For the particle size, the QICPIC parameter EQPC was used. This parameter is based on the measurement of a particle’s projection area and expresses the diameter of a spherical particle of equal projection area. The sphericity(s), being an indicator of the particle shape, relates the circumference of the projection area of this calculated spherical particle with the circumference of the real particle. The values are, thus, between 0 and 1; the more spherical the particles are, the closer this value is to 1.

Additionally, all samples were tested for bulk density and angle of repose in the style of the respective guidelines. In the style of ISO 60 [43], the bulk density was measured by means of stirring the specimen through a funnel into a cylindric vessel. The vessel had an inner diameter of 59 mm and an inner height of 36.4 mm, which amounted to a capacity of roughly 100 mL. The funnel was placed above the vessel so that its 11 mm outlet was 60 mm above the vessel’s upper edge. Before weighing the specimen, all excess powder was removed by scraping it off with a straight metal blade without densifying the specimen. This procedure was repeated three times per sample and the average density was rounded to 10 kg·m⁻³. The angle of repose was measured in the style of ISO 4324 [44] on a Dr. Pfrengle type of apparatus. The mentioned vessel was turned upside down and served as an elevated circular base with a radius of 33 mm on which the specimen was stirred again through a funnel. The funnel outlet’s distance to the base was set to 75 mm. The specimen was stirred through the funnel until a proper powder cone fully covering the base formed. By measuring the distance between the funnel’s outlet to the tip of the cone, its height was determined, and, consequently, the outer angle of the powder cone could be calculated by the arctan of cone height to the base radius.

The NF and PC samples were tested in a ring shear cell (Schulze Ringschergerät RST Mk II) to compare their flow characteristics in a non-fluidized state. Both tests were performed with a shear cell with a surface of 94 cm², with 3500 Pa as the normal load for pre-shear and 200, 900, 1600, and 3000 Pa for the shear points.

Lastly, to estimate the grindability of the semi-industrial bioreductant NF, an equally produced sample was tested according to ISO/TS 21596 [45] by DMT GmbH (Essen, Germany) for its “thermally treated biomass grindability index” (TTBGI), which is understood to be correlated with the Hardgrove Index (HGI) and, thus, provides equal grindability values but is evaluated with different sieve sizes which are thought to better suit torrefied biomass.

2.4. Fluidization and Conveying Properties

A PCI dense-phase conveying system essentially consists of fuel fluidization in a pressurized vessel, generating a mixture of solids and part of the conveying gas prior to feeding the fuel into the conveying pipe system, which conveys the fuel to the blast furnace injection. Therefore, fluidizability is an indispensable prerequisite of each bioreductant; in general, the fluidization behavior is the first aspect to be tested with an unknown material considered for dense-phase pneumatic conveying [30]. A series of measurements were performed by means of a custom-built fluidization device [46], which allows fluidization tests to be performed as described in the literature (e.g., [30] pp. 95–97). This device consists of a vertical acrylic glass tube with an inner diameter of 110 mm which is equipped with a permeable steel disk (GKN sinter metals, 1.440 steel, 10 mm thickness, type: -R 5 AX) at the bottom, a closable sample outlet at the center of this disk with an inner diameter of 9.5 mm, and a measurement of differential pressure from the upper edge of the disk to the top of the tube (one value per second). A sketch of the device is shown in Figure 1. By placing specimens on the permeable disk and applying a carefully controlled gas stream through the disk and, thus, also through the specimen, the force exerted by the air on the sample and vice versa in dependence on the superficial air velocity was determined by measuring the pressure difference above and below the specimen. Respective signals were recorded with 1 Hz frequency. Equally, by visually observing the change in bed height, the corresponding expansion of the powder bed was evaluated. A measurement was taken by incrementally increasing the gas velocity in small steps, with each step waiting for a steady pressure drop across the bed after initial and sometimes spontaneous fluctuations and denoting the steady-state bed height, until full fluidization was established. Then, the gas velocity was lowered stepwise to zero again. The superficial gas velocity, where full fluidization was achieved, was noted as the sample’s minimum fluidization velocity (U_m,f) [30]. The very limited amount of time the bioreductant has to combust in the raceway [47] requires it to be comminuted thoroughly; in fact, the Sauter mean diameter (SMD) of all samples was very close to the Geldart A/C border differentiating between good fluidization behavior and excess cohesiveness. Fluidization tests are difficult to interpret for this kind of fuel dust. While interpretations of a so-called pseudo-U_m,f, where an intercept of the linearized pressure signals of the transitional phase and of the fluidized state is inferred, would be a possibility [48], another strategy is presented in this paper. Comparisons based on a pseudo-U_m,f are potentially misleading in our respect as the main interests are—first—the fluidizability below a certain superficial gas velocity, 2 cm/s here, and—secondly—the general reliability of achieving fluidization. Apart from standard fluidization analysis, additional information from the pressure drop fluctuations was included. Certain presumptions were made:

(1)

While both the state at rest and the fluidized state are characterized by homogeneity regarding the bed porosity and resistance to a gas stream, the phase until U_m,f is reached is considered inhomogeneous and transitionary.

(2)

Gas velocities approaching U_m,f should result in increasing homogeneity; otherwise, the risk of randomly encountering troubles in fluidization cannot be ruled out.

(3)

Lower pressure-drop fluctuations signify a higher degree of homogeneity within the powder and gas mixture.

(4)

The fluidization graph should either signify the pressure drop value encountered most often per gas velocity or averaged values; however, a certain level of congruence should be observable. Both procedures are influenced by the operator, but the former is less influenced by spontaneous semi-fluidization.

(5)

During fluidization, additional adhesive forces must be overcome, which are not relevant in de-fluidization [48]. Thus, the pressure drop signal at decreasing gas velocities from U_m,f should not be higher than at increasing gas velocities. Contrary observations would signify a faulty fluidization behavior—such as blow-throughs—up to U_m,f.

U_m,f was understood to be reached in the following instances:

• The pressure drop stopped increasing with increases in superficial gas velocity;
• The bed height stopped increasing with increases in superficial gas velocity;
• The pressure drop did not fluctuate anymore according to (1) and (3).

Reliability in fluidization was defined by comparing the spread of the pressure drop signal from one gas velocity to the next: samples leading to decreasing spread when drawing near to U_m,f (according to (2)) exhibited a very good repeatability in fluidization results. The comparison of pressure signals from increasing and decreasing gas velocities (according to (5)) and m_fluidized (calculated as the ratio of the measured pressure difference at U_m,f to the counteracting area-specific weight force of the powder) were used to evaluate the correctness of the postulated U_m,f.

The ratio of the bed height (rel_rho) at U_m,f to the initial bed height was also denoted. Each test run comprised a first sequence of fluidization and a subsequent stepwise decrease in gas velocity, immediately followed by a repeated fluidization and deaeration without exchanging the material. During this deaerating stage, the gas retention capacity (grc) was measured by suddenly stopping the gas flow and measuring the time until the powder fully de-aerated. Finally, a third fluidization step was performed and the central sample discharge plug was removed at U_m,f. The extent to which the material was able to exit the device under these conditions was observed.

Fluidization and dense-phase conveying have very similar requirements for the material to be processed; however, requirements from the latter are slightly more restrictive [30]. Thus, although fluidizability is an indispensable prerequisite for dense-phase conveying, conveyability itself must be validated by actual conveying tests. The NF sample produced by semi-industrial equipment was tested for its ability to be conveyed in a dense-phase pneumatic conveying system by E.S.C.H. (Unterwellenborn, Germany). The test installation comprised two pressure vessels connected by a pipe with a length of 51 m and an inner diameter of 25.7 mm. This pipe comprised mainly horizontal but also vertical passages and seven 90° angles. A vertical cylindrical vessel with a conical lower end was filled with 380 kg of the sample, pressurized with N₂, and used for partially fluidizing the sample. Only within the conical lower end are gas velocities sufficiently high to achieve this—therefore—partial fluidization; whilst conveying, the sample is discharged at the bottom so that the remaining sample slowly flows into the fluidization zone and is also discharged and conveyed via the pipe to the receiving vessel. In several repetitions of the conveying procedure, parameters, such as gas flows and pressures, were altered to maximize the specific mass stream conveyed.

3. Results

3.1. Chemical Compositions

The results of the proximate analyses for all seven fuels of interest are shown in

Table 1. Proximate analyses and torrefaction solid yields (* as temperature measurement is not comparable to MF samples, only an estimation based on product analyses is indicated; ** solid yields could not be measured, due to high losses in the warm-up phase of the reactor).

	Pellets	MF285	MF310	MF325	MF340	NF	PC
T peak/°C	-	285	310	325	340	320–340 *	-
Moisture/%mass	7	2	2	2	3	2	1
Ash/%mass,db	2	3	4	5	5	5	7
Volatiles/%mass,db	82	64	53	42	41	40	20
Fixed Carbon/%mass,db	16	33	43	53	53	55	73
Fuel Ratio/-	0.20	0.52	0.83	1.28	1.3	1.4	3.65
yieldsolid/%mass	-	65	50	42	38	- **	-
yieldsolid,db/%mass	-	70	54	45	41	- **	-
yieldsolid,daf/%mass	-	70	53	44	40	- **	-

. Additionally, the torrefaction temperatures and solid yields are indicated for the samples produced in the muffle furnace. The characteristics of the pilot-scale sample NF which was produced according to the preliminary laboratory test results are shown as well to compare it to the lab samples. Proximate analysis of PC serves as a baseline of a PCI coal for further comparison. These data are also shown and are markedly different from the bioreductant types, as expected.

Considering the MF samples, a couple of observations are worth pointing out. As expected, the volatile content and the solid yield decrease with increased torrefaction temperature. Consequently, the fixed carbon content increases. It has to be stressed that the bioreductants’ organic volatile content is two to three times higher than that of the PC, which is potentially unfavorable. The yield_solid,daf is calculated by relating the proportion of organic matter of each sample to that from the input pellets; by doing so, the yield calculations are unaffected by the educts’ ash content and thus better comparable. It has to be emphasized that the analyzed torrefaction temperature range causes the steepest specific decrease in yield_solid,daf: towards 285 °C, a mass loss of 30% on the daf basis is observed, and the decomposition of a further 30% to a yield_solid,daf of 40% takes place by only increasing the peak temperature by 55 °C to 340 °C. With the further increase in the treatment temperature to pyrolysis conditions, the decrease in yield again significantly levels off [49].

The semi-industrial NF sample was meant to replicate the MF340 sample, and the resemblance regarding the volatile and fixed carbon content was satisfying. In contrast to the PC sample, all torrefied samples have a significantly higher volatile content. However, MF325, MF340, and NF are close to the mentioned 40 m-% of volatiles, which are understood to be acceptable for blast furnace injection.

The results for the chemical analysis are shown in

Table 2. Elemental composition (C/H/N/O) and reduction potential (ReP).

	Pellets	MF285	MF310	MF325	MF340	NF	PC
C/%mass,db	48.7	59.5	64.3	68.2	69.7	68.6	78.8
H/%mass,db	6.0	5.2	4.8	3.9	3.6	4.4	3.8
N/%mass,db	0.2	0.2	0.3	0.3	0.3	0.3	2.2
O/%mass,db	42.9	32.2	26.8	23.0	21.2	21.3	8.2
RePdb/tdb−1	98.2	134.5	151.0	161.3	165.5	166.3	205.6
RePdaf/tdaf−1	100.4	138.4	157.2	169.1	174.8	175.9	221.1

. All MF samples exhibit a continuous increase in carbon content with temperature. Even from MF325 to MF340, this increase is observable, although their proximate analyses are almost identical. The NF sample comprises a composition very similar to MF340, as intended. Notably, PC has a higher C/O ratio and thus also a higher ReP. When maintaining equal specific injection rates on the daf basis, the ReP injected per mass of iron would drop to at most 81% when replacing PC with one of these bioreductants. Therefore, an evaluation of the reasonability of this coal substitution approach must consider potentially increased coke demand, which certainly minimizes this approach’s ecological and economic impact. One approach of economically evaluating this implication on bioreductant design was presented elsewhere. There, the resulting difference in fuel costs and efficiency was shown to be outweighed by presumably lower emission certificate costs for prices expected at the end of this decade [42]. Nevertheless, the emission savings do not fully correspond to the amount of coal substituted but have to take increased coke rates into account. Moreover, a reasonable usage of the non-solid wood carbonization products—i.e., liquids and gasses—is another relevant factor for an ecological assessment.

The reduction potential ReP also increases with torrefaction temperature, as the carbon content increases and oxygen content decreases. The hydrogen content, which positively contributes to the ReP, also decreases by a similar factor, yet is less relevant due to its lower mass fraction. In [42], a regression analysis with a multitude of published torrefaction and pyrolysis tests resulted in an estimated ReP_daf as a function of the yield_daf. The current samples are plotted together with this function in Figure 2. Apart from MF285, all samples are pretty close to the regression result, with MF325 and MF340 even being within the 95% confidence interval. This comparison affirms the results’ appropriateness to published torrefaction and pyrolysis data.

Several suggested formulae were used to calculate the CRR for PC and all bioreductant samples. Figure 3 shows the CRR of the bioreductants relative to that of the PC. CCR ‘a’–‘c’ were calculated with formulae from [40], while CRR ‘a’ needed further input from a blast furnace mass and energy model [25]. CRR ‘d’ used a formula solely based on ash and moisture content [41], CRR ‘e’ relates the respective RePs already calculated, and CRR ‘f’ and ‘g’ are based on empirical formulae [25]. As the CRRs are only tentative in nature, the vast differences are to be expected; however, it shall be highlighted that the mean value of all calculated relative CRRs is almost equivalent to CRR ‘e’, which is based on the ReP calculation. This observation again emphasizes that the bioreductants are expected to contribute 20% less to the reduction process compared to the PC sample on an equal mass basis. Considering samples MF325 and MF340, the overlap of the error indicators shows only a slight improvement with higher torrefaction temperatures in this regard.

In the Van Krevelen representation, as shown in Figure 4, it is apparent that the chemical composition of the bioreductant types becomes more similar to PC with higher treatment temperatures. Comparing this diagram with other, more general diagrams of that kind, such as [50], shows that the pellets already comprise an unusual low O/C ratio, probably explaining the higher-than-expected ReP_daf for lower-temperature samples in Figure 2. Interestingly, this appears to affect the products’ RePs at lower treatment temperatures but not at elevated ones. An aspect complementing the ReP is the H/C ratio: the lower the H/C ratio is, the less steam is generated by oxidation of the bioreductant in the raceway, and the energy demand of the blast furnace is decreased [12]. The samples MF325 and MF340 are almost at the same level as PC and more favorable in this respect. The O/C ratio of all bioreductant samples is higher than that of PC, which is already reflected in the inferior ReP values.

3.2. Bulk and Particle Characteristics

In contrast to the comparative specimen PC, a median particle diameter of 70 µm was targeted for the bioreductant samples due to their tendency to produce a large fraction of ultra-fines and thereby resulting in SMDs below Geldart’s “A” class [29]. This diameter is significantly larger than that of PC. Generally, the largest particle diameter suitable for sufficient combustion under raceway conditions is chosen to decrease the comminution expense and conveying difficulties. It has already been shown that increased particle size is acceptable for bioreductants concerning their raceway burnout, as their higher porosity outweighs the size’s influence to some extent [51].

Table 3. Pulverized samples: bulk characteristics and QICPIC results.

	MF285	MF310	MF325	MF340	NF	PC
Angle of repose/°	48	46	43	39	41	44
bulk density/kg·m−3	400	420	480	500	450	480
x10,3/µm	14.5	11.55	10.88	10.68	11.67	8.01
x50,3/µm	101	67.17	71.18	68.94	58.39	35.79
s50,3/-	0.766	0.806	0.818	0.822	0.821	0.845
s < 0.7/m-%	29	16	11	11	10	6
SMD/µm	39.31	30.45	29.67	29.22	29.73	19.65

lists all measured mechanical characteristics of the pulverized samples. As can be seen, MF285 drastically fell short of the targeted x_50,3. Considering the particle size range of Geldart’s class “A” particles and the transition range to the cohesive class “C”, the particle size distributions of all other MF samples were practically identical; most importantly, the proportions of fines are equally similar [23] as are the mean diameters x_50,3. Particle size distributions are shown in Figure 5a. The NF sample’s particle size distribution is slightly lower than intended due to restrictions at the semi-industrial mill. With an increase in torrefaction temperature, the angle of repose and bulk density could also be reliably de- and increased, respectively. Measuring these aspects, however, involved stirring the samples through a funnel with a spoon, as all samples exhibited a tendency to bridging. A low angle of repose benefits the bulks’ flow- and fluidization behaviors [52,53], and high bulk density facilitates achieving high gas loading ratios. Interestingly, the NF sample’s bulk density fell short of that of the MF340 sample. Smaller particle sizes and slightly narrower size distributions serve as explanations for this observation [54], but probably not to that extent. It can be speculated that due to the rotary drum unit, the NF sample might be carbonized less uniformly than the MF samples, and less carbonized fractions exert greater influence on the overall bulk density. In both respects, the coal sample has intermediary qualities compared to the MF samples.

The sphericity, also shown in Figure 5b, does also improve significantly with higher torrefaction temperature. For powders with low sphericity, bad flow- and fluidization characteristics can also be expected [55]. A significant improvement is achieved especially between the two lower-temperature samples MF285 and MF310. A further increase in temperature does result in even higher sphericities; however, the improvement is less pronounced from MF310 to MF325. There is no significant difference between MF325 and MF340; it appears as if a certain threshold in this respect is approached in this investigated region of torrefaction temperatures. Similar improvements in sphericity with treatment temperatures have already been reported [12]. In comparison to the coal sample, all produced bioreductant samples are inferior regarding the sphericity.

Grindability and shear tests have only been performed with NF and PC. The grindability of the NF sample provided a TTBGI of 112, which is significantly higher than the usual HGI of <70 of a comparable PCI coal. Similar results were obtained elsewhere [24]. This means that the NF sample takes very little effort to grind. On one hand, this positively affects the energy demand of the milling equipment, but on the other hand, the high grindability might cause problems in material handling and transport due to particle breakage and dust generation [15]. It has been reported that lower treatment temperatures result in a lower TTBGI and less grindability [56], and might help to tackle possible problems. The co-grinding and co-injection of the bioreductant and coal might be of interest for industrial implementation but are not within the scope of this study. Studies on the co-grinding of bioreductants and coal questioned the equivalence of HGI and TTBGI values, and advised performing such assessments on an HGI basis alone [57]. Since information from the literature is not fully conclusive on this matter, specific tests need to be performed to see whether and, if so, how and to what extent the bioreductant’s high TTBGI (as a precursor for HGI) influences pneumatic transport properties in potential co-grinding applications.

Shear cell tests showed a significant difference in the bulk flow characteristics between the NF and PC regarding the flow function (compressed) (ff_c) value. The NF sample had an ff_c of 8.6 and PC had a value of only 4.8. Both values fall within the range of ‘easy-flowing’ powders from 4 to 10; however, PC’s 4.8 is close to the ‘cohesive’ class [58]. This might be due to the significantly smaller particle size. Elsewhere, even higher flow values for thermally treated biomass were retrieved, and potential problems such as dust leakages in valves were pointed out [12].

3.3. Fluidization and Conveying Results

The lab samples were tested for their fluidization behavior in order to determine a minimal torrefaction temperature needed for a large-scale conveying test.

Table 4. Fluidization results.

	MF285	MF310	MF325	MF340	NF	PC
Um,f/cm·s−1	>3	1.3	1	1	1.3	1.3
mfluidized/%	81–90	92	91.5	90	90	88.3
relrho/%	73–78	59	54	58	55	51
grc/s·cm−1	0.45	0.8	1	0.89	1.39	1.33
discharge/%	87	95	95	94	93	91

shows all corresponding results as well as the results from the NF and coal sample. The fluidization diagrams are shown in Figure 6.

The most important parameter in fluidization testing is the minimum superficial gas velocity U_m,f needed for fluidizing a powder bed. However, this parameter is not precisely determinable for samples which have a low and broad particle size distribution and a tendency for bridging. For samples with perfect fluidization behavior, this U_m,f is easily detectable as a peak in the differential pressure measurement at the moment of transition from the fixed to fluidized bed [30]. The presented measurements did not result in such peaks, but the transition to the fluidized bed appears to occur gradually. Therefore, U_m,f was estimated as described in Section 2.4. The MF285 sample did not fluidize properly, most probably because of its decidedly lower particle sphericities. With superficial gas velocities of up to 3 cm·s⁻¹, the bed height did not stabilize nor did the differential pressure. Its U_m,f was indicated with >3 cm s⁻¹, although a partial fluidization was observed and material discharge was possible; but with Δp main dec being higher than Δp main, it must be concluded that no prior sufficient fluidization was achieved. These results also explain why the interpretation cannot rely on only one indicator: The value of Δp main and the corresponding m_fluidized are sufficiently high; however, all other observations indicate that friction between the sample and the cylinder probably affected this measurement. The U_m,f of all other samples could be estimated more precisely and were in the same range as that of PC. The m_fluidized of the PC sample is slightly different: although the pressure signal along with the visual observation of constant bed height signifies a fully fluidized bed, m_fluidized is slightly lower. This could indicate that a larger portion of particles rests on the steel disk, neither participating in fluidization nor interfering with it. Indeed, this was visually observable in some portions of the powder bed’s perimeter. However, in coherence with the MF285 interpretation, this lower m_fluidized can also signify less friction between PC and the device’s cylinder.

In the fluidized state, the bed height of all samples increases significantly. The relative fluidization density is calculated as a fraction of the initial bed height to the fluidized bed height. It is less than 60% for all fluidized samples. The measured values for gas retention capacity grc are related to the initial bed height. Only the NF sample comprises a grc comparable to PC’s.

The peak in pressure drop in Figure 6 at the start in MF310, MF325, NF, and PC is due to the initial levitation of the whole powder bed. These levitations continue occurring with most samples, but mostly collapse again quickly, and are most probably caused by too small an inner diameter of the testing device for this kind of powder. Fortunately, the behavior with initial gas velocities does not affect the assessment of the general fluidizability and the minimum fluidization velocity U_m,f. The latter can also be roughly estimated by decreasing the gas velocities after reaching fluidization [30], as shown in Figure 6 as Δp main dec. Except MF285, all samples exhibited similar behavior in this regard, largely resembling Δp main well below 1 cm·s⁻¹ u_gas. Data for PC are peculiar as Δp main dec is above Δp main at lower u_gas; this relates to PC’s little resistance to lower gas streams after collapsing from a levitating bulk. As stated in Section 2.4, such instances might signify a tendency for unreliable fluidization behavior; however, as the respective gas velocities are low and bulk levitations are unlikely in partial fluidization, as in PCI, this observation appears irrelevant.

The lowest recorded pressure drops often do not relate to the resistance of the powder bed as a whole, but rather signify the channeling of the applied gas stream through one or more channels across the bed. All graphs have in common that the measured maxima and minima, shown as dashed lines in Figure 6, draw closer with increasing gas velocities. This phenomenon is interpreted to signify growing homogeneity regarding the fluidization state within the powder bed. The facts that this is reliably observable, the good congruence of Δp main and Δp main dec, and that the measured specific U_m,f is sufficiently repeatable from one test to another foster confidence in the sufficient fluidization behavior of all bioreductant samples excluding MF285. Possibly, this failure in fluidizing at gas velocities below 3 cm·s⁻¹ is due to the larger particle sizes after comminution, and a repetition with MF285 with x_50,3 = 70 µm might have provided positive fluidization results with already lower gas velocities. However, as other characteristics, especially the volatile content, were also inferior, a repetition of testing MF285 with lower particle size was deemed expendable in the course of this study.

Finally, the testing of the NF sample’s conveying characteristics was successful in terms of verifying its general ability to be conveyed pneumatically in the dense phase. Moreover, it was possible to achieve a solid loading ratio of more than 150 g_solids/g_gas by attuning pressure levels and conveying gas amounts. This is higher than the usually employed loading ratios in the PCI system, and the achieved conveying rates on the test apparatus were also sufficiently high with acceptable gas rates. In general, the conveying behavior of the NF sample largely resembled the coal conveying test results with the same equipment.

4. Discussion of the Minimal Torrefaction Temperature

The discussion of the results is separated into two sections, namely a discussion of the aspects relating to the injection system—that is, grinding, fluidization, and conveying—and aspects relating to the blast furnace itself. In both sections, a synopsis of the presented results and additional observations reported in the literature is offered. Where applicable, the implications of the bioreductant types on full or partial replacement are differentiated.

4.1. Implications of Torrefaction Temperature on Usage in PCI System

The comminution of the bioreductant-type NF is expected to demand significantly less specific energy than coal, as the TTBGI test shows. This is in accordance with many other results [59,60,61]. For bioreductant types of lower torrefaction temperature, somewhat lower grindability values are reported [56]. However, the mill type has proven to affect the particle shape and size distribution of thermally treated biomass [15,46]. As these studies were performed with wood chips, their applicability on already pelletized material, as in this study, is unclear. Nevertheless, it is important to emphasize that changing milling equipment is not only undesirable but will also most probably result in a decreased energy efficiency. Roller mills are very common in PCI systems [25] and they are understood to perform very efficiently [62]. It has been reported [63,64] and also observed in previous tests that it is more difficult to avoid oversized particles after the classifier with bioreductants, perhaps due to irregular particle shapes and very low density. Again, it is not yet clear to what extent this also applies to torrefied pellets, as the density is most influential on the classifier’s selectivity. The grindability TTBGI is beyond the usual values for PCI systems [25]. Implications of this were already discussed in Section 3.2. High-HGI coals tend to exhibit stronger caking behavior and are therefore usually not considered for PCI applications [65]. No such correlation between the grindability and caking behavior of bioreductants is known to the authors; thus, a high TTBGI should not cause such reservations. For possible co-grinding applications in a partial substitution approach, a vastly differing grindability might be problematic nevertheless. In this case, opting for lower treatment temperatures in an attempt to decrease grindability might prove reasonable in this respect [66].

The fluidization of most samples worked properly. Only MF285 had no proper fluidization behavior. Using this bioreductant type in a dense-phase conveying system is not recommended without further fluidization aids. Considering a partial substitution, however, might be reasonable for very low proportions in blends. MF310 exhibited similar fluidization behavior to the higher-temperature samples, and also the particle shape was only slightly inferior. Apparently, major changes take place within the bioreductant in the torrefaction temperature range between 285 and 310 °C. The chemical analyses back this assumption. Other parameters, such as the angle of repose and bulk density, do not encourage considering MF310 for usage for the targeted purpose; clearly, some drawbacks regarding flowability in contrast to the higher-temperature samples and PC are evident, although the measured U_m,f was equal to PC’s. Again, while MF310 might be an option for the partial replacement of coal, the minimum advisable torrefaction temperature for full substitution in this respect should be higher than 310 °C up to 325 °C. From MF325 to MF340, no notable improvement in all tests regarding the fluidizability were encountered, where the results of both samples were not at least equivalent to the PC’s.

The conveying of the bioreductant was tested only for the NF sample, which largely resembles the MF340 sample. MF285 and MF310 were excluded because of the aspects mentioned in the previous paragraphs. MF325 was also excluded as MF340’s bulk density was notably higher, although, unfortunately, NF’s bulk density was lower again. What caused this unexpected drop in bulk density is unclear, as the influence of the different mill type could be ruled out by small NF educts pulverized on the lab-scale mill as well: here, the bulk density was even lower than that of the final semi-industrial product. However, the conveying test was successful inasmuch as the fluidization step worked reliably and conveying rates and solid loading ratios were satisfying. No observations regarding negative implications on the conveying behavior compared to coal were made; only the slightly inferior bulk density to the MF340 sample should be addressed in further test runs.

4.2. Implications of Torrefaction Temperature on Usage in Blast Furnace

A general implication on the whole PCI system’s contribution to the blast furnace process is the lower energy density and reduction potential in the bioreductant compared to PC. Potential improvements in, e.g., the energy demand in grinding might be offset by higher overall expenditures in relation to the reductants’ avail in the blast furnace’s fuel supply. The most important parameters discussed for these implications are the ReP and the CRR, with both being an estimation of an alternative reducing agent’s contribution to the reduction of iron ore and the thermal needs of the blast furnace. Both, and by definition, especially the ReP, can only estimate the theoretical maximum contribution based on the chemical composition and stoichiometric requirements. Concerning this maximum, the samples fall short from the PC results, with MF325 and MF340 exhibiting a CRR of 0.8 relative to PC. Along with the reduced bulk- and conveying density, this will result in a drop in injection performance of roughly 30% in this regard when maintaining the same volumetric injection rate. For the samples of lower treatment temperature, this drop is even more severe.

However, there are more factors affecting the effective replacement ratio. Leaving it at that would oversimplify the processes in the raceway which comprise pyrolysis [67], devolatilization of the alternative reducing agent (ARA), cracking and oxidation of the released volatiles and the remaining char, and solution loss reaction (i.e., Boudouard reaction) of the char with the generated carbon dioxide [26,68]. While ReP and CRR valuably estimate a theoretical optimum, the ability of ARA to achieve this optimum within given conditions cannot be assessed easily, let alone generically. Cameron et al. [25] insist that only a prolonged period of in-process testing can provide reliable conclusions. Certain basic requirements can be postulated, nevertheless. One such requirement is the general reactivity, describing the readiness of ARA to oxidize with O₂ and CO₂ [65]. High reactivity can provide gains in effective CRR as losses of unused chars in the bosh gas or slag can be minimized [68], but it also contributes positively to CO generation at lower temperatures than coke [69], enabling an increase in overall efficiency [70]. There appears to be a consensus that thermally treated biomass and their chars have a higher reactivity than coal [71,72]. Its positive effect on burnout behavior has already been demonstrated [73,74], as well as for torrefied materials with especially higher torrefaction temperatures [75]. Another parameter on which to base the bioreductant assessment is the volatile content. This volatile matter is related to a reduced temperature in the raceway, due to devolatilization and endothermic cracking [26], which was related especially to the injectant’s tar content in an early study [76]. Arising additional thermal demand reduces the ARA’s aggregated contribution to the process. Neither aspect allows for a precise definition of a recommended minimal torrefaction temperature. However, both imply that the lower the treatment temperature is below 325 °C, the more caution in process configuration is advisable.

Aspects relating to ash composition, such as potentially improved blast furnace efficiency due to the basicity of ash [77], or the detrimental effects of the biomass’ alkali content [24], are not considered, as the ash composition should not be affected by torrefaction temperature and, thus, does not contribute to defining the minimum suitable torrefaction temperature.

5. Conclusions

To define a minimum torrefaction temperature for bioreductant production, four different samples of bioreductants were produced with intermediate and high torrefaction temperatures, comprising great variations in chemical composition and characteristics in bulk handling. Testing the mechanical characteristics and fluidization behavior provided satisfying results for the samples with torrefaction temperatures of 325 °C and above. The results do not provide evidence that, with these types of bioreductants, a substitution of coal in an ordinary PCI system should not be feasible. On the contrary, pilot-scale conveying tests of a 340 °C semi-industrial sample emphasize the similarity to coal in this respect. Because of the lower reduction potential of the samples of all tested treatment temperatures and their lower bulk density, the need for higher-than-usual injection rates might arise to maintain the PCI system’s avail. To address this topic without increasing injection rates, bioreductant treatment temperatures must be above the presented temperature range; assessing the reasonability of doing so, however, is an economic consideration beyond the scope of this report.

Although the analyzed range of torrefaction temperature is small, a pronounced difference in volatile matter content was observed. The lowest measured value for the 340 °C sample was emphasized to be merely within the usual range considered for injection coals. The resulting possible implications on the blast furnace efficiency were delineated, but they have not been fully evaluated yet. Future research is suggested to focus on the impact different bioreductants—produced at the presented minimal carbonization temperature and above—have on the in-raceway conversion, the respective flame temperatures, and blast furnace thermal control, in order to fully assess a potential fuel substitution.