Alkali Circulation in the Blast Furnace - Process Correlations and Counter Measures

Alkali Circulation in the Blast Furnace - Process Correlations and Counter Measures

Joel Carlsson

Sustainable Process Engineering, master's level 2018

Luleå University of Technology

Department of Civil, Environmental and Natural Resources Engineering

Acknowledgements

This master thesis was performed at Swerea MEFOS in Lule˚ a during the spring semester of 2018 and was the last step in my master’s degree at Lule˚ a University of Technology in Sustainable Process Engineering.

This thesis would not have been possible without the support I got and there are several people that I would like to thank. Amanda and Lena my supervisors at Swerea MEFOS and Anton my supervisor from LTU. For putting up with all my questions and helping me with the report. MEFOS and everyone there for helping and allowing me to perform my master thesis at their facilities. Hesham and Britta for helping with all my TGA experiments. My family for just being there for me through all these years and Linnea for all her support during the thesis. All the friends that I gotten to known through all my years at LTU and that I believe I will know for many years to come. A final thanks to Gelbe for being our dark, cold and dingy basement through many years of study. Hopefully I will never see you again.

Lule˚ a, August 2018

Joel Carlsson

Abstract

In blast furnace ironmaking one major challenge is to control and measure the alkalis circulating and accumulating in the blast furnace (BF). Alkali enter the BF with the primary raw material and will form a cycle where it is first reduced to metal at the lower parts forming gas. Alkali then follows the gas flow up where it oxidizes and solidifies as the oxide form has a higher melting and volatilization temperature.

Condensation then occurs on burden material and in their pores and by that it is following the burden downwards. The circular nature of the reactions leads to a build-up of alkali in the form of potassium in the BF that is hard to control or measure. Condensation of alkali compounds can also occur on the BF walls functioning like a glue to which particles attach, forming scaffolds that can rapidly increase and disturb the burden descent. The increased alkali catalyzes gasification of coke with CO

that increases coke consumption and leads to disintegration of coke. A common method today to control alkali is by varying the basicity in the BF. As lower basicity increases the amount alkali removed through slag while at the same time reducing the amount of sulfur that can be removed with the slag.

This project was divided into two parts. The first part was a continuation of a previous study performed at Swerea MEFOS. Where to control the effect of alkali on coke gasification a method was tested using coke ash modification to inhibit the catalyzing properties of alkali bound on coke. The method has previously shown that alkalis are bound in the desired form but the added amount was not sufficient for inhibition of all picked-up alkalis. In this study, additional trials with higher additions of kaolin was performed. 2 wt%

kaolin was added to the coal blend for producing coke that was then added to LKAB’s experimental blast furnace (EBF) as basket samples in the end of a campaign. The excavated samples were analyzed using XRF, XRD, SEM-EDS and TGA to find if the alkali was bound in aluminum silicates in the coke ash, if the addition was sufficient for binding all alkalis and if the catalytic effect in coke gasification had been achieved.

The second part was a novel approach with a statistical process analysis using SIMCA to connect top gas composition of SSAB Oxel¨ osund’s BF No. 4 to alkali content using process data. The approach investigated the correlation between NH

(g) and HCN(g) in the top gas to alkali content. Expanding on the possibility to measure alkali content quickly for the operators using top gas measurements. Top gas composition was measured using a mass spectrometer (MS) and where complimented with process and tap data provided by SSAB. Data was analyzed using the multivariate analysis tool SIMCA 15 to find possible correlations.

Results from the first part showed that the alkali that was found was present as alkali aluminum silicates

independent of kaolin addition after the EBF. As temperature along gas composition was the main factors

behind alkali uptake in coke. Main differences in alkali uptake and development of coke properties in the

BF was linked to the temperature and gas composition profile during tests campaigns compared. Results

from TGA showed that the reaction rate of coke with CO

increases with increasing K

O and that start of

reaction was lower with increasing alkali. The results from the second approach did not find a correlation

between HCN(g) and K

O in slag. Positive correlation could be seen between HCN(g) and increased SiO

in slag and that H

O(g) would affect HCN(g) negatively.

Sammanfattning

En av de st¨ orre utmaningarna vid j¨ arnproduktion i en masugn ¨ ar att kontrollera och m¨ ata alkali som cirkulerar och ackumuleras i masugnen. Alkali f¨ oljer med det prim¨ ara r˚ amaterialet in i masugnen d¨ ar det cirkulerar enligt en process som startar med att alkali reduceras till metallform och f¨ orgasas. Alkaligasen f¨ oljer sedan gasfl¨ odet upp i masugnen d¨ ar den oxiderar och ¨ overg˚ ar till fast form eftersom oxiden har h¨ ogre sm¨ alt- och f¨ or˚ angingstemperatur. Den fasta fasen kondenserar p˚ a beskickningen och i dess porer vilket g¨ or att alkali kan f¨ olja med beskickningen ner i masugnen igen. Vilket skapar den cirkul¨ ara processen som g¨ or att alkalim¨ andgen i masugnen byggs upp ¨ over tid fr¨ amst med avseende p˚ a kalium. Kondensation av alkali p˚ a masugnens v¨ agar kan ocks˚ a ske och bilda p˚ akladdningar vilka hindrar beskickningen fr˚ an att f¨ ardas korrekt genom masugnen. Alkali katalyserar ¨ aven f¨ orgasningen av koks vilket leder till ¨ okad kokskonsumption och ¨ okad nedbrytning av koks. Dagens metod f¨ or att hantera alkali i masugnen ¨ ar att s¨ anka basiciteten i slaggen vilket ¨ okar m¨ angden alkali som kan transporteras ut med slaggen. D¨ aremot minskar m¨ angden svavel som samtidigt kan tas bort i fr˚ an r˚ aj¨ arnet genom slaggen.

Projektet ¨ ar uppdelat i tv˚ a delar. Den f¨ orsta delen ¨ ar en forts¨ attning av en tidigare studie fr˚ an Swerea MEFOS i vilken en metod testades f¨ or att minska den katalyserande effekten alkali har p˚ a

koksf¨ orgasningen genom att modifiera koksaskan s˚ a att effekten inhiberades f¨ or alkali bundet till koksen.

Metoden har tidigare visat att alkali ˚ aterfinns i den ¨ onskade formen i askan men att m¨ angden av tillsats varit f¨ or liten f¨ or att inhibera all alkali som hittats i proven. I den h¨ ar studien utf¨ ordes ytterligare f¨ ors¨ ok med h¨ ogre halter av kaolin ¨ an vad som testades tidigare och resultaten j¨ amf¨ ordes. 2 viktprocent av kaolin tillsattes till kolmixen som anv¨ andes f¨ or att producera koksen, koksen tillsattes sedan till LKAB:s experimentmasugn (EBF) som korgprover i slutet av en f¨ ors¨ okskampanj. Efter avslutad kampanj kyldes EBF:n med kv¨ ave och gr¨ avdes ut varvid proverna togs ut. Korgproverna analyserades med XRF, XRD, SEM-EDS och TGA f¨ or att fastst¨ alla om alkali var bundet till aluminiumsilikater i koksaskan, om m¨ angden tillsatt kaolinit var tillr¨ acklig f¨ or att binda all alkali och om den katalyserande effekten p˚ a koksf¨ orgasningen hade p˚ averkats.

Den andra delen i projektet utforskade en ny metod som utgick fr˚ an att med hj¨ alp av en statistisk processanalys erh˚ alla indikationer p˚ a m¨ angden alkali som cirkulerar i SSAB Oxel¨ osunds masugn. Metoden unders¨ okte m¨ ojligheten att korrelera m¨ angden NH

(g) och HCN(g) i toppgasen mot alkalihalten i masugnen, den senare baserat p˚ a processdata fr˚ an masugnen. En s˚ adan metod skulle ¨ oka m¨ ojligheten f¨ or processoperat¨ orer att enkelt och snabbt kunna uppskata m¨ angden alkali och vidta n¨ odv¨ andiga ˚ atg¨ arder.

Toppgassammans¨ attningen m¨ attes med en masspektrometer, vilket kompletterades med process- och tappningsdata. Data analyserades med flervariabelanalys verktyget SIMCA 15 med avsikt att kunna identifiera m¨ ojliga korrelationer i data.

Resultaten fr˚ an f¨ orsta delen av projektet visade att alkalin som hittades ˚ aterfanns som alkalialumini- umsilikater i alla prov efter att de varit i EBF:n. Huvudfaktorerna som p˚ averkar alkalihalten i koks

¨

ar temperaturen och gassammans¨ attningen i masugnen. Vilket gjorde att skillnaderna som kunde ses

mellan korgproverna fr˚ an de tv˚ a kampanjerna var l¨ ankad till temperatur- och gassammans¨ attningsprofilen

i EBF:n. TGA-resultaten visade att reaktionshastigheten f¨ or koks i CO

(g) ¨ okade med ¨ okande alkalihalt

och att starttemperaturen f¨ or reaktionen minskade. Resultaten fr˚ an den andra delen visade att det inte

gick att hitta en signifikant korrelation mellan HCN(g) och halten kaliumoxid i slaggen. Det kunde ses en

positiv korrelation mellan HCN(g) och ¨ okad halt SiO

i slaggen och att H

O(g) kom att p˚ averka halten

HCN(g) negativt.

Table of Contents

1 Introduction 1

1.1 Background . . . . 1

1.2 Objective . . . . 2

1.3 Scope . . . . 2

2 Theory 3 2.1 Blast Furnace Operation . . . . 3

2.1.1 Basic Reactions of Iron and Coke . . . . 3

2.1.2 The Alkali Cycle in the Blast Furnace . . . . 4

2.1.3 Alkali Effect and Removal . . . . 6

2.1.4 Ammonia and Hydrogen Cyanide Formation in the Blast Furnace . . . . 6

2.2 Multivariate Analysis . . . . 8

2.2.1 Principal Component Analysis . . . . 8

2.2.2 Partial Least Squares Projections to Latent Structure . . . . 10

2.3 Coke Characterization . . . . 11

2.3.1 Coke Graphitisation . . . . 11

2.3.2 Reaction Rate and Activation Energy . . . . 12

3 Method 13 3.1 Basket Samples in EBF . . . . 13

3.1.1 Preparation of Coated Coke and Coke With Modified Ash . . . . 13

3.1.2 Basket Samples in the EBF . . . . 13

3.1.3 Characterization of Samples . . . . 14

3.1.3.1 XRF . . . . 14

3.1.3.2 SEM-EDS . . . . 14

3.1.3.3 XRD . . . . 14

3.1.3.4 TGA . . . . 14

3.2 Correlation of Ammonia and Cyanide in the Top Gas . . . . 14

3.2.1 Top Gas Measurements . . . . 14

3.2.2 SIMCA . . . . 15

4 Results 16 4.1 Basket Samples From EBF . . . . 16

4.1.1 Basket Location . . . . 16

4.1.2 XRF . . . . 16

4.1.3 SEM . . . . 18

4.1.4 XRD . . . . 20

4.1.5 Coke Graphitisation . . . . 22

4.1.6 TGA . . . . 23

4.2 Correlation of Ammonia and Cyanide in Top Gas . . . . 26

4.2.1 Process Data Excel . . . . 26

4.2.2 SIMCA . . . . 28

4.2.2.1 MS Data . . . . 28

4.2.2.2 MS Data + Process Data . . . . 32

4.2.2.3 MS Data + Process Data + Tap Data . . . . 35

5 Discussion 40 5.1 Coke Analysis . . . . 40

5.1.1 Alkali Uptake Depending on Position and Coke . . . . 40

5.1.2 SEM Results . . . . 40

5.1.3 XRD for analyses and determination of L

. . . . 41

5.1.4 Coke Reactivity and Reaction Rate . . . . 41

5.1.5 Final Coke Discussion . . . . 41

5.2 Process Data Analysis . . . . 41

5.2.1 Data handling and SIMCA . . . . 41

5.2.2 Final Process Data Discussion . . . . 42

6 Conclusions 44 6.1 Coke Analysis . . . . 44 6.2 SIMCA Analysis . . . . 44

7 Future Work 45

References 47

8 Appendix 48

8.1 Appendix A . . . . 48

8.2 Appendix B SEM images . . . . 52

Abbreviations and Chemical Formulas

Al

O

Alumina

BF Blast furnace

C Carbon

CO Carbon monoxide

CO

Carbon dioxide

EBF Experimental blast furnace

EDS Energy dispersive X-ray spectroscopy

Fe Iron

FeO W¨ ustite

Fe

O

Hematite Fe

O

Magnetite

HCN Hydrogen cyanide

H

O Water

K Potassium

K

O Potassium oxide

(K,Na)CN (Potassium or Sodium) cyanide (K,Na)

CO

(Potassium or Sodium) carbonate

MS Mass spectrometer

MVA Multivariate analysis

Na Sodium

NH

Ammonia

PCA Principal component analysis PLS Partial least squares

SEM Scanning electron microscopy TGA Thermal gravimetric analysis

tHM Ton hot metal

X(g) Component X in gas form X(s) Component X in solid form X(l) Component X in liquid form XRD X-ray diffraction

XRF X-ray fluorescence

1 Introduction

The background, objective and scope for the master thesis is presented in this section.

1.1 Background

During the production of iron with a blast furnace (BF), the quality of the charged raw material is important to avoid problems in the process caused by unwanted compounds entering the furnace. As the competitiveness between steel mills have increased globally the need to be able to charge lower quality material have increased while still keeping the iron quality high. The main unwanted compound present in the charge that can be troublesome with respect to removal and the performance of the BF is alkali compounds (K, Na) [1, 2].

The problem with alkalis is that they form a circulating load in the BF as previously shown in literature studies and books about the BF [1, 3] as well as in experimental work [2, 4]. Alkali will continuously be cycled in the furnace unless the basicity is low enough [5] and the charge contains silicates as the main form of alkali added to a BF. The circulation can be summarized as: alkali compounds follow the burden to the lower parts where they are reduced and forms alkali vapors at approximately 1600

C. Those vapors react further in the lower parts with the blast gases of the furnace to form alkali cyanides. The cyanides then rises quickly through the furnace due to the high gas flow condensating when the temperature goes below the boiling point for (K,Na)CN at 1625

C and forming a liquid phase. Alkali cyanides can also form alkali carbonates through oxidation when the temperature decreases and thus the cycle starts again.

Most of the alkali circulates rather than exiting through the top gas [1, 3, 4].

The presence of alkali leads to lowered production and higher coke consumption in the BF, approximately 4.5% and 2.3%, respectively, for each kg/tHM alkali added with the top charge of raw material [5]. Multiple sources report the same reasons behind the decreased production. Alkali decreases the production by:

lowering the threshold for the Boudouard reaction, increased coke gas and reduced strength of coke. Gas permeability is decreased due to coke degradation and scaffolding on the walls can happen reducing the volume of the BF [1, 5, 6, 7]. Thus, it would be of interest to develop a process support tool that can help operators to control the operation for optimizing the production depending on alkali load and for the operators to know when it is necessary to bleed more alkali by running it with parameters that decrease the amount of alkali. One possible way to remove alkali is to lower the basicity so that more alkali goes to the slag phase. While at the same time giving a worse iron quality due to increased sulfur levels so a balance would be needed [4, 5].

As measurements on the top gas from the BF can be used to see the amount of NH

and HCN there. It could allow operators to have more control over alkali as alkali participate in reactions together with CO and H

O producing NH

and HCN. Using the knowledge of the reactions coupled with measurements of the top gas performed by MEFOS on SSAB Oxel¨ osund’s BF, the total alkali could possibly be derived directly from the top gas content.

A previous master thesis done at Swerea MEFOS investigated how three different coke additives would change the effect alkali had on coke [8]. With purpose to help bind alkali to the coke stronger so it will not start the alkali cycle as easily. In the present report further work was performed to see how kaolin addition would affect the properties of coke with regards to alkali. Kaolin was either mixed to the coal blend or coated on the coke. A major process data analysis was also done as the second part of the report to connect the top gas composition of the BF to the total alkali present in the BF. The work to characterize, measure and remove alkali present in the BF was part of a larger RFCS-funded project that Swerea MEFOS had a part in called ALCIRC.

Alkali is today generally removed by lowering the slag basicity to increase the uptake of alkali or decreasing

the amount of recycled material [5, 7]. Knowing when to perform those process changes cannot be known

until a tapping is done and the alkali in the slag can be coupled with a mass balance to know the true

alkali load in a BF. If the amount of alkali could be derived at real time for the operators, the processes

against alkali could be deployed on a more controlled basis. This is important as the methods used to

lower alkali can affect iron properties and productivity negatively by e.g. increasing the sulfur included in

the hot metal.

1.2 Objective

The main objective of this master’s thesis was to:

• See how addition of 2 wt% kaolin to coke and a kaolin coating would affect the way alkali react with coke compared to 1 wt% kaolin.

• Investigate the different process variables that can affect alkali circulation and top gases in the BF through a literature review.

• Investigate if NH

and HCN levels in the top gas of Oxel¨ osund’s BF No. 4 can be correlated to the alkali load in the BF

1.3 Scope

This thesis investigated how the chemical composition of the top gas could be connected to the alkali level in the BF using process data, mass balances and multi-variate analysis. Noise in the data had to be handled, investigating the equilibrium between HCN and NH

was also necessary and the different parameters that can affect the reactions. A thorough research in to the different measuring methods used and what type of analysis that must be performed also lay in the scope of this thesis.

Practical work using previously developed methods at Swerea MEFOS on coke samples from the EBF.

The goal was to investigate the difference of using a kaolin coating compared to a 2wt% kaolin addition

by using XRF, XRD, TGA and SEM-EDS with respect to the retention of alkali in coke. The results

from the coke samples could then be compared with a previous study on this subject by Olofsson [8].

2 Theory

The theory covers a short description of the basis of the BF process and give a background on how alkali problem affects it. Further study is done on how alkali cyanides present in the BF will affect the top gas composition. It is important to know the reactions and ratios between them to control the alkali in the BF.

2.1 Blast Furnace Operation

The most common base unit used around the world for production of iron from primary raw material is the BF. A BF can generally be described as a counter-current heat exchanger where the inside is buildup of multiple layers of coke and iron burden, where the iron burden generally consists of either sinter or pellets [6].

2.1.1 Basic Reactions of Iron and Coke

The iron burden will react, melt and be tapped as hot metal at the bottom of the furnace along with slag.

The process is a reduction process where Fe

O

(s) is reduced to Fe(s). The reduction of iron goes through several steps as the burden descends in the BF. The first step is an indirect reduction of hematite into magnetite:

3F e

O

+ CO 2F e

O

+ CO

(2.1)

Followed by indirect reduction of magnetite:

F e

O

+ CO F eO + CO

(2.2)

The final indirect reduction of iron is:

F eO + CO F e + CO

(2.3)

If reaction 2.3 does not completely reduce all w¨ ustite a direct reduction with C will happen in the lower part of the BF [9, 10]:

F eO + C F e + CO (2.4)

The reactions will happen in the different thermal regions of the BF. They are generally denoted as preheating zone, thermal reserve zone, and melting zone as can be seen in Figure 2.1. In the first two zones reduction is indirect, as shown in reaction 2.1-2.3, while in the last zone reduction is direct [9]. Part of the material added to a BF will be tapped as slag, a phase that consists mostly of the gaunge from the burden that enter the BF as silicates and alumina compounds. Slag has a lower density than iron and floats on top of the hot metal in the bottom of the BF. A small amount of slag formers are added with the charge to get the desired chemical properties of the slag. Common slag fluxes are CaO and MgO depending on what basicity that is desired by the operators [6].

As the iron burden descend, so will coke. Coke in the BF has several important functions. It has to have the physical properties and strength to both hold the iron burden and at the same time be permeable enough to allow gas to flow through it. Down in the melting zone coke is the only material still in solid form until it reaches the raceway where it will be consumed together with injected reducing agents according to,

C + O

CO

(2.5)

and both be the fuel giving heat to the melting of iron, and the chemical reductants in the form of CO(g) needed for the process. The flame temperature in the raceway will be above 2000

C depending on the ratio of coal/natural gas/oil used in the BF [6]. The CO

(g) produced in reaction 2.5 will further react with unburnt carbon when the temperature is between 900-950

C through reaction with carbon:

C(s) + CO

(g) 2CO(g) (2.6)

Reaction 2.6 has two names, the Boudouard or solution loss reaction and is an endothermic reaction

[1, 9, 10, 11]. The reaction is vital for the efficiency of a BF as it will ensure that the CO

(g) produced at

higher temperatures will be transformed back into CO that can further reduce iron oxides further up the

BF [12]. Without the reaction, reaction 2.1-2.4 would not be as efficient.

Figure 2.1 Thermal zones found in a BF [9].

2.1.2 The Alkali Cycle in the Blast Furnace

Alkali will inevitable enter the BF with the iron material and with the coke in the form of silicates. Iron producers generally want to limit the amount of alkali to around 1.5-5 kg/tHM [6] and a literature study on typical alkali limits in plants around the world showed that the real limit could vary between 2.5 to 7.5 kg/tHM depending on plant [13]. Of the two alkali substances sodium and potassium, potassium is generally the main compound entering the BF [5, 6]. Most of the alkali will exit with the slag while some will follow the top gas out as dust and gas. Recirculating alkali can either be removed by the slag or the gas. Potassium will follow the top gas to a higher degree as it is more volatile compared to sodium that will follow the slag more [14, 15].

According to Abraham and Staffansson [3] the behavior of alkali can be explained as following. Alkali enters into the BF in the form of silicates which can be simplified as (K,Na)

SiO

. Further text will only speak of potassium as sodium can be assumed to react in a similar way as potassium. Research in the alkali cycle show that the silicates will descent with the burden and the cycle will start with the alkali silicate being reduced by the coke in the melting zone according to:

K

SiO

+ C(s) 2K(g) + SiO

+ CO(g) (2.7)

The reactions take place at approximately 1550

C according to thermodynamic data for the reactions.

Any alkali oxides that enter or are formed in the BF react further up in the BF at lower temperatures according to:

K

O + CO(g) 2K(g) + CO

(g) (2.8)

As they are not stable [3, 16]. K

O can also dissolve into the primary slag [6].

Further the potassium vapors produced at the hearth of the BF shown in reaction 2.7 will react with the coal and nitrogen injected with the hot blast.

2K(g) + 2C(s) + N

(g) 2KCN (g, l) (2.9)

The boiling point for KCN is 1625

C so as the potassium cyanide rises away from the hot blast from the tuyeres, it transforms into a liquid phase when the temperature drops. The time in the tuyere zone is very short due to the high gas flow so the alkali cyanides have time to move up the BF before transforming into liquid phase. Further up in the furnace alkali cyanides will react with carbon dioxide to form more stable carbonates at temperatures below 1100

C.

2KCN (l) + 4CO

(g) K

CO

+ N

(g) + 5CO(g) (2.10) The carbonates will either follow the top gas out as gas, or be deposited on the burden as they start to condensate below 900

C. Compared to alkali silicates, alkali cyanides are unstable so any silica present in the hearth part of the furnace can react with the alkali cyanides to again form alkali silicates [3]. The process of alkali silicates reducing into alkali vapor, that ascend in the BF, exit with the top gas, or react with carbon dioxide to form carbonates can be summarized as the alkali cycle. Several authors have done slightly different summarizing of the process differing exactly which reactions that take place. There are doubts whether carbonates actually are formed at all at the top of the BF as carbonates are not found during excavation of BF:s. Nonetheless the main process that alkali cyanides are formed and that alkali circulate in the BF is agreed on [1, 3, 6, 13, 16]. In figure 2.2 an example of the alkali cycle can be seen. The charged material will descend to the high temperature zone before alkali silicates either will decompose to alkali vapors or be absorbed by the primary slag phase in the form of K

O or Na

O. The cycle also indicates approximately when the alkali vapor would react with silicates to again form silicates [7]. The distribution of alkali vapors through the BF depends on the gas flows path and the extent of central gas flow. Gas flow have a great effect on how heat is distributed in a BF. More central flow means more melting in the middle and less in the periphery of the BF [6].

Figure 2.2 Alkali circulation in a BF. Lines that are solid indicates the solid flows and

broken lines the gas flows in a BF [7].

2.1.3 Alkali Effect and Removal

One of the main negative effects of alkali is that it catalyzes the Boudouard reaction, lowering the temperature for the reaction from 900-950

C down to approximately 750-850

C and increasing the coke reactivity depending on the coke. It will also affect the coke structure negatively [12, 17]. The lowered threshold for the Boudouard reaction means that more carbon will be consumed in the BF in a strongly endothermic reaction. Thus, increasing the coke addition needed to the BF to keep a stable operation with 2 to 10 kg coke per kg alkali or with 6 to 11 kg depending on sources used [6, 7]. A third effect of alkali is the increased chance of scaffold formation in the shaft as alkali condense on the lining and may bind fine material to it. Which can lead to either erratic burden descent and/or slipping [1].

Removal of alkali is mostly done with the slag and is best performed at lower basicity values. Of the alkali removed over 90% is removed through the slag [5, 13]. Basicity is a concept that has several definitions depending on which compounds used to calculate it. The basis of basicity is wt% basic oxides in the slag divided by wt% acid oxides and basicity is presented as a fraction. Two basicity definitions were used in this thesis and they were B2:

B2 = CaO

SiO

(2.11)

And Bell’s ratio [18]:

Bell

s ratio = CaO + 0.69 ∗ M gO 0.93 ∗ SiO

+ 0.18 ∗ Al

O

(2.12) Several articles have investigated how the basicity would affect alkali pickup of the slag and the general consensus is that lower slag basicity will increase the amount of alkali in the slag [5, 6, 14, 15]. A problem with too low basicity is that a higher level of sulfur will stay in the iron, as the sulfur can be counter-acted by CaO present in BF slag and CaO will be lower when the basicity is lower. Therefore, an analyze have been performed to determine the limit for lowest possible basicity while keeping the iron quality under control for one plant. The limit is dependent on the BF parameters and raw material used at the specific plant. A basicity value just above or around 1 could be seen as the limit if alkali is to be removed and iron quality kept [5].

Reaction 2.7 indicates that to hinder the gasification of alkali silicates, the partial pressure of CO should be kept high. The high temperature for the reaction at 1550

C means that a lower flame temperature also could be used to hinder the reduction and gasification and thereby lower the alkali circulation [7].

Removal of alkali would require decreased re-circulation of alkali containing materials to the furnace as alkali otherwise just would be reintroduced to the BF, which has been previously suggested in a study [13].

Lowering the catalyzing effect of alkali on coke gasification could be done by coke ash additions that can bind the existing alkali in more stable forms which have been tried with certain mineral addition before [8]. As alkali diffuse through the coke a coating of the minerals addition could stabilize the alkali at the surface of the coke stopping it from degrading the inner parts of the coke.

2.1.4 Ammonia and Hydrogen Cyanide Formation in the Blast Furnace

Work performed by Turkdogan et al [19] lay the foundation for how ammonia is believed to be formed in a BF. Reaction 2.13 and 2.14 show the reactions behind the ammonia and hydrogen cyanide formation in the BF. The basic reaction is:

2(K, N a)CN + 3H

O (K, N a)

CO

+ 2N H

+ C (2.13) A second reaction occur between ammonia and carbon monoxide as follows:

N H

+ CO HCN + H

O (2.14)

The ratio between them depends on several parameters: the amount of moisture available, amount of

available KCN in the top and temperature during the reactions. The temperature threshold for NH

was

around 600

C and the NH

formation would continuously decrease exponentially until 500

C afterwards it

was not detected. Further NH

that is formed would be oxidized by either Fe

O

or CO

and the amount

of ammonia formed would decrease. Figure 2.3 show how Fe

O

or MnO

would oxidize NH

depending

on temperature. At lower temperature MnO

is a stronger oxidant and at higher temperature Fe

O

is

the stronger oxidant.

Figure 2.3 Amount NH

oxidized depending on the temperature for different reactor beds [19].

As HCN and NH

can be found in the top gas the oxidization kinetics for NH

is not fast enough to remove it completely. Which could be seen from the data from SSAB where both NH

and HCN is found [20]. The more water found in the top gas the more formation of NH

can take place according to reaction 2.13 [19]. The formation of ammonia in the BF is complex as several parameters will affect the amount formed:

• Top gas temperature

– The temperature will both affect the moisture content, lower temperature could lead to increased solubility of NH

in water and HCN is miscible in water, so presence of water could decrease its presence in the top gas [21].

– The ratio between endothermic/exothermic reactions in the BF.

• Flame temperature

– The flame temperature will have a minor effect on the amount of alkali vapor produced and the total alkali load. As a high temperature is needed to reduce the alkali silicates in to alkali gas according to reaction 2.7 which starts the alkali circulation. Lowered flame temperature thus leads to more alkali exiting the BF through the slag [7].

• Basicity

– A lower basicity would lead to higher alkali uptake in the slag, thus lower circulating alkali in the BF and less ammonia produced according to 2.13.

• Moisture content

– Less moisture introduced with the charge or through other ways to the BF would give less water for reaction 2.13 to happen.

The multitude of parameters that can affect the BF process make it a hard process to analyze if only a

few select parameters and variables are looked at using 2-dimensional graphs or simple tables. A larger

picture generally has to be formed to see possible patterns and which parameters that have a large effect

on e.g. alkali.

2.2 Multivariate Analysis

A good tool to be able to sort through possible connections from large data sets affected by several parameters is multivariate analysis. One common methods used to evaluate data graphically in multivariate analysis is Principal Component Analysis (PCA) that is useful for obtaining an overview of the data set by projecting the data on a coordinate system. A second is Partial Least Squares Projections to Latent Structure (PLS) that work in the same way but is more useful to link several variables together to see how predictor variables can give predict response variables.

The methods will be described from the work done by Eriksson et al. [22].

2.2.1 Principal Component Analysis

PCA is the first step in the process and give a quick overview if the different variables for made observations differ or have relationships among each other. PCA works by inserting the observation, often time points or similar points in a process, and variables like temperature or chemical analysis. Forming a matrix X of observations N and variables K like shown below.

X =















x

x

x

. . . x

x

x

x

. . . x

.. . .. . .. . . . . .. . x

x

x

. . . x















Modern multivariate analysis is good at handling matrices like above where the observations are followed by a lot of variables. A dataset used with PCA generally must be pre-treated to normalize the dataset, as variables that have a large internal difference would over power variables with low internal difference.

Further on they will be denoted as large or small variables. Scaling will fix this problem by shrinking the large variables and increasing the size of small variables. A standard method used is called Unit Variance Scaling (UV). The dataset is normalized by calculating the standard deviation for each K and multiplying each column K with the inverse of the standard deviation. Thus, obtaining variables with the same variance in size for the entire dataset. It is possible to also weight the scaling performed on each variable based on previous knowledge of each variable. The risk is to give variables more weight in a model than they should have, affecting the model negatively. To make the produced model easier to evaluate a second method is generally deployed as pre-treatment in the form of Mean-centering.

Mean-centering is simply performed by calculating the average for each variable and then subtracting it for each variable. The effect of normalizing a dataset is graphically represented in figure 2.4 [22].

Figure 2.4 Graphical representation of data normalization performed by SIMCA before PCA [22].

From the matrix X a space is formed where each variable forms a dimension in the space and the

observations form the points in the space. As each variable form a dimension the space will have K dimensions where the average vector point of the dataset will be in the middle of all the points. The space will afterwards be called the K-space. By using mean-centering the average vector will be in the origin of the space. Figure 2.5 show how the vector points are relocated after the mean-centering.

Figure 2.5 (left) Non mean-centered variables. (Right) Mean-centered variables with the average vector at origin in the K-space [22].

When all the variables are centered in the K-space it is necessary to transform the projection into one with less dimension. Making it easier to analyze the dataset. From the K-space a first principal component (PC1) is calculated that form a line that will represent the largest variance in the K-space. To further approximate the data a second principal component (PC2) that is orthogonal to the first and have the largest variation possible is calculated. Together the two principal components form a plane of the K-space that can easily be plotted in 2D-space with each observation projected onto it as shown in figure 2.6.

Figure 2.6 The two principal components PC1 and PC2 form the plane that all the obser- vations then can be projected onto [22].

The plot is called a score plot as each observation can be given a score from how it is located in relation to the principal components. Observations grouped together on the score plot will have similar properties.

Meaning that observations laying far apart will have less similar properties and observations located in the middle of the score plot will be average observations. A problem with the score plot is that it will not show how and which variables that affect the observations, only the relation between the observations. To see that a loading plot is calculated and used together with the score plot. The loading plot is calculated first from the angles between the variables and PC1 and then between the variables and PC2.

The variables are the axis shown in figure 2.6 in the K-space. Similar features in interpreting the loading

plot exist compared to the score plot. Variables grouped together will be correlated, variables laying

on the edge of the loading plot will affect the calculated model more, and variables on diagonal sides

of origin will affect each other inversely. Score plot and loading plot should be examined together as

loading position correlate to which observations are affected by which variables and how they are affected.

Observations in the score plot will be affected most by the variables closest in the loading plot.

The score plot have to be observed to identify any possible outliers, and also identify which outliers that can be discarded easily or which have to be taken into account for a correct model. The program SIMCA have a tool to find outliers that are not immediately visible in the score plot, called DMod, in which observations that are above a calculated critical value can be considered as moderate outliers. Those outliers should as well be further investigated for their effect on the process as they indicate a small shift in the data that could e.g. be due to natural shifting process parameters or invalid data due to measuring errors. The opposite of moderate outliers found using DMod, are strong outliers which are the one found in the score plot. The strong outliers will affect the model. Which moderate outliers will not do, they rather represent brief changes in the process. The problem with strong outliers is that they can affect how the PCA model calculated the principal components and drag the model out unnecessarily in one direction.

The model can be evaluated by R

X and Q

X. Where R

X explain the fit of the model to the dataset and Q

X explain how well the model predict variation. When looking at Q

X it should be larger than 0.5 and the difference in R

X and Q

X should maximum be around 0.2. It is hard to get both a good fit and good prediction as R

X will go to 1 as more parameters are used which can be seen in figure 2.7 [22]

while Q

X have a maximum that should be aimed for when modeling.

Figure 2.7 Increased fit R

X by increasing the number of parameters in the model have a maximum for Q

X where more variables is not better for the model [22].

2.2.2 Partial Least Squares Projections to Latent Structure

PCA is the first step in analyzing a process using multivariate analyze and shows the observations relationships with both each other and with the variables. When monitoring a process, it is of interest to see how measured variables X affect one or more variable Y called response variable in the data set over an observed time scale. PLS can be used for that type of analyze where it is needed for variables X to be able to predict the responses Y. The dataset will be pre-treated the same way as for PCA with unit variance scaling and mean centering. When using PLS however more care should be taken in how the dataset is prepared. If it is known that certain variables carry more weight in a process the dataset should be scaled accordingly as the normalization otherwise can affect the resulting model in a negative way.

The PLS model uses two spaces, meaning that all the observations have two points connected to them.

Figure 2.8 show how all observations in the space X will have a corresponding observation in the space Y.

The observations in the space X will like PCA have a first component calculated that approximates the

observations in one direction, difference being that in PLS the component will also have to correspond to

the observations in the Y-space. When using two or more responses component one will be calculated for

the X-space called t

and one for the Y-space called u

. When calculating the second component for the

X-space it will be orthogonal to the first component while the second component added to the Y-space

does not necessarily have to be orthogonal to the first component. The two new scores are called t

and

u

respectively. The second component is not necessary when using a PLS model, though often used as a

single component may not explain the spaces X and Y properly on its own.

Figure 2.8 The space X and the corresponding response variable space Y used for PLS and how they look like when using three variables X and three variables Y [22].

From the projections of the observations on to component 1 and 2 score values t

and u

is obtained for both spaces X and Y. A score plot can be produced for either score values t

or u

to see how the observations group look like. Several different components can be calculated for the spaces to see which fit the dataset in term of correlating response variables Y to factors X. To investigate if the correlation is there, a plot of equation 2.15 can be done:

u

= t

+ h

(2.15)

How close the plotted scores follow a diagonal line with slope 1 indicate the degree of correlation between X and Y. The residual distance h

should be low for a better correlation. To see which variables in X that affect the variables in Y the most a tool called Variable Importance for the Projection (VIP) can be used. VIP can be generalized as a calculation of the weights w

for the PLS model and finding the sum of squares while also considering how Y varies in the model [22].

Like PCA, PLS also have a maximum between good fit and good predictability as shown in figure 2.7. In PLS R

Y and Q

Y are investigated instead as it is more important to have a good fit and predictability for the Y variable.

OPLS stands for Orthogonal PLS. The basis is the same as for PLS with the difference being that the model is separated into two different Y. One that explains Y and several orthogonal rotations that are not explained by Y. The number of rotation will vary depending on how many are deemed necessary by SIMCA to get a fitting model. The resulting explainability and prediction is the same as for PLS with the benefit of OPLS giving more easily interpreted models.

2.3 Coke Characterization

As coke moves through the BF its crystalline properties will change and its reactivity increases with temperature. Thus, different methods have to be used that focus on characterizing coke. Methods such as chemical analysis to get the amount of alkali in the coke or TGA to see the apparent reaction rate or mass loss of the coke are used or can be used. One such method is to investigate the increase in the stacking height of the crystal structure of coke which can be calculated as a value, L

[23].The L

value aims mainly to find the relative temperature that the coke has been subjected to. Alternatively, the apparent reaction rate k

of coke can be investigated using TGA [24]. Moreover, changes in chemical composition and ash content are other important variables indicating the properties of coke collected from a BF.

2.3.1 Coke Graphitisation

When coke descends in the BF it will undergo graphitisation. Practically the coke will become more

crystalline/ordered the more heat it have been exposed to. The reaction will happen gradually showing

the thermal history of the coke through its structure. Graphitisation is also dependent on time and if catalysts like iron were present. A way to measure the graphitisation is through calculating the L

value from the XRD data using Scherrer’s equation as shown in equation 2.16. Where K is 0.89 for the coke peak (002) shown in figure 2.9, λ is X-ray wave length and is 1.5409 for copper, β is the full width half maximum value (FWHM), and Θ is the Bragg angle.

L

= K λ

β cosΘ (2.16)

FWHM is calculated as the width of the investigated peak (002) at half its maximum amplitude. A typical coke XRD diffractogram as shown in figure 2.9 further illustrate the peak (002) that will be ”sharper” the more heat the coke sample been exposed to. Which in turn would lead to a higher L

value. The figure also shows the SiO

peaks that will often be present when characterizing coke [1, 23].

(002) SiO

0 2000 4000 6000 8000 10000 12000 14000 16000

10 20 30 40 50 60 70 80 90

Co u n ts

Position °2Θ

Figure 2.9 A XRD plot showing the (002) peak and the two SiO

peaks often located alongside the peak.

2.3.2 Reaction Rate and Activation Energy

The reaction rate of coke with CO

(g) will change depending on the catalyst present in the coke, increases with more catalyst present. The type of catalyst can vary e.g. alkali compounds, iron compounds or calcium compounds [1, 8, 24]. Increased reactivity of coke can be measured as an increase in apparent reaction rate. Apparent reaction rate must be calculated over the temperature range when the gasification start as that is the point when the reaction would be chemically controlled and not diffusion controlled.

The apparent reaction rate was expressed as k

and was calculated using equation 2.17:

k

= 1 W

dW

dt (2.17)

W is the coke sample weight at time t and reaction rate was expressed unit wise as [g g

s

] as the reaction was assumed to be of first order. To find the activation energy E

, Arrhenius equation 2.18 can be used along equation 2.17:

k

= Ae

(2.18)

As the plot of the natural logarithm of k against 1/T in kelvin give A at the intercept of the line on the X

axis in the form of ln(A) and the slope of the line will be −E

/R [24].

3 Method

The following text describe the methods deployed to obtain the results for this thesis. The methods used for the coke samples preparation was the same as a previous thesis done in ALCIRC so that the results for the coke could be compared directly to data from that thesis that was performed during the 31st and 32nd EBF campaign [8].

3.1 Basket Samples in EBF

Coke basket samples had been previously prepared at Swerea MEFOS from steel mesh/wire and added to the 33rd campaign of LKAB’s EBF which was then excavated by LKAB. [25].

3.1.1 Preparation of Coated Coke and Coke With Modified Ash

Three types of coke with kaolin was prepared at MEFOS alongside a reference coke. The composition of the coke used in the experiments can be seen in table 3.1. The test and reference coke were produced by DMT Gmbh & Co. KG to match coke produced by SSAB, the coated coke was prepared from the reference coke using the same type of kaolin. Coke samples consisted of a mix of coke sized <19 mm and >22.5 mmCoke samples CC1 and CC2 got a coating in slurry consisting of 23 and 33 wt% of kaolin mixed with water respectively. The slurry was prepared by weighing water and mixing in kaolin to get the correct weight percent in the slurry. After all kaolin had been incorporated in to the slurry, coke was placed between two handheld screens, and lowered into the slurry. The screens ensured that no coke was lost in the slurry, that the slurry could reach the coke properly, and the top screen hindered coke from floating up and not get coated completely. Mesh size on the bottom screen was 16 mm. Care was taken when coating the coke not to pack it too tight in the screen, as contact between coke pieces could hinder the kaolin coating to cover the entirety of the pieces [25].

Table 3.1 Types of coke added to the EBF in campaign 33.

Kaolin

Added wt% Coating wt%

Reference coke RC - -

Test coke TC 2 -

Coating 1 CC1 - 23

Coating 2 CC2 - 33

3.1.2 Basket Samples in the EBF

The steel wire baskets were connected two and two with a divider in the middle of each basket, separating reference and test coke. All basket had RC in the top half and the type of experimental cokes in each basket varied. Basket K1 and K5 contained TC, basket K2 contained CC1, and basket K6 contained CC2.

The baskets found during excavation of the EBF is shown in table 3.2 and layer indicate in which coke charge layer that the baskets where added in. Layer 20 samples where found at a depth below charging between 3.6 to 3.7 m and layer 28 samples at 4.25 m below charging.

Table 3.2 All baskets and in which layers they were added in the EBF. Baskets found during excavation and which coke types that were in the baskets.

Basket

Layer K1 K2 K5 K6

N20 RC/TC - RC/TC RC/-

N24 - - - -

N28 RC/TC - - -

N32 - - - -

Basket K6 in layer N20 only had the top part that contained RC left, the rest of the basket had melted of

during the campaign. The campaign only gave a total of 7 coke samples from layer N20 and N28. None of the coated coke samples were found during the EBF excavation.

3.1.3 Characterization of Samples

Several different methods were used to characterize the samples and to get data comparable to previous coke characterization performed at Swerea MEFOS and at LTU.

3.1.3.1 XRF From the coke baskets samples an approximate of 45 g was prepared by grinding it to a homogeneous powder using a ring mill for 30 seconds or until a powder was formed. 30 g of the samples were sent to SSAB Lule˚ a for chemical analysis using XRF, the X-ray source was Rh based. The other part of the coke samples was used in the XRD analysis.

3.1.3.2 SEM-EDS SEM analysis was done at LTU to see how the alkali was distributed inside the coke samples using a Zeiss Gemini Merlin SEM equipped with a X-Max EDS from Oxford Instruments.

Samples were prepared by mounting them in resin, polishing them in several steps and finally coating them with carbon.

3.1.3.3 XRD XRD analysis was performed at LTU using a PANalytical Empyrean XRD unit on powdered coke samples to see which phases and crystalline compounds that was present in the samples.

The XRD used the following conditions: Cu Kα radiation, electron emission current 40 mA, accelerating voltage 45 kV, measurement in the 2Θ range 10-90

and using a step size of 0.0260

. The L

analysis was performed using an in house developed program at MEFOS that would automatically calculate the L

-value from the diffractograms and produced a figure so that it was possible to see that the analysis had been performed correctly by the program. The L

-value was recalculated for Olofsson’s [8] results as a new method was used in this report so that the results would be comparable.

3.1.3.4 TGA The two coke basket samples recovered from the lowest part of the EBF were analyzed with TGA as highest potassium contents was expected on coke further down in the BF. Coke was crushed using a jaw crusher and 1 g of the fraction between 1-2 mm was extracted using a pair of sieves. The sample tested was placed in an alumina crucible and the atmosphere in the TG was CO

. CO

was continuously supplied at a rate of 300 ml/min the entire time the samples were in the TG. The CO

was added as the TG trials are designed to see when the Boudouard reaction 2.6 happened and in extension how alkali would affect the reactivity. Excess CO

was therefore kept in the TG so that the reaction would not be hampered by lack of CO

. The TG was first heated up to 600

C at 20

C/min, the heating rate was then lowered to 3

C/min between 600

C to 1200

C, after which the heating was turned off. From TGA apparent reaction rate and activation energy could be calculated using the equations specified in section 2.3.2. Start of reaction was defined as the point where mass started decreasing steadily for the samples and was determined visually from the figures as that was the method used previously [8].

3.2 Correlation of Ammonia and Cyanide in the Top Gas 3.2.1 Top Gas Measurements

Sampling of data for the top gas composition was done at SSAB in Oxel¨ osund’s BF No.4 and analyzed using a mass spectrometer (MS). Further data was provided by SSAB on process parameters and gas composition like CO, H

and CO

collected during that time period. A probe was inserted into one of four exhaust pipes at the top of the BF and the gas was extracted approximately at the same height as the burden feeding system. BF No.4 is a low-pressure BF, with a gauge pressure of 50-100 mBar above atmosphere pressure in the top gas, facilitating the need to use a pump to feed the gas to the MS. The gas was transported from the probe to the MS through heated pipes to avoid condensation in the system as both NH

and HCN could dissolve in liquid water [21]. The entire system including the pump was heated. Before the gas was fed to the MS it entered a sample preparation unit from which a probe was inserted that could collect the small amount of gas the MS needed. The unused gas then exited on the other side of the sample collection unit as shown in figure 3.1.

The box was there to clean the gas from particles that followed with the BF gas probe so that it did not

Figure 3.1 The sampling box used to extract gas to the MS from the gas flow and to clean the gas from the particles

plug the MS to fast. Periodically a back-blow procedure was performed to clean out the probe collecting the gas from the BF, the interval was gradually increased as the measuring process continued. The back-blow can be seen as dips down towards 0 each time it was performed in the data graphs produced later on. The MS used was a V&F Analyse- und Messtechnik GmbH Airsense Compact (newer models use the name ”Combisense”). The MS uses a combination of ion-molecule reaction ionization (IMR-MS) and electron-ionization mass spectrometer (EI-MS) to measure the gas composition. The benefit of using the IMR-MS part of the unit was that less fragmentation of the measured molecules could be achieved giving less overlapping peaks. Making it easier to separate species with similar molar mass or properties.

The samples entering the MS were ionized by first ionizing a carrying gas like Kr

or Xe

by bombarding it with electrons. The carrying gas then get transported through an electrical field and the sample gas was introduced and allowed to interact with the carrying gas. Leading to ionization of the sample gases that then could be transported through another electrical field to a Channeltron detector [26].

3.2.2 SIMCA

A dataset was prepared from the process data by averaging the data before taping and correlating it to the tap time. PCA and PLS analysis was then used to find correlations between the variables and the observations in the dataset as described in the theory. SIMCA 15 was used.

The data sets were prepared by removing all the observations when the measured MS values for NH

and

HCN were below zero and the observations when the blast flow was below 80 kNm

/h was removed as the

BF was then not in active production. Analysis was performed using data sets from MS measurements,

minute data and tap data using the parameters above.

4 Results

4.1 Basket Samples From EBF 4.1.1 Basket Location

Excavation protocols from the EBF was provided by LKAB and showed the positioning of the baskets as they were found in the EBF. In figure 4.1 the positions are visualized. The most central basket was N20K6 and it had no bottom part as it had melted due to high heat where it had been positioned. The basket N28K1 was the sample found furthest away from the center of the EBF. Figure 8.1 in the appendix showed how the temperature was in the EBF and indicated that at the lower probe (approximately 3.3 m below charging depth) the gas flow had been in the middle. The upper shaft probe (approximately 1 m below charging depth) had a temperature profile that spread out and peaked to the sides more. Shaft probe height taken from [1].

N20 N20 N28

Figure 4.1 Location of the four baskets found in the EBF during the excavation.

No basket was found in the left part of the EBF close to F3 in figure 4.1. Meaning that the samples did not represent the entire EBF and its condition. None of the coated samples were found during the excavation so no results could be extracted from them.

4.1.2 XRF

The basic chemical analysis of the samples was done using XRF and are presented in table 4.1. Results showed that components P

O

, S, MgO and TiO

was similar independently of sample and location found. Fe content was similar in all coke samples except for coke in basket 6 sample RC (reference coke), where it was approximately twice as large compared to the other samples. CaO was higher in RC samples compared to in the TC (test coke) samples. Further, SiO

was lower in RC samples and higher in TC samples. The reference coke that had not been in the EBF differed most from the others by having the lowest K

O content. It was otherwise fairly similar to the other coke samples.

The data of specific interest in this study is the alkali content in the samples. Alkali is plotted in figure 4.2 to show how K

O and Na

O varied in the samples.

Figure 4.2 showed that when the two RC samples N20K5 and N28K1 was compared the amount of K

O

differed by 0.01 percent units. The same test cokes TC had a difference of 0.13 percent points showing

that the TC samples differed more in alkali indicating exposure to different gas compositions. TC had

the highest uptake of both K

O and Na

O for all comparable samples. Figure 4.2 also showed that

approximately 4-5 times more K

O than Na

O was found in the coke samples.

Table 4.1 XRF data for the coke samples retrieved from the EBF and original coke.

Fe CaO SiO

MnO P

O

S Al

O

MgO Na

O K

O TiO

Sum

Basket [%] [%] [%] [%] [%] [%] [%] [%] [%] [%] [%] [%]

Layer 20 (Depth below charging 3.6 to 3.7 m)

Ref. coke 0.31 0.00 6.55 0.02 0.03 0.71 3.17 0.06 0.05 0.15 0.18 11.00 K1 RC 0.23 0.22 6.16 0.04 0.03 0.68 2.97 0.07 0.14 0.71 0.17 11.18 K1 TC 0.21 0.04 7.46 0.05 0.03 0.63 3.93 0.10 0.23 1.14 0.15 13.73 K5 RC 0.29 0.02 6.69 0.05 0.02 0.69 3.16 0.07 0.17 1.01 0.17 12.12 K5 TC 0.23 0.01 7.39 0.07 0.02 0.64 3.84 0.10 0.25 1.32 0.15 13.80 K6 RC 0.52 0.12 6.51 0.11 0.03 0.69 3.14 0.09 0.29 1.49 0.16 13.01

Layer 28 (Depth below charging 4.25 m)

K1 RC 0.29 0.07 6.60 0.14 0.03 0.69 3.10 0.09 0.20 0.99 0.17 12.14 K1 TC 0.24 0.03 7.39 0.18 0.02 0.65 3.83 0.12 0.29 1.15 0.16 13.82

Figure 4.2 XRF results for total alkali found with XRF.

In figure 4.3 the sum of total alkali showed that no clear correlation could be seen between which layer the baskets were found at and total alkali in the coke. The TC samples in basket N28K1 had an alkali level in between the two other TC samples that was found further up the EBF. Two TC samples also had similar alkali levels despite being found at different depths. Basket sample N20K6 RC found closest to the middle of the EBF had the highest total alkali content.

The figure shows the previously mentioned trends that coke with kaolin added (TC) had higher alkali

content compared to the RC samples. N20K6 had the overall highest alkali content. N28K1 and N20K5

had similar alkali contents even though they were found on different levels in the EBF. Indicating that

horizontal positioning mattered more than vertical i.e. the temperature exposure and gas composition of

the EBF mattered more than position. The temperature probes indicated that the heat profile had been

central above the height where the samples were found and the melting of basket N20K6 found in the

middle indicated that the highest gas flow was at the center of the EBF. During the previous campaign 32 the temperature profile in the EBF had been flat [8].

Figure 4.3 Alkali content for the different baskets depending on exact height found.

A comparison with previous results produced at Swerea MEFOS [8] showed more clearly that baskets found further down in the EBF had had higher alkali levels. The results presented here shows that different gas composition and heat exposure had significant effect on the alkali levels in the baskets.

4.1.3 SEM

From the SEM results a visual overview on how the different phases was present in the coke samples could be achieved. The SEM pictures were complemented with EDS analysis to get the atomic ratios at points of interest such as particles or in the middle of the coke matrix.

The kaolin and reference samples where similar in structure with a porous coke matrix with particle grains spread out in the matrix along occasional larger particles, and larger clusters of particles. The differences that could be seen in figure 4.4 was that the TC sample had slightly more of smaller particles (coke ash) in the coke matrix at the area investigated. Both samples had the presence of larger particles in the size range of 50 microns in all areas investigated.

Figure 4.4 (Left) Coke basket N20K1-T with RC. (Right) Coke basket N20K1-B with TC.

Further comparing the N28K1 basket’s top (RC) and bottom (TC) part in figure 4.5 showed the same results in structure as shown in previous figure 4.4. The coke had a porous structure with particle grains spread throughout. The N28K1 TC sample (2 wt% kaolin) in the right part of figure 4.5 showed an example of how larger particles appeared in the sample. The left particle labeled A in the sample was identified as a phase corresponding to KAl

Si

O

, also known as Potassium mica. Which is formed from dehydroxylation of muscovite at approximately 780

C according to [27]. The right particle labeled B was identified as a phase corresponding to (K,Na)AlSi

O

which is known as K-feldspar and specifically in the form of Sanidine due to the high heat and that it has been found before in a BF [28, 8].

Figure 4.5 (Left) Coke basket N28K1-T with RC. (Right) Coke basket N28K1-B with TC.

Reference coke that had not been through the EBF was investigated along the other samples to get a reference of how untreated coke’s structure and composition was. The sample was named 1R<19. The results are presented in figure 4.6. The sample was similar in structure when compared to the other samples that had been in the EBF. In figure 4.6 the presence of a larger particle structure in the right part of the RC coke sample was seen that was not present in the 1R<19 sample. The red shape marks the mentioned structure.

Figure 4.6 (Left) Reference coke 1R<19 (Right) Coke basket N28K1-T with RC coke for comparison.

The phases identified in the sample were summarized in table 4.2 and if possible matched with phases

with corresponding composition. The results showed that the phase corresponding to Quartz, excluding

the coke matrix, could be found in all samples. The particle identified with the composition Al

Si

O

was

found in all samples except for N28K1 TC and was identified in literature to correspond to the phase

Metakaolin [29]. Smaller amounts of the composition matching the phase Kalsilite was found in two

different coke samples, N20K1 RC and in N28K1 TC [30]. Three potassium containing compositions

was found in all the samples that had been in the EBF, those where corresponding to the phases Leucite,

Dehydroxylated muscovite, and Sanidine [24, 30, 28]. The most common phases that could be

identified by EDS was Dehydroxylated muscovite, followed by Leucite and finally Sanidine.

Table 4.2 Identified compounds in the coke samples using EDS.

N20K1-T N20K1-B N28K1-T N28K1-B 1R<19

Compound RC TC RC TC Ref

SiO

- Quartz X X X X X

Al

Si

O

- Metakaolin X X X X

(K,Na)AlSiO

- Kalsilite X X

KAlSi

O

- Leucite X X X X

(K,Na)AlSi

O

- Sanidine X X X X

KAl

Si

O

- Dehydroxylated muscovite X X X X

Fe

(Si,Al)

O

X

The Carbon matrix had no clear phase and differed from place to place investigated. The base composition consisted of a high carbon content with a sulfur content that varied between spectrum points. The carbon matrix in the reference sample 1R<19 had a lower ratio of potassium compared to the four other coke samples, and no sodium was found in the sample. The samples also contained Si and Al in different ratios without having any clear compounds that would show up as white points in the SEM pictures.

An average of the data set used for the compositions and phases can be seen in appendix A in table 8.3, 8.4 and 8.5 along all SEM images in appendix B. Comparing the results with the one previously produced in ALCIRC [8], showed similar results both in the structure inside the coke samples and some of the phases that could be found in the coke. The phases K-feldspar(Sanidine), Leucite, Quartz and Kalsilite was found in both this work and in Olofsson’s [8]. No clear connection between the coke samples that had kaolin added could be found. All samples that had been through the EBF had formed some form of potassium alumina silicate.

4.1.4 XRD

The XRD diffractograms produced for this thesis followed the typical coke diffractogram generally seen in

XRD results [1, 8, 12, 31]. A broad peak present at 26

is indicating that the coke was of semi-amorphous

structure. Figure 4.7 shows the normalized XRD diffractograms plotted for comparison. The diffractograms

are similar, and no major difference could be spotted which was expected as all samples are made from

the same coke with the composition only differing by 2 wt% kaolin. Only trend that could be seen was

that the samples that had been found further down the EBF had a higher (002) peak. All coke had been

sampled from the EBF.

Figure 4.7 Normalized XRD diffractogram results for all coke samples. -T=RC and -B=TC The mineral phases that was identified from the XRD are summarized in table 4.3. The crystal phases found with XRD correlated to some of the possible phases found with EDS. The results were separated into tables, one where the results showed compounds identified with XRD and SEM. The second table shows minerals found only with XRD. Two types of feldspar was found with XRD, ”Feldspar, alkaline”

and Albite. The first where most of the alkaline contribution was from potassium, and the later where only sodium is present. The results further illustrated that some type of alumina silicate with different levels of alkali would be found in the coke ash after it has been through the EBF.

Table 4.3 XRD identified minerals in the coke samples that were analyzed with SEM-EDS.

-T samples where RC coke and -B samples where TC coke

N20K1-T N20K1-B N28K1-T N28K1-B

Compound RC TC RC TC

C - Graphite X X X X

SiO

- Quartz X X X X

KAlSi

O

- Leucite X X X

Al

Si

O

- Mullite X X

Al

SiO

- Sillimanite X X

NaAlSi

O

- Albite X

K

Na

AlSi

O

- Feldspar, alkaline X

The samples that had not been investigated using SEM-EDS was instead investigated in the XRD. The

results were summarized in table 4.4. Two different types of Mullite was found and both Leucite and

Feldspar was present in the samples.