Fouling in Biomass Fired Boilers

Home , Fouling

Mälardalen University Press Dissertations No. 116

FOULING IN BIOMASS FIRED BOILERS

Jan Sandberg

2011

School of Sustainable Development of Society and Technology Mälardalen University Press Dissertations No. 116

FOULING IN BIOMASS FIRED BOILERS

Jan Sandberg

Akademisk avhandling

Fakultetsopponent: Professor Alain Liné, LISBP-INSA de Toulouse

Akademin för hållbar samhälls- och teknikutveckling Mälardalen University Press Dissertations No. 116

Mälardalen University Press Dissertations No. 116

FOULING IN BIOMASS FIRED BOILERS

Jan Sandberg FOULING IN BIOMASS FIRED BOILERS

Jan Sandberg Akademisk avhandling

som för avläggande av teknologie doktorsexamen i energi- och miljöteknik vid Akademin för hållbar samhälls- och teknikutveckling kommer att offentligen försvaras tisdagen den 20 december 2011, 13.30 i Delta, Mälardalens högskola, Västerås. Akademisk avhandling Fakultetsopponent: Professor Alain Liné, LISBP-INSA de Toulouse som för avläggande av teknologie doktorsexamen i energi- och miljöteknik vid Akademin för hållbar samhälls- och teknikutveckling kommer att offentligen försvaras tisdagen den 20 december 2011, 13.30 i Delta, Mälardalens högskola, Västerås.

Fakultetsopponent: Professor Alain Liné, LISBP-INSA de Toulouse

Akademin för hållbar samhälls- och teknikutveckling

Summary Abstract This thesis describes a detailed investigation into fouling in biomass fired boilers according to fuel mixture, combustion conditions, transportation of particles by the flue gas and the probability of particles impinging and sticking onto heat transfer tubes. The effects of fouling on overall boiler performance and the efficacy of soot blowing are also investigated. Both theoretical simulations and practical experiments on a 157 MW circulating fluidized bed boiler are presented. The deposit thickness on and around a heat exchanger tube is shown to be mainly dependent on the The use of biomass as fuel for boilers is promoted as a way of reducing the ash particle size, as particles larger than 10 µm (Stokes number larger than 0.1) mainly impinge on the discharge of the greenhouse gas CO2. In comparison to boilers fired with windward side of tubes. The study also shows that fuel containing small amounts of chlorine and zinc – common elements in recycled wood – may cause both higher deposit growth rates and rapid increases coal and oil, biomass-fired boilers are associated with increased complica- in corrosion rates. These elements (chlorine and zinc), together with alkali metals from the biomass have tions related to fouling and corrosion on the heat transfer surfaces. After the potential to form sticky compounds that increase the deposit growth rate. combustion, unburnt inorganic matter in the vapour state, melts and solid Reducing deposits by soot blowing is very effective at removing loose deposits but the hard sintered particles are transported in the flue gas and can form deposits on heat trans- part of the deposits is almost unaffected. The use of recycled wood has a larger impact on the deposit fer surfaces. growth rate than the soot blowing interval. In order to understand the build up of deposits and how it affects boiler Numerical simulations show that deposits on the superheater tubes redistribute the heat transfer rate performance, the entire combustion and transport process has been ana- from the superheaters to reheater 1 and partially redistribute turbine power from the high pressure turbine to the intermediate pressure turbine lysed. The analysis includes aspects such as boiler design, fuel properties, combustion conditions, particle transport phenomena and the probability of particles adhering to heat transfer tubes. This thesis describes numerical simulation of particle trajectories, measurements of deposits using a specially designed deposit probe and a seven year investigation of on-site measurements of deposit depth and corrosion on the superheater tubes in a circulating fluidized bed (CFB) boiler in Väs- terås, Sweden. The deposit growth rate and the corrosion rate are correlated

to the fuel used during the investigation period. These deposit measurements are also used as inputs to a dynamic model that simulates how deposits affect boiler performance. Numerical simulations of particle trajectories in the vicinity of two superheater tubes were conducted in an Eulerian-Lagrangian mode that considered the flue gas and ash particle phase. Particle impingements on the

tubes were investigated for different particle sizes. The results from the particle trajectory simulations showed that particles larger than 10 µm (Stokes number larger than 0.1) mainly impinge on the windward side of the first tube, but they also impinge on the sides of the second tube in the flue gas flow direction. In theory, two tubes can merge together due to this build up of deposits. This effect has been observed during boiler operations. Smaller particles are usually more dispersed due to turbulence and thermophorectic forces, resulting in a more even distribution on the surface of the tubes. Probe measurements reveal that the deposit layer growth rate has a significant temperature and time dependence. After the initial deposit buildup a

ISBN 978-91-7485-047-5 ISSN 1651-4238

Summary

The use of biomass as fuel for boilers is promoted as a way of reducing the discharge of the greenhouse gas CO2. In comparison to boilers fired with coal and oil, biomass-fired boilers are associated with increased complications related to fouling and corrosion on the heat transfer surfaces. After combustion, unburnt inorganic matter in the vapour state, melts and solid particles are transported in the flue gas and can form deposits on heat transfer surfaces. In order to understand the build up of deposits and how it affects boiler performance, the entire combustion and transport process has been analysed. The analysis includes aspects such as boiler design, fuel properties, combustion conditions, particle transport phenomena and the probability of particles adhering to heat transfer tubes. This thesis describes numerical simulation of particle trajectories, measurements of deposits using a specially designed deposit probe and a seven year investigation of on-site measurements of deposit depth and corrosion on the superheater tubes in a circulating fluidized bed (CFB) boiler in Väs- terås, Sweden. The deposit growth rate and the corrosion rate are correlated to the fuel used during the investigation period. These deposit measurements are also used as inputs to a dynamic model that simulates how deposits affect boiler performance. Numerical simulations of particle trajectories in the vicinity of two superheater tubes were conducted in an Eulerian-Lagrangian mode that considered the flue gas and ash particle phase. Particle impingements on the tubes were investigated for different particle sizes. The results from the particle trajectory simulations showed that particles larger than 10 µm (Stokes number larger than 0.1) mainly impinge on the windward side of the first tube, but they also impinge on the sides of the second tube in the flue gas flow direction. In theory, two tubes can merge together due to this build up of deposits. This effect has been observed during boiler operations. Smaller particles are usually more dispersed due to turbulence and thermophorectic forces, resulting in a more even distribution on the surface of the tubes. Probe measurements reveal that the deposit layer growth rate has a significant temperature and time dependence. After the initial deposit buildup a

sintering process occurs. This is also shown to be dependent on temperature and exposure time. Sammanfattning Soot blowing is the most common method of reducing the effects of deposits on the heat transfer tubes. The efficacy of soot blowing is therefore also investigated in this thesis. Soot blowing shows a strong positive effect on the heat transfer rate shortly after a soot blowing cycle is completed. This positive effect is much weaker after an interval of three years. This is because soot blowing mostly removes the loose component of the deposit material leaving the hard sintered deposits unaffected. The use of recycled För att minska utsläppen av växthusgasen CO2 från fossileldade pannor er- wood has a larger impact on the deposit growth rate than the soot blowing sätts dessa idag i ökande grad av biobränsleeldade pannor. Jämfört med kol interval. och olja har dock förbränning av biobränsle medfört en ökad andel kompli- The seven year investigation of deposits and corrosion shows that the de- kationer vad gäller beläggningar och korrosion på värmeöverförande ytor. posit growth rate on superheater tubes approximately doubled when the Efter förbränning av biobränsle följer oförbrända oorganiskt material med recycled wood content of the fuel was increased to 10–15%. Small amounts rökgaserna i form av förångat material, smält material eller som fasta ask- of chlorine and zinc were found both in the recycled wood and in the depos- partiklar. Dessa ämnen kan fastna på värmeöverförande ytor och bilda it layer. These elements, together with alkali metals from the biomass have beläggningar. the potential to form sticky compounds that increase the deposit growth rate. För att förstå hur beläggningar bildas och hur beläggningar påverkar en The corrosion rate of the superheater tubes varied over the study period. A pannas prestanda behövs både kunskaper om driftsdata och pannors kon- number of possible explanations for this phenomenon are discussed in the struktion. Analyser behövs också av bränsleinnehåll, förbrännings- thesis. förutsättningar, rökgasflöden, sotpartiklars egenskaper m.m. The dynamic simulations of the boiler performance show that fouling on I denna avhandling behandlas och redovisas både numeriska simuleringar superheaters redistributes the heat transfer rate from the superheaters to re- av partikelbanor, och hur beläggningar påverkar en pannas prestanda liksom heater 1 and partially redistributes turbine power from the high pressure en lång rad mätningar av beläggningstjocklekar, beläggningars kemiska turbine to the intermediate pressure turbine. When the boiler is running at sammansättning, bränslesammansättning, korrosionsangrepp m.m. Mätning maximum load, water injection after reheater 1 has to increase to maintain med en beläggningssond har genomförts liksom studier av hur värmeöverfö- temperatures below the permitted limit. The dynamic effects of fouling are ring i en överhettare påverkas av beläggningar under en tidsperiod av sju år. small and the total efficiency of the boiler is only marginally affected as En biobränsleeldad panna på 157 MW, konstruerad med en cirkulerande long as the fouling is restricted to the superheaters. fluidiserande bädd (CFB), har använts som studieobjekt (Panna 5 i Väs- The study of deposit buildup on superheater tubes in large scale boilers terås). Beläggningar på överhettartuber har korrelerats mot bränsle- involves multi-disciplinary knowledge and historically this research has sammansättning. Dessa observerade beläggningar har också använts som mostly been conducted by performing measurements and experimenting on indata till en dynamisk simuleringsmodell av pannan. operating plants. Theoretically-based online simulations are likely to play Stokastiska beräkningar av sannolikheter för att askpartiklar av olika bigger roles in research on deposit buildup and how it affects boiler perfor- storlek ska träffa en tubyta har genomförts. Resultaten från simuleringarna mance in the future. visar att partiklar större än ca 10 mikrometer (Stokes tal större än 0.1) i hu- vudsak kommer att träffa den första tubens vindsida, i rökgasen flödesriktning, men partiklar kommer även att träffa sidorna av efterföljande tuber i flödesriktning. Genom verkliga observationer kan man se att om be- läggningarna är omfattande kan tubernas beläggningar med tiden växa ihop. Mindre partiklar (mindre än 1 mikrometer) påverkas mer av turbulens och ger en jämnare fördelning av träffytan runt hela tubernas omkrets. Mätning- ar med en beläggningssond visar också att beläggningar med tiden

omvandlas från en porös beläggning till en hård, sintrad beläggning. Denna process är både tid- och temperaturberoende. Acknowledgements Sotblåsning med högtrycksånga är en vanlig metod för att minska be- läggningar på överhettartuber. Studier i denna avhandling visar att sot- blåsningar har en mycket positiv inverkan på värmeöverföringen om man studerar timmarna före och efter att en sotblåsningscykel genomförts. Denna positiva effekt är mycket svagare när man studerar en tidsperiod av tre år. Detta beror på att sotblåsning mest avlägsnar lösa sotpartiklar och lämnar This thesis has been carried out at the School of Sustainable Development den hårda sintrade beläggningen till synes opåverkad. Bränslesammansätt- of Society and Technology, Mälardalen University, in collaboration with ningen verkar vara viktigare än antal sotblåsningar per dygn. Mälarenergi, Västerås. I would like to express my gratitude to: En sju år lång undersökningsperiod av beläggningstillväxt och bränsle- sammansättning visar att tillväxttakten av beläggningar på överhettartuber My supervisors Prof. Rebei Bel Fdhila, Prof. Erik Dahlquist and Prof. fördubblas när 10–15% returflis blandades med övrigt biobränsle. Små Lars Wester for inspiration and guidelines. mängder klor och zink påträffades både i returflisen och i beläggningarna. Klor och alkalimetaller i biomassa liksom klor och zink, har potential att My colleague Dr. Ulf Sand for good cooperation during many years. bilda klibbiga föreningar som kan ökar beläggningshastigheten. Även kor- rosionshastigheten på tuber har studerats och varierar över dessa sju år. Ett My colleagues Dr. Christer Karlsson and Anders Avelin for good coop- antal möjliga förklaringar till denna variation diskuteras i avhandlingen. eration in topics covered in the thesis. En dynamisk simuleringsmodell av pannan har utvecklats. Simuleringar av olika grad av beläggningar på olika delar av överhettartuber visar att be- All the people at Mälarenergi Västerås, especially Peter Karlsson, Ingrid läggningar omfördelar överförd värmeeffekt från överhettare till Byström, Erik Holmen, Evert Lundqvist, Einar Port and Jens Moberg for mellanöverhettare 1 och delvis omfördelar utvecklad effekt från högtrycks- supporting me during the measurements on Boiler 5. turbinen till mellantrycksturbinen. Om pannan körs på maximal värmeeffekt behövs en ökad vatteninsprutning i ångledningen efter mellanöverhettare 1 All my colleagues at Mälardalen University, especially Björn Widarsson för att begränsa ångtemperaturen inom tillåtna gränser. De dynamiska effek- for helping me with the gathering of measured data from the supervision terna på pannan, av beläggningstillväxt, är små så länge beläggningar är system of Boiler 5, as well as Christina Ingwall-Johansson for helping koncentrerade till överhettare och den totala pannverkningsgraden påverkas me with chemical analysis. endast marginellt. Former students at Mälardalen university Lennart Eriksson and Börje Gästlöf for helping me with measurements and conclusions.

Jenny Larfeldt at TPS, Studsvik for supplying experimental equipments.

Anders Nordin at Umeå University for conducting chemical analysis on the tube deposit material.

Marcus Slotte at Foster Wheeler for supporting me with design data of Boiler 5.

And finally all my friends and family for always supporting me!

Jenny Larfeldt at TPS, Studsvik for supplying experimental equipments.

Anders Nordin at Umeå University for conducting chemical analysis on the tube deposit material.

Marcus Slotte at Foster Wheeler for supporting me with design data of Boiler 5.

And finally all my friends and family for always supporting me!

vective flow field. Scandinavian simulation society SIMS 2003, Septem- List of appended papers ber 18–19, Mälardalens Högskola, Västerås.

Sand U., Sandberg J. & Bel Fdhila R. (2006). A two-phase transport model for the pyrolysis process of a vertical dry wood cylinder exposed to thermal radiation, including the surrounding flow field. International Journal of Green Energy, Volume 3, Issue 1, p. 63–78.

This thesis is based on the following publications: Bel Fdhila R., Sand U., Sandberg J. & Larfeldt J. (2008). Numerical prediction of the transport and pyrolysis in the interior and surrounding of Sandberg J., Sand U. & Bel Fdhila R. (2006). Measurements, theories dry and wet wood log, Applied Energy 85 1208–1224. and simulations of particle deposits on superheater tubes in a CFB biomass boiler, International Journal of Green Energy, 3:43–61.

Sandberg J., Sand U. & Bel Fdhila R. (2006). Long time investigation of the effect of fouling on the superheaters in a circulating fluidized biomass boiler. International Journal of Energy Research 2006; 30:1037–1053.

Sandberg J., Sand U. & Bel Fdhila R. (2002). Numerical simulation of fouling on superheater tube walls, Proceedings of the 10th workshop on two-phase flow predictions, Merseburg, April 9–12. ISBN 3-86010-641- 4.

Sandberg J., Karlsson C. & Bel Fdhila R. (2011). A 7 year long measurement period investigating the correlation of corrosion, deposit and fuel in a biomass fired circulated fluidized bed boiler. Applied Energy 88 (2011) 99–110.

Sandberg J., Bel Fdhila R., Dahlqvist E. & Avelin A. (2011). Dynamic simulation of fouling in a circulating fluidized biomass fired boiler. Ap- plied Energy 88 (2011) 1813–1824.

The author has also participated in the following publications:

Sand U., Sandberg J. & Bel Fdhila R. (2002). Modelling of gas-solid fluid dynamics and pyrolysis in a biomass-fired municipal CFB boiler. Proceedings of the 10th workshop on two-phase flow predictions, Merse- burg, April 9–12. ISBN 3-86010-641-4.

Sand U., Sandberg J. & Bel Fdhila R. (2003). Simulation of volatile gas release from a small dry wood particle undergoing pyrolysis in a hot con-

vective flow field. Scandinavian simulation society SIMS 2003, Septem- List of appended papers ber 18–19, Mälardalens Högskola, Västerås.

Sandberg J., Bel Fdhila R., Dahlqvist E. & Avelin A. (2011). Dynamic simulation of fouling in a circulating fluidized biomass fired boiler. Ap- plied Energy 88 (2011) 1813–1824.

The author has also participated in the following publications:

Sand U., Sandberg J. & Bel Fdhila R. (2003). Simulation of volatile gas release from a small dry wood particle undergoing pyrolysis in a hot con-

g Gravity acceleration [m s-2] Nomenclature h Heat transfer coefficient [W m-2 K-1] h Enthalpy [kJ kg-1] H Heating value for fuel [kJ kg-1] k Thermal conductivity coefficient [W m-1 K-1] k Specific turbulent kinetic energy [m2 s-2] -1 ks Calibration constant for steam flow [kg s ]/[MPa] Kn Knudsen number [---] Symbol Meaning Unit K Ratio of fluid to particle thermal conduc- [---] A Area [m2] tivity -1 A Le Dissipation length scale [m] sinter Pre exponential factor for sintering [s ] M Mass [kg] C Turbulent dissipation coefficient [---] µ m Mass [kg] [---] -1 Cte Eddy lifetime coefficient m Mass flow rate [kg s ] C Turbulent dissipation factor [---] Nu Nusselts number [---] -1 P Turbine power [W] C Sintering rate [s ] sinter Pr Prandtl number [---] [---] CD Drag coefficient p Pressure [Pa] [---] Q Heat transfer rate [W] CC Cunningham correction coefficient R Specific gas constant [J kg-1 K-1] [---] Cm Numerical factor (kinetic gas theory) Re Reynolds number [---] [---] -1 -1 Cs Numerical factor (kinetic gas theory) s Entropy [kJ kg K ] [---] St Stokes number [---] Ct Numerical factor (kinetic gas theory) -1 -1 T Temperature [K] cp Specific heat at constant pressure [kJ kg K ] -1 -1 ∆T Temperature difference [K] cv Specific heat at constant volume [kJ kg K ] d Diameter [m] t Time [s] -2 -1 D Diameter [m] U Overall heat transfer coefficient [W m K ] -1 D Thermophoretic coefficient [N m] u Local velocity [m s ] T -1 -1 Mean velocity, x-direction [m s ] E Activation energy [kJ kg ] u e Internal energy [kJ kg-1] u′ Fluctuating local velocity, x-direction [m s-1] -1 FAM Added mass force [N] v Mean velocity, y-direction [m s ] v′ Fluctuating local velocity, y-direction [m s-1] FB Brownian force [N] w Mean velocity, z-direction [m s-1] F Buoyancy-gravity force [N] BG w′ Fluctuating local velocity, z-direction [m s-1] 3 -1 FD Drag force [N] v Specific volume [m kg ] x Coordinate [m] FE Electrostatic force [N] y Coordinate [m] F Saffman lift force [N] L z Coordinate [m] FT Thermophoretic force [N] f Mass fraction [---]

Greek Meaning Unit Table of contents α Absorptivity [---] -1 α Parameter for viscosity calculation [Pa s K ] [K] β Parameter for viscosity calculation ∂ Derivatives [---]  Rate of turbulent dissipation [m2 s-3]  Emissivity [---] γ Probability of sticking [---] 1. Introduction ...... 17 η Deposit layer dynamic viscosity [Pa s] 1.1 Background ...... 17 η Efficiency [---] 1.2 Biomass fuels ...... 18 λ Molecular men free path [m] 2. Thesis outline ...... 19 µ Dynamic viscosity [Pa s] -3 ρ Density [kg m ] 3. Theories for developing deposit layers on the heat absorption tubes in a 2 -1 ν Kinematic viscosity [m s ] boiler...... 22 -2 -4 σ Stefan-Boltzmann constant [W m K ] 3.1 The difference between fouling and slagging ...... 22 [---] ξ Gaussian distributed random number 3.2 Effect of fuel properties and combustion conditions ...... 23 τ Time constant [s] 3.2.1 Alkali metals, chlorine and sulphur ...... 24 3.2.2 Metals in the fuel ...... 25 Subscripts Meaning 3.2.3 Agglomeration and trapping of alkali in the bed ...... 26 air Air 3.3 Material transport from the furnace to the heat absorption tubes. .... 27 bed Sand bed 3.4 Surface effects ...... 28 c Critical, at crystallisation 3.5 Sintering ...... 30 drum Drum 3.6 The coupling between deposit and corrosion...... 31 e Eddy flue gas Flue gas 4. Measurements ...... 33 fuel Fuel 5. Flow field simulations ...... 37 gas Gas 5.1 Particle force balance ...... 38 i Species, Fraction or Number 5.2 Stokes number ...... 43 i Tensor index 5.3 Particle energy balance ...... 45 in Into a control volume 5.4 Turbulence simulation using a particle tracking method ...... 46 is Isentropic 5.5 Physics of the flow and particle trajectories ...... 49 j Tensor index out Out from a control volume 6. Dynamic simulations of a CFB boiler ...... 52 p Particle 6.1 Simulation model ...... 52 RH2 Reheater 2 (in Intrex) 6.2 Combustion model ...... 54 SH3 Superheater 3 (in Intrex) 6.3 Energy and continuity equations...... 55 steam Steam sur Surrounding 7. Results ...... 60 water Water 7.1 Paper 1 ...... 61 ∞ Free steam properties 7.2 Paper 2 ...... 65

7.3 Paper 3 ...... 67 1. Introduction 7.4 Paper 4 ...... 70 7.5 Paper 5 ...... 79 8. Discussion ...... 85 9. Conclusions ...... 88 10. Future work ...... 91 11. References ...... 92 1.1 Background One of the major causes of outage of modern utility boilers is boiler tube fail- Appendix 1: Basic equations for numerical simulations ...... 101 ure. A study at the Department of Energy (DOE USA, 1998) found corrosion Appendix 2: Turbulence validation ...... 107 fatigue to be the main cause of failures, with fly ash erosion and soot blower erosion also contributing. Corrosion is mainly dependent on the components Appendix 3: Over all results from the fouling distribution simulations ... 111 of the flue gas and its temperature, the properties of the deposits on the tubes, the metal alloy used to make the tube and the tube surface temperature. As Appendix 4: Summary of deposit component analysis 2002–2003 ...... 125 well as corrosion, fouling and slagging also occur on the outside of the tubes. Appendix 5: Summary of deposit viscosity calculation ...... 128 Inorganic material from the fuel is deposited on the surface of the tubes and can cause corrosion, depending on the state of the deposit material and its PAPER 1 ...... 133 properties. Deposits also reduce the heat absorption capacity of the tubes, which increases the downstream flue gas temperature and may result in a drop PAPER 2 ...... 161 in the steam output and reduced boiler efficiency. In extreme cases the gas PAPER 3 ...... 189 flow may be restricted by heavy deposits, which affects the fluid dynamics of the boiler. PAPER 4 ...... 201 Knowledge has to be acquired from multiple disciplines in order to understand and hence minimize the effects of fouling in biomass fuelled boilers. PAPER 5 ...... 237 The aim of this thesis is to identify the main and most important factors that will make it possible to reduce the negative effects of fouling. This can only be achieved by understanding how and why fouling occurs and its effects on boiler performance. This investigation is driven by the following questions:

Question 1: What are the importance of the fuel and the combustion conditions with respect to deposition and corrosion on heat transfer tubes in the flue gas channel (in a biomass fired CFB boiler)?

Question 2: What are the major mechanisms of buildup of deposits on superheater tubes?

Question 3: How effective is soot blowing at removing deposits?

Question 2: What are the major mechanisms of buildup of deposits on superheater tubes?

Question 3: How effective is soot blowing at removing deposits?

Question 4: How is the overall boiler performance affected by deposits on heat transfer 2. Thesis outline surfaces?

The answers on these four questions are presented in chapter 9.

1.2 Biomass fuels The thesis presents an overview of fouling in biomass fired boilers using the- The use of biomass fuel to replace the widespread use of fossil fuels in com- oretical, experimental and numerical analysis. The thesis therefore involves bustion is promoted in order to reduce the discharge of the greenhouse gas discussions and chemical reasoning, descriptions of model transport equations CO2. Compared to boilers fired with coal and oil, biomass fired boilers devel- and various heat and mass balances required for deposit simulations and boil- op more complications related to both fouling and corrosion on the heat er performance calculations. The topics are introduced in chapters 3–6, and transfer surfaces (Bryers, 1996). Nordin et al. (1997) conducted a survey of the author´s publications are presented in five separate papers which are problems that occurred in Swedish biomass-fuelled boilers. They concluded summarised in chapter 7. that these problems are not easy to understand in detail and depend on a large number of different factors. The most important factor seems to be that bio- Chapter 3 includes: mass fuels have higher alkali metal contents than fossil fuels. Alkali metals, General discussions of theories for developing deposit layers on the heat ab- particularly sodium and potassium, can combine with chlorine to form low sorption tubes in a boiler. melting temperature compounds. Soot particles formed under these conditions are therefore more likely to adhere to the boiler tubes. Chapter 4 includes description of measurements: The use of fast growing biomass, which has increased absorption of alkali - Probe measurements of deposit material build up conducted in a boiler envi- metals from the ground, inevitably increases the incidence of fouling prob- ronment. lems during combustion. Boilers that use straw as fuel therefore usually - Measurements and analysis of boiler process data, involving the effect of develop more fouling and slagging problems compared to boilers that exclu- soot blowing on superheater efficiency over a three year period. sively use wood as fuel. - Measurements and observations of fuel composition, deposit growth rate, The use of recycled wood as fuel can also increase fouling problem as it is deposit chemical content and corrosion of superheater tubes over a seven year a source of chlorides and metals with low melting points such as zinc and period. lead. Chapter 5 includes: Theory behind numerical computational fluid dynamic (CFD) simulation of soot particle trajectories and impingements onto two superheater tubes exposed to cross-flow. Analysis has been performed for various parameters such as particle diameter, flue gas velocity, tube surface impingement distribution and relevance of various forces in particle momentum balance.

Chapter 6 includes: Theory underlying a dynamic model of a circulated fluidized bed boiler (CFB boiler). The main purpose of the model is to simulate and understand how deposits affect the overall boiler efficiency and performance.

18 19

Question 4: How is the overall boiler performance affected by deposits on heat transfer 2. Thesis outline surfaces?

The answers on these four questions are presented in chapter 9.

18 19

Chapter 7 includes: mainly affected by flue gas turbulence. They therefore hit all sides of a tube. A short presentation of the five publications. Submicron particles are also cooled down in the thermal boundary layer surrounding a tube, and hit the tube at the tube temperature. Larger particles hit The subject “Fouling in biomass fired boilers” was proposed by the supervi- the tube and cool down to the tube temperature if they get stuck. This investi- sors, but all investigations (both theoretical and experimental) presented in gation was performed by the author. the five publications were proposed and written by the author. Paper 4 focuses on the correlations between fuel mixture, deposit growth Paper 1 includes an overview of fouling and deposits in boilers. This encom- rate, deposit chemical content and corrosion of superheater tubes in Boiler 5, passes deposit probe measurement, chemical deposit analyses and viscosity Västerås, Sweden. The overall heat transfer coefficient (U-value) is calculated calculations. in the same way as in Paper 2, but the investigation period is extended to sev- Investigations with the deposit probe were performed in the flue gas channel en years. There is good correlation between the observed deposit thickness just upstream of superheater 2 in Boiler 5, Västerås Sweden. The flue gas and calculated U-value. The measured U-value is also correlated to current temperature was almost constant (800–850 °C) but the probe temperature var- fuel mixture. A significant decrease in U-value was observed when recycled ied between 280 °C and 570 °C. This creates a range of possible tube surface wood made up 10–20% of the total biomass fuel mixture. Recycled wood is a temperatures in the boiler. The exposure time varied from 1 to 300 hours. The source of chlorine and zinc, which were also detected in small amounts in the results showed that the deposit grow rate varied both with exposure time and deposits. The corrosion rate can also be correlated to the fuel mixture to some probe temperature. This paper also considers sintering processes, where loose extent. The author did most of the work described in this article, but the cor- deposits are converted to hard deposits. The author did most of the work de- rosion measurements were performed by the boiler company and the scribed in this article, but the probe measurements were done in cooperation summary of the fuel mixture analysis was mainly done by Christer Karlsson. with students at Mälardalen University. Paper 5 presents a dynamic model of Boiler 5 based on energy and mass bal- Paper 2 focuses on the effects of deposits on the heat transfer rate, especially ances, which simulates the effects of deposits on the boiler efficiency and in the superheater tubes. It also includes a long term investigation of the ef- performance. The model is based on a combustion model developed for “data fect of soot blowing on the heat transfer rate of superheater 2 in Boiler 5, reconciliation and quality assurance of signals”, but is further developed by Västerås, Sweden. Soot blowing has a significant effect on the heat transfer the author with the addition of heat exchangers (superheaters, reheaters, rate several hours after the soot blowing circle is completed. However, soot economiser and air preheater) and two turbines (high pressure turbine and in- blowing does not have a significant effect on the heat transfer rate after an intermediate pressure turbine) in order to make it possible to study the effects of terval of three years. An overall heat transfer rate coefficient (U-value) is fouling of heat exchanger surfaces on overall boiler performance. All equa- calculated from the heat transfer rate measurements. In order to obtain com- tions in the model are developed in Modellica 2.1 using the Dymola 6.1 parable values only measurements made one hour after a soot blowing cycle graphical interface. The model is calibrated and verified against Boiler 5 in was completed are included in the investigation. The fuel mixture appears to Västerås. The simulation shows that fouling on superheaters redistributes the be more important for the deposit growth rate than the soot blowing frequen- heat transfer rate from the superheaters to reheater 1 and also shifts turbine cy. This investigation was performed by the author. power from the HP turbine to the IP turbine. This investigation was performed by the author. Paper 3 presents theoretical and numerical studies of forces and particle trajectories in the vicinity of heat transfer tubes. The paper presents our fundamental understanding of deposit formation at different locations on tubes. From observations (presented in Paper 2 and Paper 4) it is known that the deposit thickness varies around a tube. In Paper 3 the author uses the CFD program Fluent to identify the main important phenomena for particles impinging on a tube. Large particles (> 5–10 µm) mainly hit the windward side of a tube because of inertia. Submicron particles follow the flue gas and are

20 21

3. Theories for developing deposit layers on the 3.2 Effect of fuel properties and combustion conditions heat absorption tubes in a boiler The deposits mostly contain oxides of inorganic matters such as silica, aluminium, iron, sulphur as well as alkalis such as calcium, potassium and sodium. Also complex components as silicates, sulphates, carbonates and hydroxides may exist. The first layer of the deposit nearest to the tube metal surface normally consists of different iron-oxides by reaction of the tube metal and the surrounding flue gas. All the deposit components originally come from the fuel or from materials in the combustion chamber. During and after the combustion of different types of fuel, inorganic and A deposit layer on a heat absorption tube is an effect of many different fac- sometimes also organic matters existing in the fuel are left as residues or ash. tors; fuel chemical composition in combination with the boiler design Depending on the combustion temperature these residues can be in vapour, including combustion parameters such as the temperature, reducing or oxida- melt or solid state but also other factors affect the residues as boiler design tion combustion environment, the particle residence time during combustion and combustion conditions. Today many large biomass boilers are built as etc. The deposit material is mainly unburned matters in state of vapours, bubbling (BFB) or circulating (CFB) beds. The bed material is normally silica melts and solid particles, transported in the flue gas. The size of the particles based sand, which makes the combustion rather stable with respect to temper- in the flue gas channels depends on the original fuel characteristics and on the ature. The bed material itself may also absorb some of the residues and combustion parameters. It may also depend on the bed material, if the boiler depending on the particle size the residues can stay in the bed or follow the is designed as a BFB (bubbling fluidized bed) or CFB (circulating fluidized flue gas. In the cyclone, further down the flue gas path, the separation of the bed). In a CFB boiler, the sizes of the particles in the flue gas channel also particles occurs and only very small particles will be transported along the depend on how effective the cyclone separation is. flue gas path, eventually be cooled at the heat absorption tubes. One difference between biomass fuel and fossil fuel is that biomass fuel normally contains more alkali metals (particularly potassium) compared to fossil fuel. These alkali metals in biomass fuels are also organic bounded to 3.1 The difference between fouling and slagging the fuel and therefore easily released compared to fossil fuels. Alkali metals Fouling occurs in the colder section of the boiler where inorganic volatiles like potassium (K) and sodium (Na) have lower melting and evaporation condensate as compounds on the colder tube surface forming a deposit layer. point then other inorganic materials in the fuel. This is one reason for in- Also particles in melt or partly melt phase may hit the tube, be cooled down creased fouling problems using biomass fuels. Many boilers also use recycled to solid state and get stuck on the tubes. The deposit layer may be porous or it wood as a fuel. Recycled wood is a very inhomogeneous fuel and can contain may be sintered to a hard and compact layer as it interact with the surround- elements such as chlorine (Cl), zinc (Zn) and lead (Pb). These elements to- ing gases or with other components in the deposit layer. Over time the deposit gether with alkali metals may form components as potassium chloride (KCl) layer grow in thickness and the temperature of outer surface increase as the which has two major disadvantages; it has a low melting point (below deposit itself insulate the cooler water or steam inside the tube from the hotter 800 °C) and it is highly corrosive. At common combustion temperatures in flue gas. The temperature will then eventually reach the melting point of at biomass fired boilers (800–900 °C) a large part of KCl is in vapour state as least some the constituents of the deposit. This process gradually accelerates KCl has high partial pressure at temperatures above 800 °C. In biomass com- and nearly anything that hits the deposit on the tube may get stuck. Ash parti- bustion also all fuel chlorine is released to the flue gas at temperatures above cles in solid state as well as small particles from the bed material in the 800 °C (Knudsen et al. 2004). combustion chamber may then continue to build up a thick deposit layer on Today it’s common to mix different fuels in order to get as good combus- the tubes. Thus the surface temperature continues to increase and the possibil- tion conditions as possible. Elements such as Al, Si and S are able to trap ity that a larger part of the deposit is in the molten phase increase and the alkali metals, limiting the formation of alkali chlorides. These elements (Al, deposit material may start to flow. At time the deposit becomes so heavy that Si and S) exist naturally in coal and it is therefore favourable to mix coal and it falls off the tube under its own weight. This process is called slagging. biomass, (Aho and Ferrer, 2004; Aho and Sivennoinen, 2005). Peat is a

22 23

source of sulphur and especially when using recycled wood as fuel it can be fuels and sewage sludge. Also other elements can influence the reactions, e.g. favourable to also use peat to achieve a high peat/recycled wood ratio in the if phosphorus are available in the fuel, phosphorus can react with calcium not biomass fuel mixture. capture sulphur.

3.2.1 Alkali metals, chlorine and sulphur 3.2.2 Metals in the fuel The availability of sulphur (S) in the fuel effects the reactions and the amount When burning recycled wood, zinc (Zn) and lead (Pb) and sometimes titani- of produced KCl, (Andersson and Högberg, 2001; Andersson, 2003). Sulphur um (Ti) can be observed in the deposit on the surfaces in the boiler. is reacting with potassium forming potassium sulphate, K2SO4 which has a Especially zinc is assumed to be a problem as it, in the form of chloride, melting point above 900 °C. As K2SO4 is formed HCl is also produced and is which in eutectic with alkali metals components gives very low melting transported, in vapour state, with the flue gas, sometimes getting trapped in points. Andersson et al. (2001) have indicated that in CFB boilers, zinc is the flue gas textile filter or in the flue gas condenser. mostly attracted to the bed material and is therefore concentrated in the bottom ash. However, other studies show that the amount of zinc in the flue gas A possible and likely reaction is: trapped in the flue gas filter, may be important (Steenari et al. 2008). In a recent study by Steenari and Noren (2009) where municipal solid waste was used as fuel in a fluidized bed, the zinc content was much higher in the flue 2KCl(s) + SO2 (g) + ½O2 (g) + H2O(g)  K2SO4 (s) + 2HCl(g) (3.1) gas ash than in the bottom ash. Typical values were 1400 mg Zn/kg dry ash in the bottom ash and 6000 mg Zn/kg dry ash in the fly ash. The authors pointed out that the zinc was likely to be predominantly in the form of Zn2SiO4 and K2SO4 is mainly in solid phase in the flue gas and is less inclined to stick to the tube surface. Zn5(CO3)2(OH)6 in the silicate rich bottom ash. Fly ash contained the same It has been proposed by Krause (1986) and Salmenoja (2000) that when Zinc species as well as small amounts of ZnCl2 and ZnSO4. the molar ratio of S/Cl in the fuel is greater than 4, deposit and corrosion by A study by Elled et al. (2007) of trace elements in waste combustion in a chlorine will be negligible. A similarly statement has been suggested by Rob- fluidized bed showed that the amount of zinc in the flue gas increased under inson et al. (2002) that if the molar ration 2S/(K+Na) > 4, the chlorine content reducing conditions but not under oxidizing conditions at 800–850 ºC. The in deposit is negligible. The term (K+Na) expresses the amount of easy re- first stage of combustion normally occurs under reducing conditions, making leased alkali in the fuel (not firmly bound in minerals). In biomasses, alkali it favourable for zinc to flow as a vapour in the flue gas and form compounds metals are primarily organically associated or present as simple salt. This in small ash and sand particles that follow the flue gas. means that alkali metals are easily released to the gas phase during combus- Elled et al. (2008) conducted experiments in a CFB boiler fuelled by a tion. mixture of wood pellets and recycled wood with additional injections of zinc Additions of Sulphur direct to the bed or as ammonium sulphate after the and chlorine. Zinc levels were measured with a deposit probe at 500 ºC ex- combustion zone have also been tested successful shifting alkali from chlo- posed to flue gas at 825 ºC. Ash particle sizes and their chemical contents rides to sulphates. The sulphating reaction is fast in the gas phase, and occurs were also analysed at 270 ºC. At low chlorine concentrations, the zinc was concentrated in particles of around 6 µm diameter, probably as zinc oxides. between the alkali chloride and SO3 (Lisa et al. 1999). If chlorine and sulphur are not present, alkali hydroxides (e.g. KOH) are When both chlorine and zinc were added, particles of around 0.7 µm diameter the major stable gas-phase species in combustion gases (Baxter et al. 1996) were detected, presumably composed of zinc chloride. Combustion of recycled (demolition) wood contaminated with zinc alone generated a modest but also alkali carbonate (e.g. K2CO3 ) could be present (Knudsen et al. 2004). Recent researches have shown that the correlation could be even more amount of deposit as measured by a deposit probe exposed for 8–12 h. When complicated. If the amount of calcium is high in the fuel, calcium can react both chlorine and zinc were added to the fuel, the deposit rate increased by up to three times compared to combustion without added zinc and chlorine. The with sulphur (to form CaSO4) and thereby not decrease the effect of potassium chloride in the flue gas. Åmand et al. (2006) reported large amount of authors (Elled et al. 2008) pointed out that KCl and NaCl are more stable both CaSO and KCl in ash particles in size 1–10 m in co-combustion bio- compounds compared to zinc chloride but zinc chloride may also be formed 4 in the presence of excess chlorine. The ratio (Cl-K-Na)/Zn can be an indica-

24 25

tion of the probability for zinc chloride to appear. Berg et al. (2003) and Hen- In this reaction HCl is released and potassium is bound to the ash. The result- derson et al. (2006) reported that zinc chloride is not a very stable component ing compound K2O Al2O3 2SiO2 has higher melting points than those in most at temperatures above 350 °C but in reducing conditions zinc chloride is fluidized bed combustion situations whereby it both traps alkali and prevents thermodynamically favored between 450 and 850 °C (Elled et al. 2008). agglomeration. Thermodynamical calculations (based on stable temperatures and con- As both aluminium and silicates are able to trap alkali the ratio (Al+Si)/Cl trolled conditions) can give good indications of the present state in the could be of interesting and Aho and Silvennoinen (2005) suggested that a mo- furnace and in the following flue gas, but there is always uncertainties as the lar ratio of (Al+Si)/Cl higher than 8–10 is needed preventing chlorine in conditions change rapidly following the flue gas path. For example, at a deposits. Also the ratio (Al+Si)/(Na+K) is of interest discussing the probabil- common flue gas velocity of 5–10 m/s the temperature drop is about 100– ity for trapping alkali metals in the bottom ash. 200 °C per second, following the flue gas path, along the superheaters and re- As the above examples show, the mechanisms for alkali trapping in bed, heaters. bed agglomeration and deposition on heat transfer surfaces may be dependent on each other. 3.2.3 Agglomeration and trapping of alkali in the bed When alkali is released during combustion it can follow the flue gas in vapour 3.3 Material transport from the furnace to the heat or melted phase (e.g. KCl), or in ash particles in solid state (e.g. K2SO4) but can also remain in the bed material. In a silica rich sand bed, alkali can be absorption tubes. transported to the surface of the bed particles forming alkali silicates. At high Fly ashes are normally very small inorganic particles. Particles less than 10 temperatures alkali silicates is a sticky component favouring agglomeration m are too fine to be separated in a cyclone and will follow very close to the (Öhman et al. 2000). In severe cases defluidization occurs and the plant has to gas phase. Larger particles can also to some extend follow the gas phase but be shut down. The process depends on time and temperature and it is im- are more effected of its own inertia. Coarser ash particles (> 1–10 m) are of- portant not to exceed 900 °C in the bed preventing agglomerations  ten formed by calcium, magnesium and silica (Benson et al. 1996). (Davidsson et al. 2008). Volatile matter released from the fuel will also follow the flue gas. As the The probability for alkali chloride depend also on the availability of sul- temperature decrease in the flue gas path or in the vicinity of heat transfer sur- phur and Hansen et al. (1995), showed that alkali silicates are formed when faces, inorganic gases as KCl can condensate homogeneously or the ratio S/(Na+K) is low and below 0.5. When the ratio is higher mainly al- heterogeneously, forming submicron particles. Submicron particles in the flue kali sulphates will be formed. gas are mainly originally volatile matters released from the fuel. Öman et al. To prevent the formation of alkali silicates in the bed, alternative bed ma- (2006) reported a considerable reduction of the amounts of fine submicron terials such that contain little silica sand can be used. Examples of alternative particles when peat is added to biomass fuels. This is well in agreement with materials are olivine and blast-furnace slag (Brus et al. 2004, Davidsson et al. the theory as peat includes sulphur and the amount of vaporised KCl is re- 2008). But the use of a bed material that doesn’t react with alkali may result duced as K SO is formed. in a higher concentration of alkali chlorides in the flue gas, thereby causing 2 4 Volatile matters can also condense on heat transfer tubes making the tubes more deposits on heat transfer tubes. sticky or condense on coarser ash particles making also solid ash particles Aluminium silicate (kaolin) is also able to trap alkali metals and is there- sticky. These phenomenons increase thereby the probability for ash particles fore used as additive to the fuel. A possible and likely overall reaction is to get stuck on the heat transfer tubes building up larger deposits (Åmand et (Åmand et al 2006): al. 2006 and Valmari et al. 1999). The main mechanisms forcing a particle to hit a tube are by inertia, ther- Al2O3 2SiO2 (s) + 2KCl (g) + H2O (g)  K2O Al2O3 2SiO2 (s) + 2HCl (g) mophoresis and by turbulent motions. Other effect like Saffman forces, (3.2) electrostatic forces and molecular diffusion may also affect the transport process. These aspects are further discussed in chapter 5.

26 27

3.4 Surface effects ηref γ c = , η >η ref (3.3) A particle hitting a colder surface may get stuck or may bounce away. The η sticking probability can be calculated by a number of ways: In the first method, the components present are very well defined and chemical and thermodynamic calculation can provide an estimate of the particle eutectic. γ c = 1 , η ≤ η ref (3.4) The method start with a definition of T0 “first melting point“ were all components are in solid phase and T100 were all components are in melt phase. It is commonly a large interval (e.g. 100–300 °C) between these points. The stick- Where η ref is the critical viscosity of a particle, at the beginning of crystalli- ing interval is than defined as T15 up toT70 (where 15–70 % is in melted phase). It is a labouring task to perform an evaluation of all possibilities of zation, which has to be determined experimentally. Particles with a viscosity chemical components in the deposit and whether or not a particle is sticky lower than the critical viscosity are regarded as trapped on the surface. The (Anderson et al. 2001). practical viscosity of the particle may be calculated according to the follow- In 2006 the method was tested by Theis et al. (2006d). Theoretical calcula- ing equation (Urbain et al. 1981, Urbain et al. 1982): tions of eutectic, including K2SO4, N2SO4 and KCl were conducted as well as experiments at 800 °C. Particles were found to be sticky when they contained 1000β  more than 15 % molten phase. KCl seems to affect the stickiness very much. = αη T exp  (3.5) When the amount of KCl varied from 1 to10 %, the molten phase increase  T  from 15 to 100 % at 800 °C. As an extension of the first method a second method is proposed in this thesis. If volatiles condensate on the cold surface, the condensate can act as whereα and β are constants, depending on the composition of the ash parti- glue for solid particles or the possibility that condensation of volatiles on a cle. These constants can be determined from the ash composition analysis: particle surface will make the whole particle sticky. In this case only a small fraction (much less than 15 % as in the first method) of the total particles is in α = f ([CaO], [MgO], [SiO ], [TiO ], [Al O ], [Fe O ], [Na O], [K O]) melted phase but gets stuck. Submicron particles in melt phase in the flue gas 2 2 2 3 2 3 2 2 channel have mainly and originally been in the volatile state after combustion (3.6) and can also interact as glue between deposit particles and the tube. This method is also developed to explained results presented in Paper 4 (A 7 year = f ([CaO], [MgO], [SiO ], [TiO ], [Al O ], [Fe O ], [Na O], [K O]) long measurement period investigating the correlation of corrosion, deposit β 2 2 2 3 2 3 2 2 and fuel in a biomass fired circulating fluidized bed boiler). (3.7) A third method is to calculate an average ash viscosity. This method has been used in the prediction of fouling in coal fired boilers (Senior and Srini- vasachar 1995). The method is based on the fact that some components form The method has usually been calibrated and used for coal ash and has been commonly applied for many years in the glass industry. The base component long stable polymers responsible for high viscosity (e.g. SiO2), other components tend to reduce the stability of the network bye bridging oxygen bonds of glass is silicon dioxide, but by mixing it with alkali metals, the viscosity is lowered dramatically. For this reason the method is interesting since biomass within the network former, thus reducing the viscosity (e.g. Na2O and K2O). By this third method the possibility for sticking mainly depend on the vis- usually have high alkali content. The detail of the method can be studied in cosity of the particle. The sticking probability is related to the ratio of the Appendix 5 and is also further explained in Paper 1. critical viscosity of the particle to the practical viscosity of the particle (Sen- A fourth method of defining stickiness is described by Basu et al. (1999). ior and Srinivasachar 1995, Zhang at al. 2001): This method is similar to the previous method and divide the components into acidic ([SiO2], [TiO2], [Al2O3]) and basic ([CaO], [MgO], [Fe2O3], [Na2O],

28 29

[K2O]). The temperature at which the ash starts softening or melt can then be ductivity. In Paper 1 (Measurements, theories and simulations of particle de- calculated as a function of the base to acid ratio. posits on superheater tubes in a CFB biomass boiler) these phenomenon is As a fifth method, chemical reactions may contribute to increase the mass further explained. Sintering may also explain why soot blowing of superheat- of a deposit. This includes reactions between gases and materials already in er tubes only shows a strong positive effect on the heat transfer rate shortly the deposited and possibly even directly between the gas and tube surface after a soot blowing cycle is completed, but not after an interval of three (Baxter 1996). Possible chemical reactions in a deposit are oxidations and years. Soot blowing is further discussed in Paper 2 (Long time investigation sulphation. For example if chlorine is depositing as alkali chlorides on the of the effect of fouling on the superheaters in a circulating fluidized biomass tubes it becomes subsequently sulphated by the SO2 in the flue gases. Also boiler). relatively low sulphur content of biofuels is sufficient to convert alkali chlorides to corresponding sulphates in the deposit (Salmenoja 2000).

Erosion 3.6 The coupling between deposit and corrosion There are also possibilities for decreasing the deposits by erosion. Ash particles can have a scavenging effect on deposits. As the temperature decrease in High temperature corrosion the flue gas path ash particles become harder and the possibility increase for Boilers in general operate with excess of oxygen in the end of the furnace and eroding the deposits building up (Ots, 2001, Thesis 2006d). in the flue gas channel (oxidation atmosphere). On the tube surface, oxidation transforms iron to Fe3O4 or Fe2O3 (Basu et al. 1999). These hard oxide layers normally protect the tube from further oxidation. At high temperatures, around or above 500 ºC, complex chemical reactions can take place leading to 3.5 Sintering porous components and components with high molar volume which may Sintering is a process making particles stick to each other resulting in strong crack the protective layer. and hard materials. As described by Bryers (1996) at least four mechanisms If the protecting oxide layer breaks, an electrochemical reaction can occur are responsible for sintering i.e. particle-to-particle bonding. The four mecha- between the cracked interior (anode) and the oxide layer (cathode). The inte- nisms are viscous flow, chemical reaction, diffusion, and surface tension. rior becomes corroded forming pits, while the cathode remains non-corroded. Sintering increase the material strength exponentially with temperature and These corrosion problems increase if the deposit is in melt or partly melt can be represented by the Arrhenius equation: phase. In superheaters the use of corrosion resistant alloys, such as highly alloyed chromium steel, is often necessary to reduce the corrosion speed. On the other = AC e − R TE hand these highly alloyed chromium (and nickel) steel are more sensitive to sinter sinter (3.8) corrosion influenced by thermal and mechanical stresses and therefore also low alloyed chromium steel still are used. In biomass fired boilers and especially boilers utilizing recycled wood as where Asinter is the pre exponential factor and E is the activation energy. Fly ash, from coal combustion, is studied by Bryers (1996). The strength fuel the amount of chlorine may be high and appear as Cl2, HCl and KCl. If of sintering is highly affected by the amount of alkali. With 80 % alkali (in chlorine is present it can diffuse through the oxide layer, producing metal (M) chlorides and thereby making the layer porous and less protective. this case Na2O) the sintering is important already at 700 °C. For lower amount of alkali the sintering process started at much higher temperatures. In “active oxidation” Cl2 or HCl diffuse through the oxide layer to the Nordin et al (1997) suggested that chemical sintering by gas components metal surface and react with the metal alloys (Nielsen et al. 2000): (e.g. SO2 or CO2) is one important mechanism in deposit developments. Another effect of sintering is altering of density and thermal conductivity M + Cl2  MCl2 (3.9) of the deposit matter. Andersson (2003) reported that sintering could increase the thermal conductivity 2–3 times. The explanation to these phenomena is that sintering gives increased density and thereby also increased thermal con- M + 2HCl  MCl2 + H2 (3.10)

30 31

If the oxygen partial pressure is low (which is likely as oxygen is consumed 4. Measurements to produce metal oxides) chlorides such as Fe2Cl and CrCl2 are thermodynamically stable. These volatile metal chlorides may diffuse to the deposit surface, where the partial pressure of oxygen is higher, leading to oxidation of the metal chloride to solid metal oxides (at the deposit surface). As chlorine is released it can diffuse back to the metal surface continuing the process. If potassium chloride (KCl) is available in melt phase it may increase the corrosion process as chemical reactions may be faster in liquid phase than in solid phase. Sulphating of (KCl) in the presence of SO2 releasing HCl can al- The subject deposit and fouling on heat transfer tubes is an experimental dis- so be an important contribution for the active oxidation process (Salmenoja cipline and normal calculations and theories have to be validated by studies in 2000). real plants. In this thesis, comparable studies have been conducted at the mu- In flue gases from combustions of waste, the concentration of HCl is very nicipal biomass fired cogeneration Boiler 5 in Västerås, Sweden. high and the “active oxidation” process is well accepted explaining corrosion. The boiler is a circulating fluidized bed boiler (CFB-boiler), with a thermal But in biomass fuelled boilers the concentrations of HCl sometimes is rather power of 157 MW, producing both heat and electricity. The boiler is low and alternative theories are developed increasing the importance of KCl. equipped with a flue gas condensation system with a maximum heat power of Segerdahl (2003) and Henderson et al. (2006) have suggested the following 48.5 MW. Other properties include: reaction: Steam data at turbine inlet: 55 kg/s, 540 °C, 173 bar. Bed temperature: 800–850 °C. 2 KCl (s) + ½ Cr2 O3 (s) + H2 O (g) + Bed material: Silica rich sand. ¾ O2 (g)  K2 CrO4 (s) + 2 HCl (g) (3.11) Flue gas cleaning system: - Reduction of NOx by catalytic cleaning. In this reaction the protecting oxide layer breaks and, in the same time, HCl - Flue gas textile filter for particle trapping. diffuse out to the flue gas and are not any more participating in the reactions. - Gas species as HCl are mainly trapped in the flue gas condenser. To run a boiler (at start) with fuels not containing high levels of chlorine may give a protection layer on the heat exchanger surfaces decreasing the cor- A main parameter investigated by the author in this thesis is the deposit layer rosion rate. This phenomenon is further discussed in Paper 4 (A 7 year long thickness on the boiler superheaters and reheaters. Superheater 2 (SH2) is the measurement period investigating the correlation of corrosion, deposit and first superheater in the flue gas path after the cyclone and is therefore ex- fuel in a biomass fired circulating fluidized bed boiler). pected to have the highest fouling problems. Figure 1 shows the positions of Also sulphur may form porous components with iron and is, in coal fired the superheaters and reheaters in Boiler 5. Superheater 2 and superheater 1 boilers where the furnace temperature is higher than in biomass fired boilers, each consist of two tube bundles denoted SH2,p1; SH2,p2 and SH1,p1; a source for severe corrosion problems. SH1,p2. Reheater 1 consists of three tube bundles (ReH1,p1; ReH1,p2 and ReH1,p3 in the flue gas direction). The tube bundles are separated by distanc- Low temperature corrosion es of 0.5 to 1 m, making visual inspections possible through the inspection Corrosion problems in air preheaters and economizers are called low- openings (shown as small circles between the tube bundles in figure 1). temperature corrosion. The main cause of this is that gases, like sulphur triox- ide, condensate when the temperature is below its dew point. The condensation follows by a reaction with water forming an acid, e.g. sulphuric acid, on the metal surface dissolving the protective oxide film.

32 33

residues, wood, bark and small amounts of salix. In Paper 4, a more detail study of the fuel utilised during the investigation period is presented. During the annual summer revision the deposits growth was measured on the heat transfer tubes in the flue gas path. To determine the chemical content of deposits, samples from superheater 2 were analysed from 2002–2007 using Scanning Electron Microscopy, with energy dispersive X-ray spectrometer (SEM-EDS). Furthermore logged measured values of the flue gas, the steam temperature and the heat transfer rate etc. is also available from the boiler operating and controlling system, on hourly bases during the whole year. To follow corrosion rate on tubes, tube thickness measurements were performed on superheater 2 from 2001 to 2006, using an ultrasonic measuring device and by the standard method SS-EN 14127. The tube material of superheater 2 (SH2,p1) is 13CrMo44, a low alloyed steel material (1.09 % Cr and 0.4 % Mo).

Figure 1. Scheme of Boiler 5 in Västerås and the positions of the superheaters and the reheaters.

The CFB boiler was started in January/February 2001 and is normally running at full load for at least 8 months a year. The boiler is only available for inspection and studies for a couple of weeks in the beginning of august every year, during the revision period. The boiler has a construction called intrex in the fluidized sand-lock, situated in the cyclone leg. This is where the circulating fluidization sand is reinjected into the boiler bed-zone and also where superheater 3 and reheater 2 is located (figure 1). During the revision in august 2003 the boiler was re- built, the gills of superheater 3 where reduced allowing only a small flow- connection from the sand-lock to the combustion bed. No other main recon- structions have been done during the investigation period 2001–2007. Figure 2. Photography of superheater 2 showing the formation of deposit The fuel types utilised in the boiler are typically a mixture of different layers, from August 2003. types of biomass (50–60 %), peat (20–40 %) and recycled wood, i.e. wood wastes (10–15 %). The usual types of normally used biomass fuels are: forest The conventional practice to reduce the amount of deposit and thereby increase the heat transfer rate and the boiler efficiency is to use soot blowing

34 35

Figure 1. Scheme of Boiler 5 in Västerås and the positions of the superheaters and the reheaters.

34 35

systems with high-pressure steam. Studies and conclusion of how effective the soot blowing system is have been conducted and is presented in details in 5. Flow field simulations Paper 2. Another method for studying deposits is to use a deposit probe. The deposit probe, which includes a small deposit ring, is cooled by water and air to achieve a settled temperature. The ring can be exposed to the flue gas during different time periods from hours up to weeks and may thereafter be analysed, both mechanically and chemically in a laboratory. In this thesis a deposit probe was used situated above superheater 2 in order to investigate the depos- In order to understand the mechanisms responsible for the transport of the it development with respect to time and temperature etc. Further details flue gas particles and the deposit on the heat absorption tubes, a number of concerning the probe measurements are presented in Paper 1. CFD (Computational Fluid Dynamics) calculations have been performed using the commercial code Fluent 5.4. The superheaters in a boiler are usually designed as rows of parallel tubes in crossflow. As the Reynolds number is usually quite high, the flow field may be turbulent. The particle mass flow rate is typically less than one percent of the flue gas flow rate, so the flow field can be considered as a dilute two phase flow. The flow field around the first tube of a superheater tube bundle has completely different flow field characteristics to the rest of the tubes situated in the wake downstream of the first tube. It is therefore reasonable to include at least the first and second tubes in a flow field simulation. A coupled Eulerian- Lagrangian approach has been chosen for the two phase flow. If a solution of the stationary flow field is developed for the gas phase a large number of particle trajectories can be estimated with a particle tracking method. In order to simulate the flow field around two tubes in crossflow, the classical k- method is used for the turbulent flue gas flow. The method of simulating the second phase by a particle tracking method has been used with good prediction results in a variety of flows, e.g. turbulent dispersion (Baxter and Smith, 1993) and particle-laden gas flows past tubes (Schuh et al., 1989). Special care has to be taken regarding the boundary conditions as the particle path near the surface is critical. In the simulations, a special two equation treat-

ment is used for the boundary calculations, including extremely small Figure 3. The probe surface with and without deposit layer. computational cells, thereby also resolving the viscous sub layer. Compared to the classical logarithmic boundary law, the cell attached to the wall is resolved into up to 25 cells, and velocities and temperature gradients acting on the submicron particles very close to the wall where the velocity approaches zero are also resolved. All the equations used to solve the stationery flue gas flow solution are presented in Appendix 1. The calculations are performed to double precision.

36 37

Outlet 3 πDp FBG ()p −= ρρ gi 6 (5.2) Second tube

FD is the force on the particle due to drag, meaning the force on the particle created due to velocity differences between the particle and the gas. The drag Frontal tube force is the most important force for particle transportation. In this case this is modelled as a steady state assumption:

Symmetry 2 ρ − − uuuu boundary πDp ( p ) p FD = CD 4 2 (5.3)

Inlet 3 Where the CD number for spherical particles at Re < 10 is as follows (Balzer et al., 1995): Figure 4. Geometry and the grid.

24 7.0 One of the main objectives of the gas particle flow field numerical investiga- CD ( 1 += 0 . 5 R e1 ) tion is to deduce a model for particle transportation, explaining the Re (5.4) distribution of different particle sizes on different regions of the tube surfaces. The probability of particles hitting the front or second tube in the flue gas direction has also been investigated. − Duu ρ Re = pp µ (5.5) 5.1 Particle force balance

The force balance terms have been derived and compared with ideas from For small Re numbers (< 1), the CD number for spherical particles is calculat- other investigations, namely Maxey et al. (1983), Li and Ahmadi (1992), and ed from the Stokes law: Park et al. (1995). 24 C = du D p Re (5.6) m p F BG += F D AM ++ FF L + F + F BT dt (5.1) or 24 1 CD = FBG is the force on the particle due to buoyancy-gravity. This is the force cre- Re C C (5.7) ated on the particle due to differences in density between the particle (ρp) and gas (ρ) where:

38 39

Outlet 3 πDp FBG ()p −= ρρ gi 6 (5.2) Second tube

Symmetry 2 ρ − − uuuu boundary πDp ( p ) p FD = CD 4 2 (5.3)

Inlet 3 Where the CD number for spherical particles at Re < 10 is as follows (Balzer et al., 1995): Figure 4. Geometry and the grid.

38 39

3 Where Cunningham's correction Cc is (Li and Ahmadi, 1992). K2CO3 ρp = 2290 kg/m 3 K2SO4 ρp = 2700 kg/m 3  D  SiO2 ρp = 2200 kg/m   p   2λ  − 1.1  3   2λ   CaCO3 ρp = 2800 kg/m CC 1+= 7.1 + 4.025 e 3 CaSO4 ρp = 2960 kg/m Dp     (5.8) Agglomeration of different species in one particle is expected to result in a more porous structure with lower particle density. For this reason the value 3 λ is the so called molecular mean free path. The size of λ can be estimated by ρp = 2000 kg/m is used in the calculations. kinetic gas theory. Saffman (1980) derived the equation The “free fall” or slip velocity may be estimated from a force balance between drag and buoyancy-gravity, FBG = FD. For a 20 µm particle, the free fall velocity is approximately 0.011 m/s, and for a 0.1 µm particle, the free fall 2µ λ = velocity is approximately 2.2 µm/s (at 700 °C). ρu molecule (5.9) The estimated free fall velocities are relatively small compared to either typical flue gas velocities (2–10 m/s), or to expected turbulence levels (5– 10 %). However, close to a surface the gas velocity approaches zero and FBG The mean molecule velocity: may therefore have an effect, particularly in wakes. FBG is included in the calculations. -11 For reference, for a 20 µm particle the FBG is estimated to be 8.2 × 10 N 8RT and for a 0.1 µm particle it is estimated to be 1.0 × 10-17 N. umolecule = π (5.10) FL is the Saffman lift force (Saffman, 1965). This force is due to velocity gradients (shear flow) acting on particles, thus creating a lifting force on the R is the specific gas constant, ρ is density and µ is the viscosity of the gas. particles. Using typical values; T = 973 K (700 °C), R = 287 J/kg K, ρ = 0.36 kg/m3, µ = 4.1× 10-5 Pas.(Mörtstedt & Hellsten, 1987): 2 Dp ∂u 1 F = 6 4 6. µ − uu )( L 4 p ∂y ν umolecule = 843 m/s (5.11)

λ = 2.7 × 10-7 m In a rough estimation of the size of the Saffman force in the actual case, a typ- For sub-micron particles the mean free path is of the same magnitude as the ical value of ∂ u/∂ y is suggested from the Blasius solution of a flow parallel particle diameter. In this range the Cunningham's correction becomes im- to a flat surface (Incropera & DeWitt, 2002): portant. The correction is included for particles in the range of 0.1–10 µm.

The density of particles (ρp) is much higher than the density of the flue ∂u u ρ gas. The ash particles are typically silicates, sulphates and carbonates. The = 0 33 2. u ∞ densities of some of these species at 15 °C are (Perry’s Chemical Engineer’s ∞ ∂y y=0 µx Handbook): (5.12)

40 41

At a position 10 mm from the edge (x = 0.01) of the flat surface, correspond- 2 pπµ CD s K + C t Kn )(6 ing approximately to a point on the windward side of the frontal tube and DT = (5.15) using the estimated free falling velocities as the typical slip velocity (u-up) be- ρ ( 1 + m ) (3 1 2 ++ 2 t KnCKKnC ) tween the fluid and the particle the Saffman lift forces are:

-12 FL = 2.3 × 10 N for Dp = 20 µm Kn = 2λ/Dp (Knudsen number). K = k/kp (ratio fluid thermal conductivity to particle thermal conductivity). -20 FL = 1.2 × 10 N for Dp = 0.1 µm Cs, Cm, and Ct are numerical factors of order unity that can be derived from kinetic gas theory. In the studied case the Saffmans force is lower or much lower than the buoyancy-gravity force and is therefore further disregarded in the simulations. The The thermophoretic force is important for applications of small submicron use of this assumption is supported by test simulations that included the particles exposed to a large temperature gradient. Under these conditions the Saffman force, which had no observable effects on the results. The impact of thermophoretic force can be much larger than the buoyancy-gravity force. A the Saffman force in different problems is described further by Saffman rough estimate of the magnitude of the thermophoretic force for a submicron -13 -14 (1965), Saffman (1968) and Wang et al. (1997). particle of 0.1 µm in the actual case shows values of magnitude 10 –10 N. FAM is the added mass, or the “virtual mass” force, required to accelerate Very close to the wall where the velocities and therefore the drag decrease to the fluid surrounding the particle. zero, a velocity independent force such as the thermophoretic force may be important. This force is therefore included in the simulation. FB is the Brownian force that may be included in laminar flow with sub- 3 πD p d micron particles. Brownian motion is random molecular movement and is ex- FAM = ρ ()− uu p pected to be of smaller magnitude than the thermophoretic force, and is thus 12 dt (5.13) omitted in this case. This assumption is supported by test simulations which included the Brownian force, which had no observable effects on the results.

The added mass force is important when p < ρρ gas . However this is not the case when looking at flue gas particles. The added mass force is therefore not included in the simulations in this work. 5.2 Stokes number FT is the thermophoretic force. Small particles suspended in a gas with a Simplifying the particle force balance (equation 5.1), only including the drag temperature gradient experience a force in the opposite direction to that of the force temperature gradient as a result of increased molecular movements (internal energy) at higher temperatures. This phenomenon is known as thermophoresis (Talbot et al., 1980): du p m p = FD dt (5.16) 1 ∂T F = D TT T ∂x i (5.14) 24 Using C = the acceleration of the spherical particle from up = 0 can be D Re estimated as The thermophoretic coefficient DT can be obtained for a sphere in a perfect gas (Talbot et al., 1980):

42 43

du p 18µ 5.3 Particle energy balance = 2 u dt D p ρ Transient storage of energy in a small particle is balanced by a heat flow of p (5.17) convection and radiation.

From this equation the particle time constant is identified as 3 ρπ D ∂Tc pp conv += QQ rad 6 ∂t (5.21) D 2 ρ τ = p p p 18µ (5.18) As the studied particles are very small it can be assumed that it is unnecessary to resolve the spatial temperature distribution inside the particles. The temperature T is therefore only a function of time. This assumption is called a If the particle time constant is divided by a typical fluid time scale p “lumped capacitance method”, where the temperature difference inside the particle is much less than the temperature difference between the particle sur- D 2/ face and the surrounding fluid. τ = t u (5.19) Convection Newton’s law of cooling states:

(Dt is the tube diameter and u the free flow velocity at the inlet)

conv h AQ p T ∞ −= T p )( The ratio of particle time constant to the fluid time scale is defined as the (5.22) Stokes number (St-number)

Besides the particle surface (Ap) and the temperature difference between the 2 fluid and the particle surface, the heat transfer rate is also a function of the τ D pp ρ p u St == heat transfer coefficient (h). τ 9µD t (5.20) Developing the basic differential equation for the flow field and the temperature field to a non-dimensional form and introducing the Nusselt number (Nu-number) as the non-dimensional temperature gradient at the surface For small St-numbers, the particles easily follow the fluid stream line around (y=0), it is possible to derive a relation to calculate the Nu- number for a the tube. For larger St-numbers, the particles have a higher probability of hit- specified geometry (Incropera & DeWitt, 2002). ting the tube (by inertia). Applied to forced convection; Setting St=1, the particle diameter is estimated to 34 µm. The Nu-number is a function of the Reynolds number (Re) and the Prandtl Typical values used in this estimation are: number (Pr).

3 ρp = 2000 kg/m Nu = f(Re, Pr) (5.23) u = 8 m/s Dt = 51 mm (Tube diameter) µ = 4.1 × 10-5 Pa s uDp ρ Re = (5.24) µ

44 45

du p 18µ 5.3 Particle energy balance = 2 u dt D p ρ Transient storage of energy in a small particle is balanced by a heat flow of p (5.17) convection and radiation.

(Dt is the tube diameter and u the free flow velocity at the inlet)

conv h AQ p T ∞ −= T p )( The ratio of particle time constant to the fluid time scale is defined as the (5.22) Stokes number (St-number)

3 ρp = 2000 kg/m Nu = f(Re, Pr) (5.23) u = 8 m/s Dt = 51 mm (Tube diameter) µ = 4.1 × 10-5 Pa s uDp ρ Re = (5.24) µ

44 45

The relation of the Nu-number to the heat transfer coefficient (h) is derived, In the particle trajectory simulation presented in this thesis, a stochastic starting from a heat balance equation based on the fact that the heat transfer is separated flow (SSF) analysis, proposed by Gosman and Ioannides (1981) transported by conduction in the fluid at zero velocity (at the surface y = 0) and developed by Shuen et al. (1983, 1983a, 1984), is used. It involves com- puting trajectories of a statistically significant number of particles as they move from the injector and encounter a succession of turbulent eddies with a ∂T TTh T )( =− −kT lifetime te and a fluid velocity u(x), v (y), w(z), where: ∞ p ∂y y=0 (5.25) = + uuu ′ , = + vvv ′ , = + www ′ (5.28)

Ranz and Marshall (1952) have for spherical particles: u , v , w =The local stationary (mean) velocity in x, y and z directions hD Nu p == + P rRe6.00.2 3/12/1 The specific turbulent energy is k (5.26) 1 = ( uk ′ 2 + v ′ + w ′22 ) Radiation 2 (5.29) Energy exchange by radiation between a small surface (Ap) and a large surrounding area is expressed as (Incropera & DeWitt, 2002) If the turbulence is isotropic the turbulent RMS value (Root mean square val-

44 ue) is: rad = p AQ σε T s u rp − T p )( (5.27)

2k vu ′2 = ′2 = wv ′2 = -8 2 4 Stefan-Boltzmann constant σ =5.67 10 W/(m K ) 3 (5.30)

ε p = Emissivity The surface (A ) is assumed to be “gray” that is ε =α p p p Corresponding to a standard deviation in a Gaussian distributed probability

α p = Absorptivity density function (pdf). To simulate “stochastic” particle tracking:

5.4 Turbulence simulation using a particle tracking u = + ξ uu ′2 (5.31) method A common hypothesis is that the deposit of small particles is caused by the ′2 (5.32) turbulence fluctuations in the boundary layer close to the surface (Basu et al., += ξνν ν 1999). The idea of simulating the particles using a particle tracking model was proposed by Hutchinson et al. (1971) and by Yuu et al. (1978). w = + ξ ww ′2 (5.33)

46 47

44 ue) is: rad = p AQ σε T s u rp − T p )( (5.27)

2k vu ′2 = ′2 = wv ′2 = -8 2 4 Stefan-Boltzmann constant σ =5.67 10 W/(m K ) 3 (5.30)

ε p = Emissivity The surface (A ) is assumed to be “gray” that is ε =α p p p Corresponding to a standard deviation in a Gaussian distributed probability

α p = Absorptivity density function (pdf). To simulate “stochastic” particle tracking:

46 47

Where ξ is a normally distributed random number. As the turbulence level seems to be critical, a turbulence validation is presented in Appendix 2. This velocity (u, v, w) is acting on a particle over the lifetime of the turbulent eddy (te). Using the particle force balance (equation 5.1) and the turbulent eddy lifetime, the particle movement to a new position can be estimated. A new estimate is made starting from this new point (with local values of velocities, 5.5 Physics of the flow and particle trajectories turbulent energy, turbulent eddy lifetime and a new random number), and so The stationary solutions of the velocity and temperature field are solved. The on iteratively. stationary solution is an approximation of the real flow, as the real flow is ex- The eddy lifetime is estimated as “the characteristic length divided by the pected to have non-stationary components. Globally, non-homogenous characteristic velocity”. The characteristic size of an eddy is the dissipation biomass fuels in combination with the controlling system of the boiler will length scale (Le). Gosman and Ioannides (1981) suggested: induce variations in the flue gas flow and the flue gas temperatures. Flow around tubes in crossflow may also generate swirling wakes that affect the

1 separation points and may thereby also affect the flow field behind the tubes 2 C 3 in the flow direction. L = µ k 2 e ε An example of a stationary solution of the velocity field is presented in (5.34) figure 5.

and Shuen et al. (1983): u (inlet) = 8 m/s, T(inlet, flue gas) = 700 °C, Tt(tube surface) = 450 °C.

3 4 C 3 µ 2 Le = k ε (5.35)

If the turbulent RMS value is used as the characteristic velocity, the eddy lifetime can be computed as:

2k t = Lee / ( ) 3 (5.36) or

k = Ct t ee ε (5.37)

Cte is typically 0.15–0.37. In this thesis Cte = 0.30 is used.

48 49

and Shuen et al. (1983): u (inlet) = 8 m/s, T(inlet, flue gas) = 700 °C, Tt(tube surface) = 450 °C.

3 4 C 3 µ 2 Le = k ε (5.35)

If the turbulent RMS value is used as the characteristic velocity, the eddy lifetime can be computed as:

2k t = Lee / ( ) 3 (5.36) or k = Ct t ee ε (5.37)

Cte is typically 0.15–0.37. In this thesis Cte = 0.30 is used.

48 49

the particles also hit one of the tubes, and in this study the impingement is simulated so that the particle is considered as trapped on the tube wall. As shown in figure 6, some particles impinge on the tube wall in the sur- roundings close to the first superheater frontal stagnation point. Other particles circumvent the superheater tubes by following the gas-phase flow round the tubes, and some of these are caught in the wakes behind the tubes.

Figure 6. Examples of individual trajectories for 30 particles of 20 µm diameter released at the inlet.

From analysis of the ash particles in the flue gas filter (Björkenfjäll & Eli- asson, 2003), it is evident that both submicron particles and particles in the range 50–100 µm can exist in the flue gas path even if the majority of the larger particles are captured by the cyclone filter. For large particles, the inertia is the dominant force and the particles impinge on the frontal tube on the Figure 5. Velocity range 0–14.7 m/s (up left), Velocities as iso-surfaces windward side. For particles below 1 µm (St << 1), the inertia has no influ- (up right), Velocity field around the first frontal tube (down). ence, and these particles follow the flue gas flow. However, the turbulence levels and the thermophoretic force affect the deposition of small particles on heat transfer surfaces. Compared to large particles, small particles can be ex- When 30 particles of 20 µm diameter are introduced (in the stationary solu- pected to be much more evenly distribution on the tubes. These results are tion) at ten points at the inlet boundary, with the same inlet velocity as the discussed in Paper 3 and in detail in Appendix 3. flow, the particles follow individual trajectories as shown in figure 6. Some of

50 51

Figure 6. Examples of individual trajectories for 30 particles of 20 µm diameter released at the inlet.

50 51

6. Dynamic simulations of a CFB boiler

In order to understand how deposits affect the boiler efficiency and performance a dynamic model is developed for a biomass fired circulating fluidized bed boiler. The model is based on energy and mass balances for the components in the boiler and on a combustion model for the fluidized bed. All equations in the model are developed in Modellica 2.1 using the Dymola 6.1 graphical interface. The model is calibrated and verified to Boiler 5 in Väster- ås.

6.1 Simulation model The presented model includes most of the steam tubes, the flue gas channel, and the combustion zone. The steam side of the boiler is connected to two turbines: the high pressure turbine (HP turbine) and the intermediate pressure turbine (IP turbine). In the real plant the turbines are shared by two boilers but in the simulation the turbines only support Boiler 5. The IP turbine is further simplified by excluding the steam extractions to the feedwater heaters. This results in identical steam flow rates through both turbines. Figure 7 shows a scheme of Boiler 5. Three main paths can be identified, the water-steam path, the air/fuel-flue gas path and the ash/sand circulation path. Figure 7. Scheme of Boiler 5 model.

The performance of the real boiler is controlled by a large number of controlling loops such as the steam controlling valves that control steam flow and the rotation speed of the water pumps that controls the feed water flow. The flue gas temperatures and the heat release in the combustion zone are controlled by the fuel flow rate and by recirculation of flue gas back to the combustion zone. The maximum steam temperatures are controlled by water injection. The presented model excludes all these controlling mechanisms except the water injection for temperature control, which is included in one case (Sce- nario 2, presented in Chapter 7.5). The controlling valve ahead of the turbines is not included. This corresponds to it being fully open at all times. The pressure in Boiler 5, which is a

52 53

6. Dynamic simulations of a CFB boiler

52 53

drum boiler, is not constant, but depends on the load (a sliding pressure boil- Hi = 34.1C + 102H + 6.3N + 19.1S - 9.85O - 2.5H2O (MJ/kg) (6.3) er). Based on an ideal turbine assumption (Traupel 1960) a linear relation is used between the drum pressure and the steam flow rate through the high pressure turbine. In the model the feed water flow follows the steam flow. The sulphur content (S) is zero. The model is simplified further by excluding pressure drops caused by friction. 6.3 Energy and continuity equations 6.2 Combustion model In the model there are three main mass species (point masses): the sand bed (M1_bed+gas), the intrex (M2_bed+gas) and the water and steam in tubes surround- Fuel, air and flue gas are divided into total nine mass fractions. ing the combustion zone (the bed) and in the drum (Mdrum). For each heat exchanger surfaces (i) the heat transfer rate Qi is calculated 1) Carbon (C) in fuel. by 2) Hydrogen (H) in fuel. 3) Oxygen (O or O2) – O in fuel and O2 in air and flue gas. = UQ ii A i ∆T i 4) Carbon dioxide (CO2) in flue gas. (6.4) 5) Water (H2O_l) in fuel. 6) Steam (H2O_s) in flue gas. 7) Nitrogen (N or NO2) – N in fuel and NO2 in flue gas. The temperature difference ∆Ti is the arithmetic average temperature differ- 8) Sand/ash in fuel and flue gas. ence between the two fluids in the heat exchanger based on inlet and outlet 9) Nitrogen (N ) in air and flue gas. 2 temperatures ( (Tin+Tout)hot/2-(Tin+Tout)cold/2). The same equation is used for all heat exchangers independent of the heat exchanger design. For heat Fuel enters the boiler at 25°C with specified mass fractions of six compo- transfer in the beds, the bed temperature is used as hot temperature nents: C, H, O, H2O_l, N, and Sand/ash. Air enters the boiler as specified (Tin=Tout)hot . mass fractions of two components: O2 (20%) and N2 (80%). The combustion model is based on two main chemical reactions a) Sand bed and combustion zone Gas and sand side:

C + O2  CO2 (6.1) The mass flow into the control volume is made up of four main components: air, fuel, sand and sand/ash (from the Intrex). The mass flow out of the control volume is flue gas including ash/sand.

4H + O2  2H2O (6.2)    i np −  )()( outp u e l if −+ QHmTcmTcm 1 ∂T _1 +gasbed As an approximation all the elementary nitrogen (N) in fuel is assumed to re- = in out ∂t M c Tf )( act with oxygen to form NO2. All water in the fuel evaporates to steam, and _1 +gasbed  v i ash in fuel is mixed with the sand and ash that flows with the flue gas. i (6.5) After combustion (in and above the fluidized bed) the flue gas includes six components: O2, CO2, H2O_s, NO2, sand/ash and N2. The lower heating value in the fuel is calculated from (Mörtstedt, Hellsten, ∂M _1 +gasbed −= mm 1987):  in  out ∂t in out (6.6)

54 55

For steam and water in the drum and in the evaporating tubes surrounding the m ,outsteam = m ,insteam combustion zone: (6.13)

∂T ( ()  hmhm ) +− Q c) Cyclone drum = water steam 1 ∂t ( + fcfcM ) ,watervdrum water ,s t e a mv steam (6.7) hm  , steamoutsteam ,out = mh  h ,, + Q iinsteaminsteam (6.14)

m ,outsteam = m ,insteam ∂M drum (6.15) water −= mm  steam ∂t (6.8)    ,outfluegas hm ,outfluegas + cm ,, T ,outsandsandpoutsand =  m c )( i np − QT i (6.16) in = MV ( v f + v f ) drum drum water water steam steam (6.9)   in =  mm out (6.17) in out b) Intrex d) SH1, SH2 and RH1 Gas and sand side: The mass flow into the control volume is made up of two main components: hm  , ,outsteamoutsteam = mh  , h , + Q iinsteaminsteam + m water ,injection hwater Air for fluidization and sand/ash from the cyclone. (6.18) The mass flow out of the control volume is made up of a mixture of the in- m =  + mm  put components. ,outsteam ,insteam water ,injection (6.19)

)(  − ()  ) −− QQTcmTcm m ,outfluegas h ,outfluegas = m  , , − Qh iinfluegasinfluegas ∂T  i np  outp SH 3 RH 2 (6.20) _2 +gasbed = in out ∂t M _2 + gasbed c vTf )( i  m ,outfluegas = m ,influegas i (6.10) (6.21)

e) Economizer ∂M _2 +gasbed  −= mm  in out m  , wateroutwater ,out = mh  h ,, + Q iinwaterinwater (6.22) ∂t in out (6.11) m = m ,outwater ,inwater (6.23) Steam side: For SH3 and RH2 respectively m h = m  − Qh ,outfluegas ,outfluegas , , iinfluegasinfluegas (6.24)

hm  , ,outsteamoutsteam = mh  h ,, + Q iinsteaminsteam m = m (6.12) ,outfluegas ,influegas (6.25)

56 57

For steam and water in the drum and in the evaporating tubes surrounding the m ,outsteam = m ,insteam combustion zone: (6.13)

∂T ( ()  hmhm ) +− Q c) Cyclone drum = water steam 1 ∂t ( + fcfcM ) ,watervdrum water ,s t e a mv steam (6.7) hm  , steamoutsteam ,out = mh  h ,, + Q iinsteaminsteam (6.14)

hm  , ,outsteamoutsteam = mh  h ,, + Q iinsteaminsteam m = m (6.12) ,outfluegas ,influegas (6.25)

56 57

f) Superheater convection m ,outsteam = m ,insteam (6.35)

hm  = mh  h + Q , steamoutsteam ,out ,, iinsteaminsteam (6.26)  steam = km p drums (6.36)

m = m ,outsteam ,insteam (6.27) mP =  hm − h outinsteam )( (6.37)

m h = m  − Qh ,outfluegas ,outfluegas , , iinfluegasinfluegas (6.28) i) Hot condenser

hmQ = m  ( − hh ) m = m steam ,insteam ,outwater (6.38) ,outfluegas ,influegas (6.29) g) Air preheater

hm  = mh  h + Q , ,outairoutair ,, iinairinair (6.30)

= mm  ,outair ,inair (6.31)

m h = m  − Qh ,outfluegas ,outfluegas , , iinfluegasinfluegas (6.32)

m = m ,outfluegas ,influegas (6.33) h) HP turbine and IP turbine Isentropic efficiency η = 0.85 (constant) is

− hh η = outin is − hh ,isoutin (6.34)

58 59

f) Superheater convection m ,outsteam = m ,insteam (6.35)

hm  = mh  h + Q , steamoutsteam ,out ,, iinsteaminsteam (6.26)  steam = km p drums (6.36) m = m ,outsteam ,insteam (6.27) mP =  hm − h outinsteam )( (6.37) m h = m  − Qh ,outfluegas ,outfluegas , , iinfluegasinfluegas (6.28) i) Hot condenser

hmQ = m  ( − hh ) m = m steam ,insteam ,outwater (6.38) ,outfluegas ,influegas (6.29) g) Air preheater

hm  = mh  h + Q , ,outairoutair ,, iinairinair (6.30)

= mm  ,outair ,inair (6.31) m h = m  − Qh ,outfluegas ,outfluegas , , iinfluegasinfluegas (6.32) m = m ,outfluegas ,influegas (6.33) h) HP turbine and IP turbine Isentropic efficiency η = 0.85 (constant) is

− hh η = outin is − hh ,isoutin (6.34)

58 59

Paper 4 is focused on the correlation between fuels, deposit grow rate, de- 7. Results posit chemical content and corrosion of superheater tubes in Boiler 5, Västerås, Sweden. The investigation period was total seven years.

Paper 5 presents a dynamical model of the Boiler 5 based on energy and mass balances, simulating how deposits affect the boiler efficiency and performance.

The work presented in this thesis shows the complexity of the fouling process in biomass fired boilers. Small amounts of chlorine and alkali metals in the 7.1 Paper 1 fuel can be responsible for both deposit growth and corrosion on superheaters and reheaters in a boiler. Soot particle size and flue gas temperature, veloci- Title: Measurements, theories and simulations of particle deposits on super- ties and turbulence are the main parameters that determine the actual deposit heater tubes in a CFB biomass boiler. shape on the heat exchanger tubes. Despite the use of soot blowing systems, the deposit thickness increases continuously during a running season from The paper starts with an overview of the subject fouling and deposits in boil- August to June. It is obvious that deposits on superheaters decrease the heat ers. The main subject is the measurements using a deposit probe, chemical transfer rate but the presented simulations show that in addition, deposits also analysis of the deposit and the following viscosity analyses. redistribute the heat transfer rate from superheaters to reheaters and, typically, During the 3–4 month long investigation period the fuel consisted of a the total boiler efficiency only is marginally affected. mixture of different kinds of biomass (50–60 %), peat (20–40 %) and recy- The strength of this thesis lies in the fact that the deposit process has been cled wood, i.e. wood wastes (10–15 %). The different types of normally used studied for over 7 years in a large commercial 157 MW boiler, and that both biomass fuels are: forest residues, wood, bark and small amounts of Salix. theoretical and experimental methods are used to generate and evaluate the The flue gas temperature was around 800 °C during the test period and the results. These methods include, among others, chemical analyses of deposits boiler was running at maximum power (>145 MWth). and fuels, continuous measurements of boiler process parameters such as The measurements show that the deposit grow rate is both time and tem- temperatures and flow rates, optical inspections of deposits, CFD simulations perature dependent. The deposit growth rate is as high as 25 g/m2h the first of particle paths, and dynamic simulations of the total boiler performance. hours starting with a clean probe. Depending on the temperature the grow rate A summary of all results has been reported in five individual publications: decrease to 1-6 g/m2h after an exposing period of a couple of days. The high value during the first hours is probably an effect of high oxidation rate on the Paper 1 includes an overview of the subject of fouling and deposits in clean surface. boilers. Its also includes deposit probe measurement, chemical deposit The figure below (figure 8) presents a significant temperature dependence analyses and viscosity calculations. of the deposit growth rate. Higher temperatures result in higher deposit grow rate. This indicate that the stickiness increase with temperature. The phenom- Paper 2 is focused on the effect of deposits on the heat transfer rate, espe- ena is more important than the thermophoretic forces which decrease with cially for the superheater tubes. Its also includes a long period investigation increasing probe temperature (Paper 3 and Appendix 3) resulting theoretically of the effect of soot blowing on the heat transfer rate of superheater 2 in in decreasing deposit growth rate with increasing probe temperature. Boiler 5, Västerås, Sweden. Recycled wood is a source of chlorine and zinc. These elements together with alkali metals from the biomass have the potential to form sticky com- Paper 3 presents theoretical and numerical studies of forces and particle pounds that increase the deposit growth rate. The stickiness can be expected trajectories in the vicinity of heat transfer tubes. The paper presents funda- to increase up to a maximum temperature were the melted phase dominates mental understandings for forming deposits on different locations on tubes. and the deposit decrease again.

60 61

Paper 5 presents a dynamical model of the Boiler 5 based on energy and mass balances, simulating how deposits affect the boiler efficiency and performance.

60 61

25 100

90 570 °C nr. 1 530 °C nr. 1 80 20 490 °C nr. 1

h] 70 2 570 °C nr. 1 Outlier: too high probe temp. (600-700 °C) 490 °C nr. 2 60 530 °C nr. 1 380 °C nr. 1 15 490 °C nr. 2 380 °C nr. 2 50 380 °C nr. 2 280 °C nr. 1 280 °C nr. 1 40

10 Gas temperature fluctuations 30 Fractionof sintered material [%] 20 Deposit layer growth rate, [g/m 5 10

0 0 20 40 60 80 100 120 Time, [h] 0 With coal 0 50 100 150 200 250 300 Time, [h] Figure 9. The fraction of sintering material on the probe ring as a function of time. Figure 8. The probe surface ring deposit layer growth rate over time. Another interesting aspect, of hard and loose deposit, is that it is not equally As a reference; The influence of probe surface temperature on the deposit distributed around the deposit probe. The looser part is dominating on the grow rate has been investigated by Theis et al. (2006c), on an entrained flow leeward side. The CFD simulations presented in Paper 3 show that only small reactor that simulates conditions in the superheater region of a commercial particles of less than about 5–10 µm can hit the back of a tube. Small particles boiler. The flue gas temperature was 1000 °C, the surface temperature range will also follow the gas, both in direction and in temperature, and as the gas is 475–625 °C and the fuel utilised peat, bark and straw. For peat the deposit cooled down in the boundary layer, close to the tube surface, the particles will grow rate was approximately constants independently of probe surface tem- also be cooled down possible to solid state before hitting the tube surface. On perature. For bark the deposit grows rate increase between 475 to 500 °C and the windward side larger particles may hit the tube surface in melted phase thereafter decreases. For straw the deposit grow rate three doubled between become cooled down on the tube surface and thereby get stuck. 475 to 550 °C. Above 550 °C the temperature dependence was irregular. Straw is a source of readily released potassium and chlorine. The rate of the sintering process could also be proven to increase with increasing probe temperature, considering modelling according to the Arrhenius equation. Here the hard part (not easily removed without a tool) of the deposit material is considered being characteristic for a sintered deposit. For temperatures above 500 °C the most part of the deposit becomes sintered after a few hours. For superheater tubes the steam temperature inside the tubes is typically around 400–525 °C. With increasing deposit depth, the deposit surface temperature is increasing, typically up to around 500–600 °C making the deposit sintered and thereby hard to remove by soot blowing. This process has been investigated and is presented in detail in Paper 2. Figure 10. Probe ring before and after removal of the loose porous particles.

62 63

25 100

90 570 °C nr. 1 530 °C nr. 1 80 20 490 °C nr. 1

h] 70 2 570 °C nr. 1 Outlier: too high probe temp. (600-700 °C) 490 °C nr. 2 60 530 °C nr. 1 380 °C nr. 1 15 490 °C nr. 2 380 °C nr. 2 50 380 °C nr. 2 280 °C nr. 1 280 °C nr. 1 40

10 Gas temperature fluctuations 30 Fractionof sintered material [%] 20 Deposit layer growth rate, [g/m 5 10

62 63

As a reference study an investigation by Miettinen et al. (2003) is used. This The presented result is used together with the Urbain viscosity method giving investigation was done in a 12 MW CFD boiler, with wood chips and coal in the following results (details in Appendix 5). different mixtures of fuel. The gas temperatures was slightly higher (850– 885 °C), with a probe temperature of 550 °C. Miettinen et al. (2003) meas- 1,E+06 ured the deposit grow rate of 2 g/m2h for wood and 1.2 g/m2h for coal. The Fly ash, from textile filter Super-heater 2, half depth, nr. 1 ash deposit was distributed on all surfaces of the deposit ring in the case of 1,E+05 Super-heater 2, half depth, nr. 2 wood, even if the depth was somewhat thicker on the windward side. The Probe, at 570 °C probe ring from the presently conducted probe measurements, depicted on the 1,E+04 Super-heater 2, at the surface left side in figure 10, show comparable distribution of the deposit layer. Using coal as fuel (Miettinen et al. 2003), a completely different deposit 1,E+03 distribution occurs with symmetrically concentrations on each side of the deposit ring (approximately at 120° as “Mickey mouse ears“). Finally, in the s] [Pa Viscosity, 1,E+02 reference study, a HCl-addition was mixed to the wood chips giving a deposit grow rate of 20 g/m2h with a similar distribution as for pure wood chip. 1,E+01

1,E+00 Chemical analyses 400 450 500 550 600 650 700 750 800 Chemical analyses have been performed on a few numbers of samples; super- Temperature, [°C] heater deposits, deposits on the deposit probe and on the fly ash samples. After chemical analysis on atomic basis and recalculation to oxides (Appen- Figure 12. The particle viscosity as a function of temperature, calculated dix 4) the results are presented in figure 11. from the chemical analysis.

40 Super-heater 2, half depth, nr. 1 The above figure show that there is a significant difference in viscosity be- Super-heater 2, half depth, nr. 2 tween the fly ash (upper curve) and the surface deposit from superheter 2 35 Super-heater 2, at the surface (lower curve), considering the studied temperature interval. Fly ash from textile filter Probe, at 570 °C Earlier studies (Senior and Srinivasachar, 1995) concerning sticking prob- 30 ability for ashes from coal combustion, shows a high sticking probability for ash viscosities lower than approximately 103 Pas. 25 For steam and tube temperatures of approximately 500 °C, typical for su-

20 perheater 2, figure 12 indicate that fly ash may have a viscosity being too high for the particle to get stuck.

Mass fraction,[%] 15

10 7.2 Paper 2

5 Title: Long time investigation of the effect of fouling on the superheaters in a circulating fluidized biomass boiler. 0 C Na2O MgO Al2O3 SiO2 P2O5 SO3 Cl K2O CaO MnO Fe2O3 TiO2 BaO In this paper an investigation has been conducted on how the deposit and the heat transfer rate for superheater 2 has developed during a three year period. Figure 11. Chemical analysis of the deposit layers. These results are further explored in Paper 4 (A 7 year long measurement pe-

64 65

20 perheater 2, figure 12 indicate that fly ash may have a viscosity being too high for the particle to get stuck.

Mass fraction,[%] 15

10 7.2 Paper 2

64 65

riod investigating the correlation of corrosion, deposit and fuel in a biomass illustrated by the fact that there is no significant difference in the deposit layer fired circulating fluidized bed boiler). structure whether a tube is situated near or far from a soot blowing lance The investigation in Paper 2 also includes analysis of how effective the ahead of superheater 2. soot blowing system is in Boiler 5. In a short time scale, hours after a soot Considering the long-term aspects there is also no significant difference in blowing cycle, the increase of the heat transfer rate in superheater 2 is quite the deposit layer growth rate and in the heat transfer rate, varying the soot significant. The soot blowing thus increases the heat transfer by approximate- blowing cycle frequency from 1 to 3 times a day. The hard part of the deposit ly 0.5–1 MW by removing some of the deposit material. This corresponds, layer material seems to continue to grow more or less independently of the judging from the results from the theoretical deposit model (presented in Pa- soot blowing frequency. Temperatures (Paper 1) and fuel mixture (Paper 4) per 2), to a change in deposit layer thickness of approximately 5 mm. This seems to be as important as the soot blowing frequency, explaining the depos- result is simply impossible as it corresponds to a 3–4 months layer growth pe- it grow rate. riod based on a calibrated thermal conductivity of 1 W/m K. A more likely explanation for the change in heat transfer rate is that the soot blowing jets remove only the loose part of the deposit, which has a much lower thermal conductivity than the hard parts. This is in agreement with Pronobis (1994) 7.3 Paper 3 who reported a thermal conductivity of 0.4 to 1.5 for high temperature hard Title: Numerical simulation of fouling on superheater tube walls. deposits but only 0.1–0.2 for dust sedimentation, considering a coal fired boiler. In this paper the result of the numerical CFD calculations is presented showing that larger particles will hit mainly on the windward side of the first

10.00 frontal tube. The result agrees very well with Baxter and DeSollar (1993) who explored that particles, larger than 10–15 µm mainly are affected by inertia 9.00 and therefore only can hit the windward side. The larger particles also hit on 8.00 two symmetrical regions at the left and the right side of the second tube pre-

7.00 sented in figure 14. This could explain the observations, from superheater 2 in Boiler 5, Västerås, where two tubes are merging together by deposit growing 6.00 Heat transfer rate Soot-blowing from the second tube towards the frontal tube. Soot-blowing steam mass flow rate 5.00

4.00

3.00

soot-blowing steam mass flow, [ton h-1] 2.00 Heat transfer rate of super-heater 2, [MW] and 1.00

0.00 200 210 220 230 240 250 260 270 280 290 300 Hours from 031201

Figure 13. Detail study (nr. 4); the heat transfer rate (MW) in superheater 2, and the soot blowing interval (December 2003). Figure 14. Fouling distribution on the two tubes with large particles, 50 µm. The hard part of the tube deposit material seems to be nearly unaffected by the soot blowing procedure and will continue to grow over time. This can be

66 67

4.00

3.00

soot-blowing steam mass flow, [ton h-1] 2.00 Heat transfer rate of super-heater 2, [MW] and 1.00

0.00 200 210 220 230 240 250 260 270 280 290 300 Hours from 031201

66 67

Small particles, with diameters less than 1 µm, can impinge on both tubes and velocity, 2 m/s, decrease the probability to hit the frontal tube but increase the on all sides of the tubes (figure 15) and are therefore more equally distributed probability to hit the second tube somewhat. This effect is most significant for compared to larger particles (> 10–15 µm). The simulations with submicron larger particles. particles also show impingements on the side of the tubes forming “Mickey For small particles, the turbulence eddy timescales as well as thermo- mouse ears”. The result is probably an effect of the fact that a stationary gas phoretic forces also affects the result. For the 1 µm particle the probability of flow field solution has been used (with well-defined separation points) to- impingement decreases from 3.2 to 2.2 % for the frontal tube, and decreases gether with the stochastically particle path simulation. If the calculations from 3.5 to 1.9 % for the second tube without the thermophoretic force. The would have been conducted in transient mode, instabilities and oscillations fact, that the thermophoretic force can increase the fouling values, has also would possibly have spread the impingements more evenly on the tube sides. been investigated by others e.g. Greenfield and Quarini (1998) and Baxter Deposit distributions such as “Mickey mouse ears” has not been observed on and DeSollar (1993). the tubes in Boiler 5 or during the measurement with the deposit probe, as re- Furthermore the simulations also show that the fouling values increase if ported in this thesis. However, Miettinen et al. (2003) reported that from the turbulence timescales increase. experiments when firing a CFB boiler with pure coal, the deposits concentrations was actually formed as “Mickey mouse ears”. 100

In observations and experiments in real plant conditions it is obviously not frontal tube 90 possible to separate and study the aspects of particles having a specific di- second tube ameter. Samples and analysis of the ash particles in the flue gas filter of 80 Boiler 5 (Björkenfjäll and Eliasson, 2003) show that both submicron particles as well as particles in the size of 50–100 µm may exists in the flue gas path 70 but the main part of particles larger than 50–100 µm are separated in the cy- 60 clone. 50

40 Probability to hit tubea (%) 30

0 0,1 1 10 100 Particle size, [µm]

Figure 16. Probability of impingement as a function of particle size (8 m/s).

Particles less than 10 µm correspond to Stokes number less than 0.1 Figure 15. Distribution of particles in size 1 µm. A complete presentation of the result is given in Appendix 3.

The probability of impingement on a tube is lower for small particles as seen in figure 16. The result presented is from simulations with an inlet velocity of 8 m/s. In the simulations up to 400 000 particles were released for each investigated particle size in order to get reliable average values. Lower inlet

68 69

40 Probability to hit tubea (%) 30

0 0,1 1 10 100 Particle size, [µm]

Figure 16. Probability of impingement as a function of particle size (8 m/s).

Particles less than 10 µm correspond to Stokes number less than 0.1 Figure 15. Distribution of particles in size 1 µm. A complete presentation of the result is given in Appendix 3.

68 69

7.4 Paper 4 Title: A 7 year long measurement period investigating the correlation of corrosion, deposit and fuel in a biomass fired circulating fluidized bed boiler.

In this paper the investigation involves a seven year (2001–2007) long study of corrosion and deposits on superheater tubes in a biomass fired circulating fluidized bed boiler. These measurements are correlated against the different fuels utilised these years.

Fuel mixtures The biomass fuel mixtures varied from week to week, and sometimes from day to day over the whole investigation period 2001–2007. The fuel mixtures utilized during the boiler start and the first season 2001–2002 consisted of a mixture of different kinds of biomass (50–60 %) and peat (20–40 %). A larger proportion of recycled wood (10–15%) was introduced in the winter/spring 2003. Figure 18. Fuel mixtures used as percentages by volume during 2002– Figure 17 shows fuel use over the season 2001–2002 (from when the boiler 2003. was started in January 2001). The boiler did not run at full load in late spring and early autumn, but the 100% values in the graph relate to the actual Figure 19 shows fuel used in the season 2003–2004. Recycled wood was used amount of fuel used in each case, independent of the capacity used. at 10–15% throughout the season except for a few days during the boiler start up, when the fuels used were wood and bark. Peat was also used in larger amounts in this period.

Figure 17. Fuel mixtures used as percentages by volume during 2001– 2002. Figure 19. Fuel mixtures used as percentages by volume during 2003– Figure 18 shows fuel used in the season 2002–2003. 2004.

70 71

7.4 Paper 4 Title: A 7 year long measurement period investigating the correlation of corrosion, deposit and fuel in a biomass fired circulating fluidized bed boiler.

70 71

Figure 20 shows fuel used in the season 2005–2006. Unlike in 2004–2005, recycled wood was first introduced several weeks after boiler start. Peat usage was lower, and recycled wood was not used at the end of the season.

Figure 21. Deposit layer thickness (mm) on the windward side on superheater 2, superheater 1, and reheater 1 (2001 to 2007), (ReH1,p2 year 2002 not measured).

The deposit was thickest on the windward side of the first part (first tube) of superheater 2 (SH2,p1) and decreased down the flue gas path as shown in figure 22 (which shows data from 2004). There was a much thinner deposit Figure 20. Fuel mixtures used as percentages by volume during 2005– layer on the leeward side compared to the windward side in all the heat ex- 2006. changers investigated.

Deposit thickness The deposit thickness on the windward side of superheater 2, superheater 1, reheater 1 (figure 21) was initially zero in January 2001 (the tubes were new), and after each annual revision during 2003–2006, due to cleaning. The tubes were not cleaned in 2001 and 2002. After the season 2003–2004 there was a maximum deposit thickness of 32 mm on the windward side of superheater 2.

Figure 22. Deposit thickness on superheater 2, superheater 1 and reheater 1 after season 2003–2004, measured August 2004 (SH2,p2 not measured).

The overall heat transfer coefficient (U) For superheater 2, which is a parallel flow heat exchanger, the heat transfer rate Q (W) and overall energy balance is given by:

72 73

Figure 21. Deposit layer thickness (mm) on the windward side on superheater 2, superheater 1, and reheater 1 (2001 to 2007), (ReH1,p2 year 2002 not measured).

Figure 22. Deposit thickness on superheater 2, superheater 1 and reheater 1 after season 2003–2004, measured August 2004 (SH2,p2 not measured).

The overall heat transfer coefficient (U) For superheater 2, which is a parallel flow heat exchanger, the heat transfer rate Q (W) and overall energy balance is given by:

72 73

hmQ = m  ∆ = mh  c T =∆ A ∆ TU (7.1) The overall heat transfer coefficient decreased between January 2001 and steam p gasflue ln August 2003. After the cleaning of the tubes in August 2003 the relative efficiency of superheater 2 returned to approximately 1, but quickly decreased again as the superheater tubes were again covered by deposits. The heat trans- In this equation A is the known circumferential area of the superheater tubes fer coefficients recovered to approximately 1 after each annual cleaning. The and ∆Tln is the logarithmic temperature difference between the flue gas and slope of the curves indicates the deposit growth rate. A steeper slope indicates steam. a higher deposit growth rate. The heat transfer rate (Q) was calculated from the first part of the equation Comparing for example the season 2003–2004 with 2004–2005, the slopes above (massflow (kg/s) multiplied by enthalpy difference (kJ/kg) for the were about the same but the season 2003–2004 was longer (at full load). steam). U (W/m2 K) was then calculated using the value of Q thus obtained. From the delivery test in December 2000 (Ericsson 2004) a reference value Tube metal thickness Tube thickness measurements were performed on four on superheater 2 from for the overall heat transfer coefficient was calculated. Here, Uref = 68.2 2 2001 to 2006 points. Points A, B and C were on the windward side of the first (W/m K). frontal tube of bundle 1 (SH2,p1) and point D was on the first frontal tube on The overall heat transfer coefficient (U) for superheater 2 were calculated bundle 2 (SH2,p2). Superheater 2 consists of 37 parallel tubes with outer di- at full load condition (> 145 MW) according to the method presented. Only ameter 44.5 mm and a nominal tube wall thickness of 7.1 mm. data recorded in the hour following a soot-blowing cycle were used for fur- The mean values of the tube wall thickness for all the 37 parallel tubes at ther investigations. These calculated values were divided by the reference U- the points A, B, C and D are plotted in figure 24. The tube wall was thinnest value and presented as relative overall heat transfer coefficient for the period at point A. 2001–2007 in figure 23.

7.50

7.00

6.50 A 6.00 B 5.50 C

Tube Thickness Thickness Tube mm 5.00 D 4.50 2001 2002 2003 2004 2005 2006

Year

Figure 24. Mean Tube thickness (mm) as a function of year, (Point D was not measured in 2005).

Figure 23. The relative overall heat transfer coefficient of superheater 2 All tubes showed a decrease in wall thickness of about 0.3 mm over the sea- over a seven years period. son 2003–2004. The low value for thickness at point A in 2005 appears

74 75

7.50

7.00

6.50 A 6.00 B 5.50 C

Tube Thickness Thickness Tube mm 5.00 D 4.50 2001 2002 2003 2004 2005 2006

Year

Figure 24. Mean Tube thickness (mm) as a function of year, (Point D was not measured in 2005).

74 75

inconsistent. The most likely explanation for this is that the measured position Table 2. Chemical analyses (weight %) of deposit on SH2, outward sur- in 2005 was not exactly the same as those measured in 2004 and 2006. face.

The chemical content of deposits on superheater 2 The chemical analyses of the deposits samples are presented in tables 1 and 2. Year C O Na Mg Al Si P S Cl K Ca Ti Mn Fe Zn The outer side of the sample corresponded to the most recent deposit (on the windward side) just before the boiler shut down in July/August each year 2002 26.8 33.6 0.6 1.9 1.1 2.2 0.9 10.7 7.1 13.1 0.4 1.5 2002 to 2007. The inner side of the same sample corresponded to a deposit surface 1–3 mm from the tube surface and therefore to deposits from early in 2003 4.5 37.8 1.3 2.3 2.1 4.6 1.2 14.5 11.0 18.4 0.5 1.7 the running season, just after the revision period in July/August. Excluding oxygen, the main components in the deposits were potassium, 2004 4.8 38.5 4.6 1.9 1.7 3.7 0.8 15.2 11.1 13.8 0.2 0.5 2.7 0.5 sulphur and calcium, and the most abundant element was calcium It is also important to note that small amounts of the trace elements chlorine, 2005 7.8 35.8 3.6 3.0 2.4 5.3 1.4 10.5 0.1 8.6 15.4 0.4 0.9 4.8 titanium and zinc were found from 2004 onwards. 2006 5.7 34.8 4.0 2.4 1.9 4.6 1.0 13.0 13.4 15.5 0.3 0.9 2.4 Table 1. Chemical analyses (weight %) of deposit on SH2, inner surface. 2007 8.5 34.5 1.2 1.5 2.1 5.5 1.0 15.0 12.2 12.3 0.3 0.6 5.3

Mean 9.7 35.8 2.6 2.2 1.9 4.3 1.1 13.2 0.1 10.6 14.8 0.3 0.6 3.1 0.5 Year C O Na Mg Al Si P S Cl K Ca Ti Mn Fe Zn Result and Discussion 2002 4.5 34.3 1.5 2.4 2.7 11.1 1.3 6.7 8.6 21.3 0.9 4.6 The results of this study suggest that the use of recycled wood as fuel has important implication for both deposits and corrosion. When recycled wood was 2003 9.1 35.5 2.9 2.7 1.9 5.3 1.4 10.6 10.8 15.3 0.8 3.8 used for longer periods in winter-spring 2003 the increase in deposit thickness was significantly slowed, as indicated by the value of U (figure 23). When the 2004 9.5 32.9 3.5 1.4 1.2 3.6 0.8 13.9 0.2 15.1 14.2 0.4 3.3 deposits were thick the surface temperature approached that of the flue gas 800–850 ºC (Paper 1). Higher surface temperatures and even low levels of el- 2005 5.2 35.8 0.9 1.4 1.6 5.3 0.7 13.6 0.0 10.1 21.1 0.4 0.7 3.1 ements that form low melting temperature substances may be able to melt the deposit. 2006 5.4 34.8 1.9 1.7 1.7 5.1 0.8 12.4 0.1 12.0 19.6 0.4 1.0 2.5 0.7 From season 2003–2004,when larger proportion of recycled wood (10– 15%) was utilised from the boiler start in August each year with cleaned 2007 6.7 36.1 1.8 1.4 2.3 7.3 0.6 12.9 11.3 13.8 0.4 0.6 4.9 tubes, the slopes (and therefore the deposit growth rates) were around double those of the first two seasons (2001–2003). Mean 6.7 34.9 2.1 1.8 1.9 6.3 0.9 11.7 0.1 11.3 17.6 0.4 0.7 3.7 0.7 The flue gas temperature is also of great importance in fouling. There was a drop in flue gas temperature from 800–850 ºC ahead of superheater 2 to around 400–450 ºC at the outlet of reheater 1. The deposition rate of potassium chloride is significantly affected in this temperature range (Åmand et al. 2006). At 850 ºC, KCl in the gas phase can condense on cooler tubes. At lower temperatures KCl may be present in the melted phase as submicron particles or on larger ash particles. At temperatures such as those in the outlet

of reheater 1, KCl is most likely to be in the solid phase, meaning it is less likely to stick to the tubes. This may explain why the amount of deposits de-

76 77

of reheater 1, KCl is most likely to be in the solid phase, meaning it is less likely to stick to the tubes. This may explain why the amount of deposits de-

76 77

creased from a maximum on SH2,p1 to nearly no deposit at all on The deposits on superheater 2 were very hard and dense and could not be ReH1,p2/p3 (figure 21,22). removed with the soot blowers. Once ash particles are stuck on a tube, a sin- The deposits on the windward side of the tubes were much thicker than tering process occurs between the particles, making the deposit more dense. those on the leeward side of the same tubes (figure 22). The theoretical simu- This process is time and temperature dependent (Paper 1). Other chemical relation of deposit thickness growth rate presented in Paper 3 showed that, due actions besides the sintering processes may also occur. If potassium chloride to inertia, larger ash particles (more than 5–10 m) mainly encounter the is present in the deposit, it may react with sulphur, either from the deposit it- windward side of a tube but smaller submicron particles also encounter the self or from the flue gas (in the form of SO 2), releasing chlorine. Chlorine leeward side of the same tube. One possible explanation is that larger parti- may remain in the deposit but may also react with other species, enabling it to cles mainly build up the deposit and components as alkali chloride and zinc escape the deposit. The amount of chlorine in the deposit would thereby de- chloride may act as glue for these larger ash particles. This could also explain crease and the amount of sulphur would increase. The amount of chlorine how small amounts of chlorine and zinc in the deposits (tables 1 and 2) may shown in tables 1 and 2 may therefore be lower than the actual amount of be responsible for large deposit growth rates. chlorine produced during the initial stages of the deposition process. The results for reheater 1 (ReH1,p2 and ReH1,p3 in figure 22) show that Excluding oxygen, the main components in the deposits were potassium, there was only a thin deposit on the windward side and nearly no deposit at sulphur and calcium (tables 1 and 2), and the most abundant element was cal- all on the leeward side. A possible explanation for this is that some larger par- cium. Calcium is often associated with coarser ash particles (Elled et al. ticles may still have been sticky when hitting the windward side of a tube and 2008). As calcium competes with potassium and sodium for sulphur (to form may have got stuck there. The submicron particles that could possibly hit the CaSO4 or CaSO3), it can play an important role in the fouling process. Åmand leeward side could have been mainly in the solid state, making them less et al. (2006) reported large quantities of both CaSO4 and KCl in ash particles sticky. Furthermore, there may have been no condensation on the windward of size 1–10 µm in co-combustion biofuels and sewage sludge. or the leeward side. An interesting observation is that the deposit growth rates (the slope in fig- Ash particles can also have a scavenging effect on deposits. As the tem- ure 23) remained large and relatively constant from 2003–2007 but the perature decreases in the flue gas path ash particles become harder and the corrosion rate was only large in the season 2003–2004 (figure 24). A possible probability of eroding the deposits increases, thereby limiting the deposits on reason for the increased corrosion may be that the longer running periods af- reheater 1 (Ots, 2001). ter the summer revision period during which fuel mixes that excluded The inertia of small submicron particles is of minor importance, meaning recycled wood were used could cause buildup of non-corrosive deposits, that these particles follow the flow of the flue gas and other small scale turbu- which could protect the tubes from further corrosion. The running period be- lent fluctuations (Paper 3). On the other hand small particles can be affected fore the introduction of recycled wood was 2 days in 2003, 5 days in 2004, 14 by molecule diffusion and thermophoresis as a result of the temperature dif- days in 2005 and 18 days in 2006. However, chlorine is a small molecule ference between the flue gas and the surface. A possibility is that the deposit which is likely to be able to diffuse through a deposit layer, rendering the pro- rate on superheater 2 is affected by thermophoresis but for reheater 1 this ef- tective oxide layer on the tube surface ineffective. fect is less important as the temperature difference between the flue gas and the tubes is much smaller. The steam temperatures are about the same for superheater 2 and reheater 1 (in the range 400–500 ºC), but the flue gas temperature is much higher for superheater 2. 7.5 Paper 5 The sulphur:chlorine ratio is greater than 2 in most of the years investigat- Title: Dynamic simulation of fouling in a circulating fluidized biomass fired ed and greater than 4 (the recommended value to reduce fouling) in others. If boiler. the sulphur content is high, sulphur is likely to react with alkali metals such as potassium to form sulphates, releasing chlorine in the process. Chlorine may In this paper the result of the simulation of how deposits affect the boiler per- react to form HCl, but may also form zinc chloride in the presence of zinc. formance is presented. The model is calibrated by adjusting the product AU However, fouling also depends on parameters as fuel type, bed material and for the heat exchanger surfaces and verified to Boiler 5 in Västerås. temperature. When the boiler runs at maximum load the maximum steam temperature is controlled by high pressure water injections. The water injections affect the

78 79

heat balances, significant reducing steam temperature. However, they also increase the complexity of theoretical studies of the effect of fouling on boiler performance. In the calibration procedure the fuel flow and airflow values are reduced as the water injections are omitted. Following the calibration procedure, three scenarios are set up for calculations:

Scenario 1 (based on the calibrated model) deposits are introduced gradually in superheater 2, superheater 1, superheater convection and reheater 1 by halving U of each area in turn (i.e. equivalent to halving the UA as the heat exchanger surface area is constant).

In Scenario 2 the model is recalibrated as the real boiler is running at full load with water injections in the superheaters and reheater to control and limit the maximum steam temperature to 540 °C.

In Scenario 3 (based on the calibration procedure in scenario 1) deposit is simulated on the evaporation tubes by halving the value of U. Figure 25. Heat transfer rates of superheaters and reheater 1 as a function of reduced U. Result scenario 1 Simulations with the presented model show that fouling on superheaters redistributes the heat transfer rate from the superheaters to reheater 1 and also shifts turbine power from the HP turbine to the IP turbine. The total efficiency of the boiler is only marginally affected and the dynamic effects are small as long as the deposits are mainly in the superheaters. The decrease in HP turbine power is mostly offset by increased IP turbine power but there is a small loss of total turbine power and a small increase in heat flow in the hot condenser.

Figure 26. Turbine power as a function of fouling.

There are two main reasons for the nearly complete recovery of heat transfer rate from the superheaters by reheater 1. First, the flue gas temperature ahead

80 81

Figure 26. Turbine power as a function of fouling.

There are two main reasons for the nearly complete recovery of heat transfer rate from the superheaters by reheater 1. First, the flue gas temperature ahead

80 81

of reheater 1 increases as the heat transfer rate for the superheaters decreases because of the decreased U. The second reason is that the inlet steam temperature into reheater 1 decreases. The inlet steam temperature into reheater 1 is the same as the outlet steam temperature from the HP turbine, and as the inlet steam temperature into the HP turbine decreases, the outlet steam temperature from the HP turbine also decreases (and the turbine power only slightly decreases). The overall effect causes the temperature difference between the flue gas and steam to increase significantly as does the heat transfer rate of reheater 1. However, the outlet flue gas temperature from reheater 1 is almost unaffected. The overall effects on the rest of the boiler variables, combustion, furnace temperature, steam flow rate etc. are marginal. As the flue gas temperature out of the air preheater is unchanged, the total boiler efficiency remains relatively constant. The increase in power of the IP turbine does not completely offset the reduction in power of the HP turbine (figure 26) and as a result there is a slight increase of heat in the hot condenser.

Result scenario 2 The simulation shows that the fouling effect can be counteracted (i.e. steam temperature at turbine inlets can be maintained at 540 °C) by decreasing and Figure 27. Steam temperature at turbine inlet as a function of fouling and stopping, the water injections in superheaters 1 and 2 and increasing them in water injection. reheater 1. The bars in figure 27 represent water injection (kg/s) and the lines show temperatures at the turbine inlet. The steam inlet temperature into the IP Examination of real data from Boiler 5 in Västerås (e.g. comparing November turbine is kept constant by increasing water injection. The temperature at the 2007 with April 2008) shows that the water injection for reheater 1 increased HP turbine inlet eventually falls below 540 °C despite reduction and stopping while those for the superheaters decreased from autumn to spring. This agrees of water injection. well with the simulations in Scenario 2, although a precise comparison is difficult.

Result scenario 3 A step change of the value of U (halving the calibrated UA) in the combustion zone results in decreased heat transfer rate, decreased evaporating temperature (figure 28) and therefore a decreased evaporating pressure. The steam flow also decreases as the pressure decreases. As the fuel flow is constant (in the simulation), the combustion temperature and bed temperature (T_bed) increase dramatically from 823 to 995 °C. The bed temperature in intrex (T_bed_intrex) also shows a delayed increase to above 860 °C. After 1000 seconds the inlet steam temperature into the turbines increases to just below 700 °C. Under these conditions in the real boiler, the water injection limits the steam temperatures to 540 °C and a number of additional control systems are activated. These include the flue gas circulation system, which maintains the bed temperature below 850 °C. These control systems also reduce the fuel flow and the inlet air flow.

82 83

In the simulation of Scenario 3, in the absence of the additional controlling systems, the increased flue gas temperature results in increased heat transfer 8. Discussion rates in all heat exchangers in the flue gas channel. The increased heat transfer rate in the economizer results in higher feed water temperature out of the economizer. As the pressure in the drum decreases the feed water begins to evaporate in the economizer.

Theoretical prediction of fouling development on a superheater tube requires prediction of a long chain of mechanisms: from fuel characteristics, combustion model, fluid dynamics, and ash chemical composition to mechanisms that control the deposition of ash, such as inertial impact, turbulence, thermophoresis, condensation and chemical reactions, etc. Both experimental and theoretical estimates are necessary to increase knowledge of the subject of fouling. One aim of the thesis is to identify some of the most important phenomena and parameters responsible for developing deposits on heat exchanger tubes. A second aim is to understand how these deposits affect the boiler performance. The results of this study suggest that the use of recycled wood as fuel has important implications for both deposits and corrosion. The elements in recycled wood can also explain some of the main deposit growth mechanisms seen when using biomass fuels. Deposit growth rate and corrosion rate both increased significantly following the introduction of recycled wood into the biofuel from the start of the autumn 2003 burning season, after the revision period when the tubes were cleaned. When burning recycled wood, there are

higher levels of chlorine (Cl), zinc (Zn) and titanium (Ti) in the fuel. In this Figure 28. Steam temperature at turbine inlet after fouling on evaporator study very small amounts of these three elements were also detected in the surfaces. deposits on the tubes when recycled wood was added to the fuel. As well as oxygen, the main components in the deposits were potassium (K), sulphur (S) and the most abundant element was calcium (Ca). Chlorine can react with potassium to form KCl, a sticky compound which causes a high deposit growth rate. Chlorine can also react with excess zinc to form ZnCl2, another sticky component. Titanium and its compounds have much higher melting temperatures compared to Cl and Zn, and are not expected to be of great importance in the deposition process. The role of sulphur has not been thoroughly investigated. The general rec- ommendation is that the sulphur:chlorine ratio must be between 3 and 4, and at least above 2 to avoid problems with alkali chlorides (Salmenoja, 2000). If the sulphur content is high, it is likely to react with alkali metals such as potassium to form sulphates, releasing chlorine in the process. Chlorine may react to form HCl, but may also form zinc chloride in the presence of zinc. The sulphur:chlorine ratio in the period under investigation in this thesis was

84 85

greater than 2 in most of the years investigated and greater than 4 in the other probability of eroding the deposits increases, thereby limiting the deposits on years. In the season 2003–2004 (in which there were high deposit and corro- reheater 1. The “wavy” pattern of the deposits that is often seen on the wind- sion rates) the ratio was remained above 3. However, fouling also depends on ward side of tubes might be explained by this erosion. parameters such as fuel type, bed material and temperature. For example, high The aim of soot blowing is to remove deposits. Over a short time scale, in levels of calcium in the fuel (such as those found in peat) can absorb the sul- the first hour after a soot blowing cycle, there is a significant increase in the phur, to form CaSO4, therefore attenuating its neutralising effect on KCl in heat transfer rate due to the removal of the loose part of the deposit. However, the flue gas (Åmand et al. 2006). if the tube surface temperature is above 500–550 ºC it only takes a few hours The deposits on the windward side of the tubes were much thicker than for the loose deposits to start to sinter to a hard deposit (Paper 1). In the long those on the leeward side of the same tubes. The theoretical simulations of term study in 2001–2003 there was surprisingly no significant difference in deposit thickness growth rate showed that ash particles larger than 5–10 m the deposit layer growth rate, when the soot blowing cycle frequency was var- in diameter mainly encounter the windward side of a tube due to inertia, but ied from 1 to 3 times a day. The soot blowers were not able to remove the smaller submicron particles also encounter the leeward side of the same tube. hard deposits. The deposit growth rate was highest in autumn 2003 in spite of One possible explanation for this is that larger particles mainly build up the three daily soot blowing. The fuel composition was not investigated in detail deposit and components such as alkali chlorides and perhaps zinc chloride in the study in Paper 2, but Paper 4 indicates that the use of recycled wood in may act as glue for these larger ash particles. This may also explain how the fuel is more significant than the soot blowing frequency for deposit de- small amounts of chlorine and zinc in the deposits may be responsible for velopment. high deposit growth rates. Eutectic salts may also be of importance because a It is obvious that deposits on heat exchanger surfaces decrease the heat eutectic alkali salt of e.g. KCl-K2SO4-Na2SO4 may have a lower melting transfer rate. Paper 2 shows that a 30 mm deposit on superheater 2 reduces points compared to an individual alkali salt e.g. KCl. the heat transfer rate by approximately half. The investigation presented in The flue gas temperature plays an important role in fouling. There is a drop Paper 5 shows that deposits on the superheaters mainly redistribute the heat in flue gas temperature along its flow from 800–850 ºC ahead of superheater transfer rate from the superheaters to reheater 1 and also shift turbine power 2 to around 400–450 ºC at the outlet of reheater 1. The deposition rate of po- from the HP turbine to the IP turbine. The total efficiency of the boiler is only tassium chloride is significantly affected in this temperature range. At 850 ºC marginally affected. There is always a cost incurred by soot blowing (loss of KCl in the gas phase can condense on cooler tubes. At lower temperatures intermediate pressure steam), and for this reason, we propose that the soot KCl may be present in the melted phase as submicron particles or on larger blowing frequency should be minimised. Conversely, deposits also redistrib- ash particles. At temperatures such as those in the outlet of reheater 1, KCl is ute the temperatures in the flue gas channel, possibly compounding the most likely to be in the solid phase, meaning it is less likely to stick to the problems with corrosion, as the temperatures are higher there than the design tubes. This may explain why deposits decreased from a maximum on the first temperatures further down in the flue gas channel. Excessive deposits can al- tubes of superheater 2 (SH2,p1) to nearly no deposit at all on tubes at the out- so decrease the flue gas free areas, increasing the flue gas pressure drop. let of reheater 1 (ReH1,p3). The results for tubes at the outlet of reheater 1 (ReH1,p2 and ReH1,p3) show that there was only a thin deposit on the windward side and nearly no deposit at all on the leeward side. A possible explanation for this is that some larger particles may still have been sticky when hitting the windward side of a tube and may have been cooled down and stuck there. The submicron particles – which are mainly influenced by turbulence and can therefore hit the leeward side – could have been in the solid state (as they are cooled down in the thermal boundary layer surrounding the tube), making them less sticky. Furthermore, there may have been no condensation on the windward or leeward side. Ash particles can also have a scavenging effect on deposits. As the temperature decreases in the flue gas path, ash particles become harder and the

86 87

and melt states. These elements are of great importance to the deposition 9. Conclusions process.

- Small amounts of chlorine and zinc were found both in the recycled wood and in the deposit layer. These elements together with alkali metals from the biomass may cause both high deposit growth rates and rapid increases in corrosion rates.

The aim of this work was to identify the main and most important factors that - Running the CFB boiler with a biomass fuel that does not contain large can reduce the negative effects of fouling. The results show that small amounts of chlorine and zinc immediately after the cleaning and revision amounts of chlorine, and possibly zinc, together with alkali metals from the period may provide some protection against further corrosion. biomass, may be responsible for both high deposit growth rates and rapid increases in corrosion rates. Fouling problems may be reduced by controlling Answer to Question 2: the fuel content of these elements. - Large particles (> 5–10 µm) form a deposit on the windward side of the The concrete research questions were: first frontal tube and on two symmetrical points on the sides of the second tube in the flue gas path. The deposit on the second tube can grow very Question 1: thick, eventually causing the tubes to merge together. Submicron particles What are the importance of the fuel and the combustion conditions with re- such as condensate potassium chloride (KCl) may act as glue for these spect to deposition and corrosion on heat transfer tubes in the flue gas larger ash particles. Eutectic alkali salts, with low melting point, may also channel (in a biomass fired CFB boiler)? be of importance in the deposit process.

Question 2: - KCl may be present in the gas phase at the inlet to the superheater zone, What are the major mechanisms of buildup of deposits on superheater condensing on cooler tubes. At lower temperatures, KCl may be present in tubes? the melted phase as submicron particles or on larger ash particles. At temperatures such as those at the outlet of reheater 1, KCl is most likely to be Question 3: in the solid phase, meaning it is less likely to stick to the tubes. This may How effective is soot blowing at removing deposits? explain why the deposits decreased along the flue gas path. In the presence of zinc, chlorine can react to form zinc chloride (ZnCl2) a sticky low melt- Question 4: ing temperature compound that can affect the deposit process. How is the overall boiler performance affected by deposits on heat transfer surfaces? - Small particles can impinge on all sides of the tubes, and the deposit rate is highly dependent on the turbulence in the flue gas flow. Small particles Based on theoretical calculations, simulations and practical investigations of a can result from condensation of gases and seems to produce a more porous biomass fired CFB boiler, the answers to these questions are: deposit which is easier to remove.

Answer to Question 1: - The deposit rate increases with increasing flue gas temperature but also - The details of the combustion conditions in a CFB boiler are very com- increases with increasing surface temperature (an effect of increasing prob- plex and are affected by parameters such as the elements in the fuel, ability of sticking) and thereby overrides the thermophoretic forces that combustion temperatures, the composition of the bed material, reducing or decrease with increasing surface temperature (at constant flue gas tempera- oxidation conditions, etc. At normal combustion temperatures of 800– ture). This is due to the decrease in the temperature difference between the 850 °C alkali metals, chlorine and zinc mainly follow the flue gas in vapor flue gas and the surfaces. In the superheater and reheater zones the flue gas temperature drops from 850 °C to below 450 °C while the tube metal tem-

88 89

and melt states. These elements are of great importance to the deposition 9. Conclusions process.

88 89

perature is approximately 400–500 °C. A possibility is that the deposit rate on superheater 2 is affected by thermophoresis but for reheater 1 this effect 10. Future work is less important.

Answer to Question 3: - The soot blowing system is very effective at removing loose porous deposits but is not effective at removing hard deposits. Changing the soot blowing interval from 1 to 3 times a day does not have a significant effect on the deposit growth rate over the three year investigation period. The use Although the presented work is both broad and deep in many disciplines there of recycled wood has a larger impact on the deposit growth rate than the are many possible areas for further investigation in the fouling field. The next soot blowing interval. step could be to do a more detailed study of soot blowing. Soot blowing decreases the thickness of the deposit layers, but also consumes steam, - The experiments with a deposit probe reveal that the sintering process is decreasing and redistributing the electricity power production. The relation- very fast, a matter of hours for surface temperatures above 500 °C, making ships between biomass fuel composition, deposits on tubes and soot blowing the deposits difficult to remove by soot blowing. has not been explored in detail and is unknown. In the longer term, the dynamic model of the boiler has to be developed Answer to Question 4: and improved in many ways. A more detailed combustion model that includes - Fouling on superheaters redistributes the heat transfer rate from the su- ash components such as alkali metals and chlorine will make it possible to perheaters to reheater 1 and shifts turbine power from the HP turbine to the follow these components through the flue gas path. The possibility of deposit IP turbine. When the boiler is running at maximum load, water injection formation depends on the chemical composition of ash, its temperature and ahead of reheater 1 must increase to maintain temperatures within maxi- phase (evaporated, melt or solid state), and also on velocities, particle sizes, mum allowed limits. turbulences, etc. A detailed boiler model will improve the knowledge of boiler performance, availability and lifetime. It will also be of great interest to - The total efficiency of the boiler is only marginally affected and the dy- operators who will be able to run a simulation in parallel with the real boiler namic effects are small as long as the deposits are mainly concentrated on in order to better understand both expected and unexpected events. the superheaters. The decrease in HP turbine power is mostly offset by increased IP turbine power but there is a small loss of total turbine power and a small increase in heat flow in the hot condenser.

- Fouling on the evaporation tubes surrounding the main bed causes dra- matic changes in the main variables in the boiler.

90 91

perature is approximately 400–500 °C. A possibility is that the deposit rate on superheater 2 is affected by thermophoresis but for reheater 1 this effect 10. Future work is less important.

- Fouling on the evaporation tubes surrounding the main bed causes dra- matic changes in the main variables in the boiler.

90 91

Baxter L. L., Miles T. R., Miles T. R. Jr., Jenkins B. M., Oden L. L., Dayton D. C, 11. References Milne T. A., 1996, Behavior of Inorganic Material in Biomass-Fired Power Boilers: Field and Laboratory experiences. In: Proceedings of the Engineering Foundation Conference “Biomass Usage for Utility and Industrial Power”, April 28th – May 3rd, Snowbird, Utah. Baxter L.L., 2004, Biomass cofiring overview, Second world conference on biomass for energy industry and world climate protection, Rome, Italy, May 10– 14. Baxter L.L. et al.; 1996, Behavior of Inorganic Material in Biomass-Fired Power Boilers: Field and Laboratory Experiences, Proceedings of the United Engi- Aho M., Ferrer E, 2005, Importance of coal ash composition in protecting the boiler neering Foundation Conference on Biomass usage for utility and industrial against chlorine deposition during combustion of chlorine-rich biomass, Fuel power 84, 201–212. , Snowbird, Utah, 28 april– 3 maj. Baxter L.L., 1993, Ash deposition during biomass and coal combustion, a mecha- Aho M., Silvennoinen J, 2004, Preventing chlorine deposition on the heat transfer nistic approach. Biomass Bioenergy 4(2):85–102. surface with aluminium-silicon rich biomass residue and additive, Fuel 83, 1299–1305. Benson S. A., Steadman E. N., Zygerlicke C. J., Erickson T. A., 1996, In application of advanced technology to ash related problems in boilers, Plenum press, Aho M., Silvennoinen J., 2005, Preventing chlorine deposits on heat transfer sur- New York, pp 1–15. faces with aluminium –silicon rich biomass residue and additive, Fuel 1299– 1305. Berg M. et al., 2003, Combustion of waste wood, Report nr 820, Värmeforsk Ser- vice AB . Åmand L. E., Leckner B., Eskilsson D., Tullin C., 2006, Deposits on heat transfer Binderup Hansen F., 1999, Deposit formation in the convective path of a Danish 80 tubes during co-combustion of biofuels band sewage sludges, Fuel 85 1313– 1322. MWth CFD-boiler co-firing straw and coal for power generation, Impact of mineral Impurities in solid fuel combustion, edited by Gupta et al, Kluwer Aca- Andersson A. et al., 2003, Combustion of Waste Wood Second phase of the collab- demic/Plenum Publisher, New York. oration project on wastewood combustion, Report nr 820, Värmeforsk Service AB. Björkenfjäll L., Eliasson K., 2003, Mätningar och datorbaserade strömningsberäk- ningar för bestämning av avskiljningsverkningsgrad I rektangulär cyklon, Andersson M., 2003, Deposit trends in biomass fired boilers, report M4-204, Mälardalens högskola, Diploma work, nr. 17. Värmeforsk Service AB. Andersson C., Högberg, J., 2001, Fouling and slagging problems at recovered wood Brus E., Öhman M., Nordin, A., Broström, D., Hedman H., Eklund, A., 2004. Bed Agglomeration characteristics of biomass fuels using blast-furnace slag as bed fuel combustion. F9-821. Värmeforsk Service AB. material. Energy & Fuels 18, 1187–1193. Andersson, C., Ljung, P., Woxlin, H. (1997). On-line Alkali Monitoring-Part 1. U3- 606. Stiftelsen för värmeteknisk forskning. Bryers R. W., 1996, Fireside slagging, fouling and high-temperature corrosion of heat-transfer surface due to impurities in steam-raising fuels. Prog. Energy Åström K.J., Bell R., 1988, Simple drum-boiler models. In IFAC international Combustion Sci. 22:29–120. symposium on power systems, modelling and control applications., Brussel, Belgium. Call C. J., Kennedy I.M., 1992, Measurements and simulations of particle dispersion in a turbulent flow, Int. J. Multiphase Flow, Vol 18, No. 6, pp. 891–903. Åström K.J., Bell R., 2000, Drum boiler dynamics. Automatica; 36, pp. 363–378. Chen Y., Xiaolong G., Dynamic modeling and simulation of a 410 t/h Pyroflow Balzer G., Boelle A. et Simonin O., 1995, Eulerian Gas Solid Flow Modelling of Computers & Chemical Engineering Dense Fluidized Bed, HE-44/95/026/A, Departement Laboratoire National CFB boiler, , 2006, Volume 31, Issue 1, Pages 21–31. d'Hydraulique, Chatou Cedex, France. Coda B., 2003, Studies on ash behavior during co-combustion of paper sludge in Bartusch C. 2002, Bed Behaviour and Solids Circulation Algorithms for Boiler 5 at Stuttgart University MälarEnergi AB. Diploma work, Mälardalens Högskola, IST,Västerås, Serial fluidized bed boiler, doctorial thesis, . Corella J., Toledo J., 2000, Incineration of doped sludges in fluidized bed Fate and Number 2002-093. partitioning of six target heavy metals, Journal of Hazardous Materials B80 Basu P., Kefa C, Jestim L, 1999, Boilers and Burners Design and Theory, Mechani- cal Engineering Series, Springer-Verlag, ISBN 0-387-98703-7. 81–105. Dahlquist E., Widarsson B., Lilja R., Avelin A., 2008, Data-Reconciliation-Quality Baxter L. L, 1989,Turbulent transport of particles, Brigham Young University. assurance of signals. Report nr 1050, Värmeforsk Service AB . Baxter L. L, DeSollar R. W, 1993, A mechanistic description of ash deposition during pulerized coal combustion: predictions compared with observations, Fuel Davidsson K.O., Steenari B.-M., Eskilsson, D., 2007a, Kaolin addition during biomass combustion in a 35 MW circulating fluidized-bed boiler. Energy & Fuels Volume 72 Number 10. 21, 1959–1966. Baxter L. L, Smith P. J, 1993, Turbulent dispersion of particles, the STP model, Energy & Fuels, 7, 852–859.

92 93

Baxter L. L., Miles T. R., Miles T. R. Jr., Jenkins B. M., Oden L. L., Dayton D. C, 11. References Milne T. A., 1996, Behavior of Inorganic Material in Biomass-Fired Power Boilers: Field and Laboratory experiences. In: Proceedings of the Engineering Foundation Conference “Biomass Usage for Utility and Industrial Power”, April 28th – May 3rd, Snowbird, Utah. Baxter L.L., 2004, Biomass cofiring overview, Second world conference on biomass for energy industry and world climate protection, Rome, Italy, May 10– 14. Baxter L.L. et al.; 1996, Behavior of Inorganic Material in Biomass-Fired Power Boilers: Field and Laboratory Experiences, Proceedings of the United Engi- Aho M., Ferrer E, 2005, Importance of coal ash composition in protecting the boiler neering Foundation Conference on Biomass usage for utility and industrial against chlorine deposition during combustion of chlorine-rich biomass, Fuel power 84, 201–212. , Snowbird, Utah, 28 april– 3 maj. Baxter L.L., 1993, Ash deposition during biomass and coal combustion, a mecha- Aho M., Silvennoinen J, 2004, Preventing chlorine deposition on the heat transfer nistic approach. Biomass Bioenergy 4(2):85–102. surface with aluminium-silicon rich biomass residue and additive, Fuel 83, 1299–1305. Benson S. A., Steadman E. N., Zygerlicke C. J., Erickson T. A., 1996, In application of advanced technology to ash related problems in boilers, Plenum press, Aho M., Silvennoinen J., 2005, Preventing chlorine deposits on heat transfer sur- New York, pp 1–15. faces with aluminium –silicon rich biomass residue and additive, Fuel 1299– 1305. Berg M. et al., 2003, Combustion of waste wood, Report nr 820, Värmeforsk Ser- vice AB . Åmand L. E., Leckner B., Eskilsson D., Tullin C., 2006, Deposits on heat transfer Binderup Hansen F., 1999, Deposit formation in the convective path of a Danish 80 tubes during co-combustion of biofuels band sewage sludges, Fuel 85 1313– 1322. MWth CFD-boiler co-firing straw and coal for power generation, Impact of mineral Impurities in solid fuel combustion, edited by Gupta et al, Kluwer Aca- Andersson A. et al., 2003, Combustion of Waste Wood Second phase of the collab- demic/Plenum Publisher, New York. oration project on wastewood combustion, Report nr 820, Värmeforsk Service AB. Björkenfjäll L., Eliasson K., 2003, Mätningar och datorbaserade strömningsberäk- ningar för bestämning av avskiljningsverkningsgrad I rektangulär cyklon, Andersson M., 2003, Deposit trends in biomass fired boilers, report M4-204, Mälardalens högskola, Värmeforsk Service AB. Diploma work, nr. 17. Andersson C., Högberg, J., 2001, Fouling and slagging problems at recovered wood Brus E., Öhman M., Nordin, A., Broström, D., Hedman H., Eklund, A., 2004. Bed Agglomeration characteristics of biomass fuels using blast-furnace slag as bed fuel combustion. F9-821. Värmeforsk Service AB. material. Energy & Fuels 18, 1187–1193. Andersson, C., Ljung, P., Woxlin, H. (1997). On-line Alkali Monitoring-Part 1. U3- 606. Stiftelsen för värmeteknisk forskning. Bryers R. W., 1996, Fireside slagging, fouling and high-temperature corrosion of heat-transfer surface due to impurities in steam-raising fuels. Prog. Energy Åström K.J., Bell R., 1988, Simple drum-boiler models. In IFAC international Combustion Sci. 22:29–120. symposium on power systems, modelling and control applications., Brussel, Belgium. Call C. J., Kennedy I.M., 1992, Measurements and simulations of particle disper- Int. J. Multiphase Flow Åström K.J., Bell R., 2000, Drum boiler dynamics. Automatica; 36, pp. 363–378. sion in a turbulent flow, , Vol 18, No. 6, pp. 891–903. Balzer G., Boelle A. et Simonin O., 1995, Eulerian Gas Solid Flow Modelling of Chen Y., Xiaolong G., Dynamic modeling and simulation of a 410 t/h Pyroflow Dense Fluidized Bed, HE-44/95/026/A, Departement Laboratoire National CFB boiler, Computers & Chemical Engineering, 2006, Volume 31, Issue 1, d'Hydraulique, Chatou Cedex, France. Pages 21–31. Bartusch C. 2002, Bed Behaviour and Solids Circulation Algorithms for Boiler 5 at Coda B., 2003, Studies on ash behavior during co-combustion of paper sludge in MälarEnergi AB. Diploma work, Mälardalens Högskola, IST,Västerås, Serial fluidized bed boiler, doctorial thesis, Stuttgart University. Number 2002-093. Corella J., Toledo J., 2000, Incineration of doped sludges in fluidized bed Fate and Journal of Hazardous Materials Basu P., Kefa C, Jestim L, 1999, Boilers and Burners Design and Theory, Mechani- partitioning of six target heavy metals, B80 cal Engineering Series, Springer-Verlag, ISBN 0-387-98703-7. 81–105. Baxter L. L, 1989,Turbulent transport of particles, Brigham Young University. Dahlquist E., Widarsson B., Lilja R., Avelin A., 2008, Data-Reconciliation-Quality Värmeforsk Service AB Baxter L. L, DeSollar R. W, 1993, A mechanistic description of ash deposition dur- assurance of signals. Report nr 1050, . ing pulerized coal combustion: predictions compared with observations, Fuel Davidsson K.O., Steenari B.-M., Eskilsson, D., 2007a, Kaolin addition during bio- Energy & Fuels Volume 72 Number 10. mass combustion in a 35 MW circulating fluidized-bed boiler. Baxter L. L, Smith P. J, 1993, Turbulent dispersion of particles, the STP model, 21, 1959–1966. Energy & Fuels, 7, 852–859.

92 93

Davidsson K.O., Åmand L.-E., Elled A.-L., Leckner, B., 2007b, Effect of cofiring Hutchinson, P., G.F. Hewitt, A.E. Dukler, 1971, Deposition of Liquid or Solid Dis- coal and biofuel with sewage sludge on alkali problems in a circulating fluid- persions from Turbulent Gas Streams: a Stochastic Model, Chemical ized bed boiler, Energy & Fuels 21, 3180–3188. Engineering Science 26,419. Davidsson K.O., Åmand L.-E., Steenari B.-M., Elled A.-L., Eskilsson D.,. Leckner Incropera F. P, DeWitt D. P, 2002, Fundamentals of Heat and Mass Transfer, fifth B, 2008, Countermeasusures against alkali-related problems during combustion edition, John Wiley and Sons, ISBN 0-471-30460-3. of biomass in a circulated fluidized bed boiler, Chemical Engineering Science Irvine T.F., Liley P.E., 1984, Steam and Gas Tables with Computer Equations. Ac- 63 5314–5329. ademic Press. Department of Energy (DOE) USA, 1998, Materials and Components News Letter. Kalisz S., Pronobis M, 2005, Investigation on fouling rate in convective bundles of 134:1. coal-fired boilers in relation to optimization of sootblower operation, Fuel 84, Dong C., Yang Y., Yang R., Zhang J., 2009, Numerical modelling of the gasifica- 927–937. tion based biomass co-firing in a 600 MW, pulverized coal boiler. Applied Kallio G.A., M. W. Reeks, 1989, A numerical simulation of particle deposition in Energy. turbulent boundary layers, Int J Multiphase Flow Vol 15 No.3 pp 433–446. Elled A.-L., Åmand L.-E., Leckner B., Andersson B.-Å., 2007, The fate of elements Kassman H., Berg M., 2006, The ChlorOut Concept – Measurement and reduction in fluidised bed combustion of sewage sludge and wood, Fuel 86 843–852. of corrosive alkali chloride, Issue 03,ThermalNet CombNet. Elled A.-L., Åmand L-E., Eskilsson D., 2008, Fate of zinc during combustion of Kataja T., Majanne Y., 2007, Dynamic Model of a Bubbling Fluidized Bed Boiler. demolition wood in a fluidised bed boiler, Energy & Fuels 22 1519–1526. The 48th Scandinavian Conference on Simulation and Modeling (SIMS 2007), Ericsson L., 2004, Parameterstudie av rökgasstråk, Hur påverkas effekten av olika 30–31 October, Göteborg (Särö), Sweden. överhettare i Västerås biobränslepanna till följd av beläggningar, Mälardalens Knudsen J. N., Jensen P. A., 2004, Dam-Johansen K., Transformation and Release högskola, Diploma work. to the Gas Phase of Cl, K, and S during Combustion of Annual Biomass, Ener- Fluent 5.4, 2000, User manual, Fluent Inc. gy & Fuels 18, 1385–1399. Galgano A., Salatino P., Crescitelli S., Scala F., Maffettone P.L., 2005, A model of Krause H.H., 1986, High Temperature Corrosion problems in Waste Incineration the dynamics of a fluidized bed combustor burning biomass. Combustion and Systems, Journal of Materials for Energy Systems 7. flame Vol.140, No.4, pp. 271–284. Kuprianov V. I., Kaewklum R., Sirisomboon K., Arromdee P., Chakritthakul S., Gestlöf B., 2004, Beläggningsmätningar med mätsond i biobränslepanna, Mälarda- 2009, Combustion and emission characteristics of a swirling fluidized-bed lens högskola, Diploma work, In Swedish. combustor burning moisturized rice husk, Applied Energy 86. Gosman A.D, Ioannides E, 1981, Aspects of computer simulation of liquid-fueled Lawrence B., Annamalai K., Sweeten J.M., Heflin K., 2009, Cofiring coal and dairy combustors. AIAA Paper No. 81-0323. biomass in a 29 kWt furnace, Applied Energy 86 2359–2372. Grabke H. J., Spiegel M., Reese E., 1995, The Effects of Chlorides, Hydrogen Li Amy, Goodarz Ahmadi, 1992, Dispersion and Deposition of Spherical Particles Chloride, and Sulphur Dioxide in the Oxidation of Steels below Deposits, Cor- from Sources in a Turbulent Channel Flow, Aerosol Science and Technology rosion Science 37, 1023–1043. 16:209–226. Greenfield C., Quarini G., 1998, A Langrangian simulation of particle deposition in Lin, W., Dam-Johansen, K., Frandsen, F., 2003, Agglomeration in bio-fuel fired a turbulent boundary layer in the presence of thermophoresis, Applied Mathe- fluidized bed combustors. Chemical Engineering Journal. 96. matical Modelling 22 759–771. Lisa K., Lu Y., Salmenoja K., 1999, Sulfation of potassium chloride at combustion Gyllenhammar M., 2007, Measures for simultaneous minimization of alkali related conditions. Energy & Fuels 13, 1184–1190. operation problems phase 2, Report nr 1037, Värmeforsk Service AB. Litchford Ron J., San-Moe Jeng, 1991, Efficient statistical transport model for tur- Hansen L. A., Michelsen H. R., Dam-Johansen K., 1995,Alkali Metals in a Coal- bulent particle dispersion in sprays, AIAA Journal, Vol. 29, No. 9. and BiomassFired CFBC – Measurements and Thermodynamic Modeling. In: Maffezzoni C., 1992, Issues in modeling and simulation of power plants. In Pro- Proceedings of the l3th international Conference on Fluidized Bed Combus- ceedings of IFAC symposium on control of power plants and power systems. tion, Vol 1, ASME, pp. 39–48. Vol. 1, pp.19–27. Henderson P. et al., 2006, High temperature corrosion in biomass and waste fired Maxey Martin R., James J. Riley, 1983, Equation of motion for a small rigid sphere boilers, Report nr 992, Värmeforsk Service AB. in a nonuniform flow, Phys. Fluids 26(4). Hendersson P., Björk A., Ljung P., Nyström O., Skog E., 2000, Superheater corro- Miettinen Westberg H, Byström M, Leckner B, 2003, Distribution of potassium, sion in combustion of biofuels, report nr. 700, Värmeforsk Service AB. chlorine and sulphur between solid and vapour phases during combustion of Hendersson P., Szakalos P., Pettersson R., Andersson C., Högberg J., 2006, Reduc- wood chips and coal, Energy&Fuels, 17, 8–28. ing superheater corrosion in wood fired boilers, Materials and Corrosion 57, Miles T., 1996, Alkali deposits found in biomass boilers, Sandia National Laborato- No.2. ry, NREL/TP-433-8142, SAND96-8225 Vol. 2. Mörtstedt S.-E., Hellsten G., 1987, Data och Diagram, Norstedts tryckeri.

94 95

Nielsen H.P., Baxter L.L., Sclippab G., Morey C., Frandsen F.J., Dam-Johansen K., Salmenoja K., 2000, Field and Laboratory Studies on Chlorine-induced Superheat- 2000, Deposition of potassium salts on heat transfer surfaces in straw-fired er, Corrosion in Boilers Fired with Biofuels, PhD dissertation, Åbo Akademi, boiler, a pilot-scale study. Fuel 79:131–9. Åbo. Nielsen H.P., Frandsen F.J., Dam-Johansen K., Baxter L.L., 2000, The implication Sand U., Sandberg J., Larfeldt J., Fdhila R.B., 2008, Numerical prediction of the of chlorine-associated corrosion on the operation of biomass-fired boilers, Pro- transport and pyrolysis in the interior and surrounding of dry and wet wood log. gress in Energy and Combustion Science 26, 283–298. Applied Energy 85, p. 1208–1224. Nordin A., Öhman M, 1996, Agglomeration and defluidization in FBC of biomass Sandberg J., Sand U, Fdhila B. F., 2002, Numerical simulation of fouling on super- fuels-Machnisms and measures for prevention, Application of advanced tech- heater tube walls, 10th Workshop on two-phase flow predictions, Meersburg. nology to ash-related problems in boilers, Plenium Press, New York, Sandberg J., Sand U., Fdhila B. R., 2006, Measurements, theories and simulations Nordin, A., Leven, P., 1997, Askrelaterade driftsproblem i biobränsleeldade an- of particle deposits on super-heater tubes in a CFB biomass boiler, Internation- läggningar. O3-515. Stiftelsen för Värmeteknisk forskning. al Journal of Green Energy 3:43–61. Oleschko H., Schimrosczyk A., Lippert H., Muller M., 2007, Influence of coal Sandberg J., Sand U., Fdhila R.B., 2006, Long time investigation of the effect of composition on the release of Na-K-Cl-and S-species during combustion of fouling on the super-heaters in a circulating fluidised biomass boiler. Interna- brown coal, Fuel 86, 2275–2282. tional Journal of Energy Research 30:1037–1053, DOI: 10.1002/er. 1202. Olsson, J. G., Jäglid, U., Pettersson, J. B. C., Hald, P., 1997, Alkali metal emission Sandeep Jain, 1998, Three-Dimensional simulation of turbulent particle dispersion during pyrolysis of biomass. Energy and Fuels. applications, Department of Chemical and Fuels Engineering, The University Ots A., 2001, Mechanism of ash deposit formation, corrosion and sulphur capture of Utah. by burning calcium and chlorine containing fuels, VGB Power Tech Sandia National Laboratory, 1996. Alkali deposits found in biomass boilers. 81(10):114-20. NREL/TP-433-8142. SAND96-8225. Vol. 2. Park W.C., J.F.Klausner, R. Mei, 1995, Unsteady forces on spherical bubbles, Ex- Scala F., Salatino P., 2002, Modelling fluidized bed combustion of high-volatile periments in Fluids 19 167–172, Springer-Verlag. solid fuels. Chemical engineering science Vol. 57, pp. 1175–1196. Perry’s Chemical Engineer’s Handbook. Segerdahl K., 2003, The breakdown of the protective oxide on 11% chromium Pronobis M., 1994, Influence of ash deposits on the heat transfer in convection steel, the influence of water vapour and gaseous KCl, PhD dissertation boiler surfaces, Coal, Energy and Environment, Mead, John S. Ed. ,Chalmers tekniska högskola, Department of Environmental Inorganic Chemis- Pronobis M., 2006, The influence of biomass co-combustion on boiler fouling and try. efficiency, Fuel 85, 474–480. Senior C. L, 1997, Predicting Removal of coal ash deposits in convective heat ex- Rahimi M., Owen I., Mistry J., 2003, Thermal stresses in boiler tubes arising from changers, Energy and Fuels, 11, 416–420. high-speed cleaning jets, International Journal of Mechanical sciences 5, 995– Senior, C. L., Srinivasachar, S., 1995, Viscosity of ash particles in combustion sys- 1009. tems for prediction of particle sticking. Energy and Fuels. 9. Ranz, W.E. ,W.R. Marshall Jr, 1952, Evaporation from Drops, ,Chem. Eng. Prog., Shirey G., Akers D., Zitron Z., 2005, Production of new biomass/waste containing 48(3):141–146 and 48(4):173–180. solid fuels, CQ Inc. Final report project No. 01E0351. Rivkin S.L., 1988, Thermodynamic Properties of Gases. Fourth Edition Revised. Shuen J.S., A.S.P. Solomon, Q-F. Zhang, , G.M. Faeth. 1984. Structure of particle- Hemisphere Publishing Corporation. laden jets: measurements and predictions, A.I A.A.J.23, 396–404. Robinson A, Buckley S, Yang N, Baxter L, 2001, Experimental measurements of Shuen J.S., L-D. Chen, G.M. Faeth, 1983, Evaluation of stochastic model of parti- the thermal conductivity of ash deposits: Part 2. Effects of sintering and deposit cle dispersion in a turbulent round jet. A.I Ch.E.J.29, 167–170. microstructure, Energy&Fuels 15, 75–84. Shuen J.S., L-D. Chen, G.M. Faeth. 1983a. Prediction of the structure of turbulent, Robinson A.L, Junker H, Baxter L.L, 2002, Pilot scale investigation of the influ- particle-laden, round jet. A.I A.A.J.21, 1480–1483. ence of coal-biomasss cofiring on ash deposition, Energy Fuels 16:343-55. Shuen M.J., Schuler C.A., Humphrey J.A.C., 1989, Numerical calculation of parti- Romero C., Sarunac N, Levy E, 2001, Sootblowing optimization in coal-fired boil- cle-laden gas flows past tubes, AIChE Journal, Vol.35, No.3 ers, The 26th international technical conference on coal utilization and fuel Sjöblom R., 2001, Hypotheses and mechanisms for development of deposits con- systems, Clearwater, Florida, March 5–8. taining zinc and lead in conjugation with combustion of wood waste, Report nr Rönnquist E.-M., 2000, Superheater corrosion in biomassboiler theories and tests in 734, Värmeforsk Service AB. Västermalmsverket Falun, Report nr 870, Värmeforsk Service AB. Skrifvars B.-J., Backman R., Hupa M., Salmenoja K., Vakkilainen E., 2008, Corro- Saffman P.G. ,1965, The lift on a small sphere in a slow shear flow. J. Fluids Mech. sion of superheater steel materials under alkali salt deposits, Corrosion Science 22, 385-440. 50, 1274–1282. Saffman P.G. ,1968, Corrigendum to “The lift on a small sphere in a slow shear Speciation of Zinc in ash investigated by X-ray absorption spectroscopy, Report nr flow”. J. Fluids Mech. 31, 624. 1063, Värmeforsk Service AB.

96 97

Steenari B.-M., Norén K., 2009, Application of synchrotron based X-ray spectros- Wester L., 2008, Charts and graphs for energy calculations, Mälardalens högskola,. copy for the speciation of metals in ash from combustion of biomass and waste, Wiinikka H., Grönberg C., Öhrman O., 2008, Titanium dioxide an effective addi- Värmeforskdagarna 2009, Stockholm, 27–28 January. tive for minimisation of alkali vaporisation, Report nr 1071, Värmeforsk Steenari B.-M., 2000, Askor-bildning och egenskaper, Ångpanneföreningens semi- Service AB. narium i Västerås 2000-03-21. Wolfshtein M., 1969, The velocity and temperature distribution in one-dimensional Sun Z.-A., Jin B.-S., Zhang M.-Y., Liu R.-P., Zhang Y., 2008, Experimental study flow with turbulence augmentation and pressure gradient. Int. J. Heat Mass on cotton stalk combustion in a circulating fluidized bed, Applied Energy 85 Transfer. Vol 12 pp301–308. 1027–1040. Wooldridge Margaret S., 1998, Gas-phase combustion synthesis of particles, Swedish Standards Institute (SIS), 2004, SS-EN 14127:2004 Ultrasonic thickness Prog. Energy Combust. Sci. Vol. 24. pp. 63–87. measurement of metallic and non-metallic materials. Yuu S. N., Yasukouchi Y., Hirosawa T., Jotaki T., 1978, Particle Turbulent Diffu- Talbot L., R.K. Cheng, R.W. Schefer, D.R. Willis, 1980, Thermophoresis of parti- sion in a Dust Laden Round Jet. AIChE Journal 24, 509. cles in a heated boundary layer. J. Fluid Mech., 101(4):737–758. Zhang W., Leckner B, 2001, Process simulation of circulating fluidized beds with Theis M., Skrifvars B.-J., Zevenhoven M., Hupa M., Trän H., 2006, Fouling ten- combustion/gasification of biomass, The European Commission, Joule III, Con- dency of ash resulting from burning mixtures of biofuels. Part 2: deposit tract JOR3CT980306. chemistry. Fuel 85, 1992–2001. Öhman M. et al., 2006, Reduced ash related operational problems (slagging, bed Theis M., Skrifvars B.-J., Hupa M., Tran H., 2006a, Fouling tendency of ash result- agglomeration, corrosion and fouling) by co-combustion biomass with peat, ing from burning mixtures of biofuels. Part 1:Deposition rates, Fuel 85, 1125– Report nr 999, Värmeforsk Service AB. 1130. Öhman M., Nordin A., Skrifvars B.-J. Backman R. Hupa, M., 2000, Bed agglomer- Theis M., Skrifvars B.-J., Zevenhoven M., Hupa M., Tran H., 2006b, Fouling ten- ation characteristics during fluidized bed combustion of biomass fuels Energy dency of ash resulting from burning mixtures of biofuels. Part 2:Deposition & Fuels 14, 169–178. chemistry, Fuel 85, 1992–2001. Theis M., Skrifvars B.-J., Zevenhoven M., Hupa M., Tran H., 2006c, Fouling tendency of ash resulting from burning mixtures of biofuels. Part 3: Influence of probe surface temperature, Fuel 85, 2002–2011. Theis M., Mueller C. Skrifvars B.-J., Hupa M., Tran H., 2006d, Deposit behavior of model biofuel ash in mixtures with quarts sand. Part 1: Experimental data, Fuel 85, 1970–1978. Toledo J., Corella J., Corella L., 2006, The partitioning of heavy metals in incineration of sludges and waste in a bubbling fluidized bed, Journal of Hazardous Materials B126 158-168. Traupel W., 1960, Thermische Turbomaschinen, Zweiter Band. Springer-Verlag. Urbain G., Bottinga, Y., Richet, P, 1982, Viscosity of liquid silica, silicates and alumino-silicates. Geochim. Cosmochim. Acta. 46:1061–1072. Urbain G., Cambier, F., Deletter, M., Anseau, M. R.. 1981, Trans. Br. Ceram. Soc. 80. Valmari T., 2000, Potassium behaviour during combustion of wood in circulating, fluidised bed power plants, VTT Chemical Technology, Dissertation for the degree of Doctor of Technology, VTT publications 414, Espoo Finland, June. Vamari T., Lind T.M., Kauppinen E.I., 1999, Field study on ash behaviour during circulated fluidized-bed combustion of biomass 2, ash deposition and alkali vapour condensation, Energy Fuel 13:390-5. Velero A., Cortes C., 1996, Ash fouling in coal-fired utility boilers, monitoring and optimization of on-load cleaning, Prog. Energy Combust. Sci. Vol 22, pp. 189. Wang Q., K D Squires, M Chen, J B McLaughlin, 1997, On the role of the lift force in turbulence simulation of particle deposition. Int J Multiphase Flow Vol 23 No.4 pp 749–763. Wester L., 2003, Tabeller och diagram för energitekniska beräkningar, Mälardalens högskola. Västerås, Sweden.

98 99

Appendix 1: Basic equations for numerical simulations

During the numerical flow field simulations including the particle path simulations described in this thesis, the classical Eulerian-Lagrangian approach has been used, in Fluent 5.4. The gas-phase is modelled via averaged field equations, applying a model for the gas-phase turbulence, and the particles motion is modelled by a force balance that interacts with the gas-phase.

Gas-phase mass balance

∂ρ ∂ + ()ρui = 0 (A1.1) ∂ ∂xt i

Gas-phase momentum balance

∂ ∂ ∂p ∂   ∂u ∂u 2 ∂u  ρu + ρ uu −= + µ i + j − δ l  + ()i ()ji  ij  ∂t ∂x j ∂ ∂xx ji  ∂x j ∂xi 3 ∂xl     (A1.2) ∂ ()− u i ′′ ρρ gu ij ++ F i ∂x j

3 Fi is the interaction term with the dispersed phase (N/m ). In the simulations the particle interaction on the fluid is included however being rather small since the particles are dilute. If the Boussinesq hypothesis is used in the k-ε model the Reynolds stresses is given by:

101

Appendix 1: Basic equations for numerical simulations

Gas-phase mass balance

∂ρ ∂ + ()ρui = 0 (A1.1) ∂ ∂xt i

Gas-phase momentum balance

101

Gas-phase energy  ∂u ∂u  2  ∂u  − ′uu ′ = µρ  i + j  −  ρ k + i δµ (A1.3) ji tj    t  ij  ∂x j ∂xi  3  ∂xi  ∂ ∂ ∂  c µtp  ∂T  ρe + ρeu p)( =+ k +  + u τ )(  (A1.7) () ()i   jj e f fi  ∂t ∂xi ∂xi  Prt  ∂xi  Gas-phase turbulent viscosity:

k 2 p lC ρ ϑµ == ρ Cl (A1.4) = he =− vTc (A1.8) t µ ε ρ

Gas-phase transport equation for k = ch pT (A1.9)

  ∂()ρk ∂(ρkui ) ∂  µt  ∂k + =  µ +   −  ∂u ∂u  2 ∂u     i j  l ∂t ∂ i ∂xx i σ  ∂xik ( effij = () µτ + µ t ) + µ +− t )( δµ ij (A1.10)   (A1.5)    ∂x j ∂xi  3 ∂xl ∂u j µt ∂T ρuiu′′ j + βgi − ρε ∂xi Pr ∂xit The two-layer zonal model In order to resolve the near-wall region, all the way to the viscous sub layer, Gas-phase transport equation for εεε a very fine grid and a two-layer model are used. In the model the domain is subdivided into two regions, a viscosity-effected region and a fully turbulent region. A turbulent Reynolds number Rey is defined as: ∂(ρε ) ∂(ρεu ) ∂  µ  ∂ε  i  t   + = µ +   + ∂t ∂ i ∂xx i  σ ε  ∂xi  (A1.6) ρ ky 2 Re y = (A1.11) ε  ∂u j µ ∂T  ε µ − ρ ′′ + β t  − ρ C1ε  iuu j 3ε gC i  C2ε k  ∂xi Pr ∂xit  k where y is the distance from the wall to the cell centre. The fluid thermo-physical data In order to accurately model the gas-phase viscosity, the gas-phase energy The gas phase turbulence is modelled using the classical k-ε model, where k equation was included in the simulation model. This is important since the is the turbulent kinetic energy and ε is the dissipation rate of kinetic energy. viscosity, density and heat transfer coefficient used in the simulation are If Rey > 200 the region is fully turbulent and the k-ε model is used. If temperature dependent, and since the tube wall temperature is 250 °C lower Rey < 200 the one equation of Wolfshtein (1969) is used. than the gas-phase inlet temperature. The standard k equation is also used but the turbulent viscosity is computed from:

102 103

Gas-phase transport equation for k = ch pT (A1.9)

102 103

Where: t = µ ρµ kC l µ (A1.12)

τ u = w , E=9.793, κ=0.419 (A1.18) ε is not estimated from the standard transport equation . It is instead ob- τ ρ tained from

+ 3 For y < 11.2 k 2 ε = l ε (A1.13) u ρ yu = τ (A1.19) uτ µ

The length scale lε and lµ are given by: Grid considerations using the standard logarithmic law or the two- layer zonal model  Re y  ε cl l y  −= e x p (1 − ) (A1.14) Using the standard logarithmic boundary law the first cell is located at  Aε  y+ = 30 (or 30–60).Using the two-layer zonal model the first cell is located at y+ = 1 (or 1–4). The latter model is recommended in fluid dynamics problem including severe pressure gradients leading to boundary layer

 Re y  separation (Fluent 5.4) µ cl l y  −= e x p (1 − ) (A1.15)  Aµ 

Where:

3 − 4 cl = Cµ κ , Aµ=70, Aε=2cl (A1.16)

Optional In one additional test case the standard logarithmic wall law was used ap- + plying a coarser grid. A factor, y = ρuτy/µ, is calculated to define where the surface sub layer region is laminar or turbulent (Fluent 5.4).

For y+ > 11.2:

u 1  ρ τ yu  = ln E  (A1.17) uτ κ  µ 

104 105

Where: t = µ ρµ kC l µ (A1.12)

τ u = w , E=9.793, κ=0.419 (A1.18) ε is not estimated from the standard transport equation . It is instead ob- τ ρ tained from

+ 3 For y < 11.2 k 2 ε = l ε (A1.13) u ρ yu = τ (A1.19) uτ µ

 Re y  separation (Fluent 5.4) µ cl l y  −= e x p (1 − ) (A1.15)  Aµ 

Where:

3 − 4 cl = Cµ κ , Aµ=70, Aε=2cl (A1.16)

For y+ > 11.2:

u 1  ρ τ yu  = ln E  (A1.17) uτ κ  µ 

104 105

Appendix 2: Turbulence validation

During the numerical flow field simulations including the particle path simulations described in this thesis a dilute turbulence approach is applied. Results from the simulations are shown in figure 29, showing isolines for velocities (left) and the turbulent kinetic energy k (right). The flue gas velocity approaches zero at the stagnation point at the first frontal tube, accelerates passing the cylinder sides and “hit” the second tube at both sides. Between the tubes the flue gas velocity is very low. The turbulent energy k is very high close to the tubes, between the tubs and especially near the flow separation points.

Figure 29. Isolines for the velocity field (left) and the turbulent kinetic energy field (right).

A flow simulation, with an inlet velocity of 8 m/s and with 10% inlet turbulence intensity, has been compared with a standard turbulent layer presented

107

Appendix 2: Turbulence validation

Figure 29. Isolines for the velocity field (left) and the turbulent kinetic energy field (right).

A flow simulation, with an inlet velocity of 8 m/s and with 10% inlet turbulence intensity, has been compared with a standard turbulent layer presented

107

by Kallio and Reeks (1989). The comparison concerns the turbulent large k + = Ct u 2 /ν (A2.4) eddies time scales and turbulent velocity fluctuations. Kallio and Reeks t ee ε τ (1989) conducted curve fitting of experimental data, showing that for + + 0 < y < 200, where = y uy τ /ν , the normal turbulent fluctuation + '' and = uu yy / u τ is described by:

3k .0 00 5. y +2 u '+ = / u (A2.5) u '+ = (A2.1) 2 τ y + 002923.01 y + 128.2 τ In order to estimate the friction velocity, uτ = shear stresses τ is cal- And for 5 < y+ < 200 the turbulent large eddy time scale ρ culated. As shown in figure 31,τ changes over the tube angle (-180 to +180 + 2 degree), where the stagnation point is at the zero angel degree. At approxi- = tt u /ν (A2.2) ee τ mately 90 degrees the shear stress is approximately 0.6 N/m2 which also is a representative mean value. A flue gas density of 0.48 kg/m3 results in is described by: uτ = 118.1 m/s. The kinematic viscosity () used in the calculations is 7.54 -5 2 + + +2 10 m /s and the constant C is 0.3. t = y −+ 00129.0573.0122.7 y te e (A2.3)

+ + Furthermore, for y < 5 a constant value is suggested: te = 10 , since near wall region is affected by bursting events. The results from the present CFD-simulations are then compared with these equations, considering the local point on the left side of the frontal cylinder as shown in figure 30.

Figure 30. The y+ coordinate on the frontal tube (at the 90 degree point). Figure 31. Variation of the shear stress around the frontal tube with an- Here the data for k and ε normal to the frontal tube (y+ direction in fig 30) gel. are used to estimate the values for:

108 109

Appendix 3: Over all results from the fouling distribution simulations

The aim of this simulation is to determine the fouling distribution on different tubes and at different tube points. To get a statistically significant result of the number of particles that will impinge on the tubes, the number of particles released from the inlet boundary at ten symmetrically distributed points has been set to 100 000, 200 000 and 400 000. The resulting fouling Figure 32. Variation of non dimensional fluctuation velocity with wall distribution curve around the tubes is smoother using 400 000 released par- coordinates. ticles than with 100 000 released particles. The ratio of the total number of particles impinging on a tube, to the total number of released particles, is approximately the same for 100 000, 200 000 and 400 000 particles. For example, with particles of 1 µm the number impinging the two tubes, if 100 000 particles is released four times, is: 2255, 2255, 2211, 2259. This gives a mean of 2245 , a standard deviation of 22.7 and a mean standard deviation (σ/√n) of 11.3 which is 0.5 % of the mean value. Therefore, in most cases, it is sufficient to use 100 000 released particles to get a significant result but for all simulations with low percent values (normally the 0.1 and 1 µm particles simulations) 400 000 released particles are used.

Distribution of particles in size 50 µµµm For a 50 µm particle, and also for 100 µm, with an inlet velocity of 8 m/s, the fouling distribution will be as shown in figure 34 (the figure is presented by a program developed by the author in Matlab code). Most of the particles will impinge on the frontal tube, at the leading edge. Only a small number Figure 33. Variation of non-dimensional turbulent eddy time scale with of particles will impinge on the second tube and only at two symmetrical wall coordinates. regions at the left and the right side of this tube. The distance between the tube outer radius and the fouling line, the growth, is proportional to the number of particles impinging on the tube, One conclusion from the above figures is that both the turbulent eddy time this for a constant number of released particles and with a 10-degree angle scale and the fluctuation velocity, behave principally as the measured of resolution on the tube perimeter. In order to get a good visibility and a boundary layer (Kallio and Reeks, 1989), considering the chosen point of clear resolution in the figure, the number of particles is multiplied by a scale comparison.

110 111

Appendix 3: Over all results from the fouling distribution simulations

110 111

factor. Observe that in order to clearly see the fouling line on the second tube, the scale factor is twenty times the scale factor of the first tube.

0.3 scale=1

0.3 scale=4

0.25 20 my

0.25 50 my

0.2 scale=0.5

0.2 scale=0.2

0.15

0.15 0 0.05 0.1 0.15

0 0.05 0.1 0.15 Figure 35. Variation of deposition for 20 µm particles with position on the frontal and second tube. Figure 34. Variation of deposition for 50 µm particles with position on the frontal and second tube. Distribution of particles in size 10 µm For the case with particle size of 10 µm, the fouling distribution will be dif- Distribution of particles in size 20 µm ferent from the case with the 50 and the 20 µm distribution. The total In the case of a 20 µm particle the fouling distribution will not differ dra- amount of particles impinging on the first tubes is in this case lower but matically from the 50 µm distribution. The total number of particles still, most of the fouling is received by the leading edge. For the second impinging the first tube is less for the 20 µm case than for 50 µm case, but tube, a higher and more complicated particle distribution is obtained. Higher most of the fouling is taking place at the leading edge of the first tube, just fouling values are obtained at low velocity regions at the back of the tubes. The scale in figure 36 is 2. like the case with 50 µm. It should be observed that the growth scale in figure 35 is 1 and 0.5.

112 113

factor. Observe that in order to clearly see the fouling line on the second tube, the scale factor is twenty times the scale factor of the first tube.

0.3 scale=1

0.3 scale=4

0.25 20 my

0.25 50 my

0.2 scale=0.5

0.2 scale=0.2

0.15

0.15 0 0.05 0.1 0.15

112 113

0.3 scale=2 0.3 scale=2

0.25 10 my 0.25 1 my

0.2 scale=2 0.2 scale=2

0.15 0.15

0 0.05 0.1 0.15 0 0.05 0.1 0.15

Figure 36. Variation of deposition for 10 µm particles with Figure 37. Variation of deposition for 1 µm particles with position on the frontal and second tube. position on the frontal and second tube.

Distribution of particles in size 0.1 and 1 µm In the case of particles of a size of 0.1 or 1 µm, the fouling distribution will be quite different from the cases with 50 and 20 µm particle distribution. The total amount of particles impinging on the two tubes is much lower, and the fouling distribution is more even and smooth around both tubes. The high turbulence just after the separation points will give “peak-values”. There is no significantly difference between 0.1 and 1 µm. The growth scale in figure 37 and 38 is 2.

114 115

0.3 scale=2 0.3 scale=2

0.25 10 my 0.25 1 my

0.2 scale=2 0.2 scale=2

0.15 0.15

0 0.05 0.1 0.15 0 0.05 0.1 0.15

114 115

In the case of particles with the size of 100 µm with a inlet velocity of 8 m/s, almost all particles will be trapped on the frontal tube, as a result of the higher inertia of the larger particles, but for particles with the size of 0.1–1 µm only 2–4 % will deposit. The inertia of the smallest particles is not large 0.3 scale=2 enough, to overcome the turbulent gas-phase flow field forces. This results in less particle impingements on the superheater tube walls.

0.25 0.1 my

0.2 scale=2

0.15

0 0.05 0.1 0.15

Figure 39. Variation of deposition with particle diameter (8 m/s and 2 Figure 38. Variation of deposition for 0.1 µm particles with position on m/s). the frontal and second tube. Sensitivity study: How a number of factors influence on the result Total fouling values At this investigation the particle size was kept at 0.1, 1, 10 and 50 µm and The deposition or fouling fraction, Ff (%), is defined as the total number of the inlet boundary velocity was constantly at 8 m/s. The thermophoretic particles impinging on a tube wall, divided by the number of particles re- force, the two-layer boundary equations, 10% inlet turbulence level and the leased at the velocity inlet boundary, and released from the projected tube time eddy constant Cte = 0.3 is used if nothing else is said. area on the inlet boundary, i.e. at a 51 mm width. The investigation tries to answer the question; What is the influence on The result of the simulations shows that there is no large variation in the the result: number of trapped particles in the particle diameter range of 0.1 to 10 µm. - without thermophoretic forces When the particle diameter is increased to 20 µm the effect of the particle - with higher inlet turbulence level inertial force is starting to show, the frontal tube is then increasingly at- - with lower turbulent eddy time scale tacked by the particles and the secondary tube impingements are decreased. - with a coarser grid and a logarithmic wall law instead of the two-layer This pattern of behaviour is amplified as the particle diameter size is in- boundary equations. creased to 50 and 100 µm. Regardless of the inlet velocity, this general pattern is the same, however the higher the inlet velocity, the greater the dif- Influence of the thermophoretic force on particle tube impingements ference between the number of impingements on the frontal and the The thermophoretic force is not affecting the results for particles in size 50 secondary tube µm but for the result with particles in size 0.1–10 µm.

116 117

0.25 0.1 my

0.2 scale=2

0.15

0 0.05 0.1 0.15

116 117

Figure 41. Variation of deposition with particle diameter, (with 10 % and 20 % inlet turbulence). Figure 40. Variation of deposition with particle diameter. (with and without thermophoretic forces). Influence of turbulence timescales on particle tube impingements A parameter study of the impact of the gas-phase eddy timescales, te, on In the case of particle diameter 10 µm and smaller sizes, the impinging val- particle tube impingements, used in the simulations, has been done. The im- ues without the thermophoretic force is lower especially at points with low pingement of particles on the superheater tubes could possibly largely be velocities e.g. at the stagnation points. governed by the time scale used in the simulation, i.e. how long the lifetime For the 1 µm particle the total fouling value (Ff) decrease from 3.2 to 2.2 of an eddy is. % for the frontal tube and decrease from 3.5 to 1.9 % for the second tube without the thermophoretic force. k The result that the thermophoretic force can increase the fouling values = Ct (A3.1) has also been investigated by others e.g. Greenfield and Quarini (1998). t ee ε

Influence of turbulence intensity at the inlet boundary on particle tube impingements In this simulation Cte = 0.30 as standard but as a sensibility analyse the value If the inlet turbulence level is raised from 10 to 20 %, only small changes Cte = 0.15 is investigated. will occur. Higher inlet turbulence level will disperse the particle but since the shear stress is small at the inlet, the turbulence level will decrease on the way to the first tube. The turbulence level close to the tubes is thus more an effect of the flow around the tubes than of the inlet boundary conditions. The relation between inlet boundary turbulence intensity and the number of trapped particles is positively proportional for small particles. E.g. for the 1

µm particle the total fouling value (Ff) increase from 3.2 to 4.2 % for the frontal tube and increase from 3.5 to 3.8 % for the second tube when the turbulence level is raised from 10 to 20 %.

118 119

µm particle the total fouling value (Ff) increase from 3.2 to 4.2 % for the frontal tube and increase from 3.5 to 3.8 % for the second tube when the turbulence level is raised from 10 to 20 %.

118 119

Using the standard logarithmic boundary model the impingement is increased for the smallest particle sizes, the fouling distribution is different and the results are not dependent on if or not the thermophoretic forces are included.

For the 1 µm particle the total fouling value (Ff) increase from 3.2 to 6.6 % for the frontal tube and increase from 3.5 to 5.9 % for the second tube when the logarithmic boundary model is used instead of the two-layer boundary model.

Figure 42. Variation of deposition with particle diameter. (with eddy time constant 0.3 and 0.15).

The simulations showed that the longer lifetime allowed, for the eddies, the more particle impingements on the superheater tubes. This is, of course, a reasonable result, as the eddies will be able to grow bigger and effect the particle trajectories more as the eddy lifetime increases, hence bigger particle velocity fluctuations and hence greater chance of hitting the tube walls. The result from the study gives a general decrease of 0.1–10 µm particle impingements, as the eddy lifetime is decreased. For the 50 µm size parti- Figure 43. Variation of deposition with particle diameter, cles, the inertia force of the particles is much greater than the turbulent (with two different boundary layer laws). forces of the gas-phase and thus the influence of the eddy lifetime is ne- glectable in this case. Results from using the logarithmic boundary layer model are presented in For the 1 µm particle the total fouling value (Ff) decrease from 3.2 to figure 44. The fouling distribution is quite different from simulations with 1.9 % for the frontal tube and decrease from 3.5 to 2.3 % for the second the two-boundary layer model (figure 37). The higher fouling values at the tube when the time eddy constant is decreased from 0.3 to 0.15. stagnation point of the frontal tube is an effect of higher turbulent kinetic energy. Influence of different grid and boundary layer equations on particle tube impingements Simulations with the standard logarithmic boundary layer are performed and the results are compared with the two-layer zonal boundary model (explained in Appendix 1). With the logarithmic boundary model the grid are more course and the y+-values (to the first cell) are around 30 which is up to 25 times larger than the values for the two-layer boundary model (y+ around 1). This affect the impingement values for small particles but not for the largest size (50 µm).

120 121

Figure 42. Variation of deposition with particle diameter. (with eddy time constant 0.3 and 0.15).

120 121

0.3 scale=2

0.25 0.1 my

0.2 scale=2

0.15

0 0.05 0.1 0.15 Figure 45. Isolines for the velocity field (left) and the turbulent kinetic energy field (right), (with the logarithmic boundary layer equation). Figure 44. Variation of deposition for 1 µm particles with position, on the frontal and second tube, using the logarithmic boundary layer equation. Conclusions Using CFD calculations including a model for particle tracking, it is possi- Isolines for the velocity distribution (to the left in figure 45) are about the ble to develop methods to predict possible fouling distribution on tubes. same as for the two boundary layer simulation (figure 29) but not the turbu- The results are sensitive to different parameters as turbulence levels and lence kinetic energy k distribution (to the right in figure 45). This solution turbulent scales for particles of small sizes (less than 10 µm). Comparing to seems not to be reliable as acceleration of the flow, at the windward side of results from others e.g. Greensfield and Quarini (1998), (who did similar the frontal tube, are expected to decrease the turbulence kinetic energy k. calculations on pipe flow and compared the simulations with experiments) shows that the method in principal is reliable but the accuracy can be improved by “calibration” through experiments. A possible calibration factor is the turbulent eddy time constant (Cte).

122 123

0.3 scale=2

0.25 0.1 my

0.2 scale=2

0.15

122 123

Appendix 4: Summary of deposit component analysis 2002–2003

Samples of deposits and fly ash have been taken during 2002–2003 from the biomass fired, Boiler 5, at Mälarenergi, Västerås and have been analysed with the two methods; XRF and SEM-EDX. The fly ashes are sampled in February 2002 (analysed at Foster Wheeler, Finland). This sample was taken from the tube filter, after the convection part of the flue gas channel. Three deposit samples were taken from the first row of tubes on superheater 2 in August 2002 (analysed at Umeå University, Sweden) where the deposit layer thickness was approximately 10-15 mm thick at the windward side of the tube. Two of these samples were taken from the middle (in radial direction) of the deposit. The third sample was taken from the surface of the deposit material. A further sample was also collected using a deposit probe, positioned close to superheater 2, in May 2003 (analysed at Analytica, Luleå, Sweden). The deposit probe measurement was conducted over 262 hours at a probe temperature of 570 ºC.

125

Appendix 4: Summary of deposit component analysis 2002–2003

125

The results above have then been recalculated from weight % to molar %. Results Table 4. Resulting component analyse molar %. Table 3. Resulting component analyse wt %.

Componets SH2 /Mid1 SH2/Mid2 SH2/Surface Flyash Probe/570 ºC Componets SH2 /Mid1 SH2/Mid2 SH2/Surface Flyash Probe/570 ºC C 20,52 21,93 19,77 4,24 C 4,18 4,44 3,78 0,80 Na2O 4,96 5,00 5,09 0,83 2,55 Na2O 5,22 5,23 5,03 0,81 1,85 MgO 2,05 2,96 3,66 3,92 4,18 MgO 1,40 2,01 2,35 2,49 1,97 Al2O3 1,80 2,44 1,39 2,47 1,22 Al2O3 3,11 4,21 2,26 3,97 1,46 SiO2 20,42 15,42 5,89 24,21 7,75 SiO2 20,83 15,64 5,65 22,89 5,45 P2O5 0,39 0,46 1,26 0,82 0,59 P2O5 0,95 1,11 2,86 1,83 0,99 SO3 15,81 18,83 30,17 14,26 20,47 SO3 21,49 25,44 38,51 17,97 19,19 Cl 0,13 1,04 0,44 0,90 Cl 0,08 0,62 0,25 0,50 K2O 10,50 13,90 13,83 2,36 7,71 K2O 16,79 22,09 20,77 3,49 8,50 CaO 21,03 16,30 16,43 42,17 27,56 CaO 20,02 15,43 14,69 37,21 18,10 MnO 0,36 0,56 0,99 0,69 0,53 MnO 0,43 0,67 1,11 0,77 0,44 Fe2O3 2,03 1,15 1,08 2,68 16,81 Fe2O3 5,50 3,11 2,74 6,72 31,40 TiO2 0,13 0,42 TiO2 0,17 0,40 BaO 0,05 0,06 BaO 0,11 0,10 Sum 100,00 100,00 100,00 99,73 89,84 Sum 100,00 100,00 100,00 99,73 89,84

The high level of Fe 2O3 from the probe sample shows that the thin oxide- shell nearest the probe surface was included since initially a clean probe is exposed to oxidation before the deposit layer has been built up.

126 127

The results above have then been recalculated from weight % to molar %. Results Table 4. Resulting component analyse molar %. Table 3. Resulting component analyse wt %.

126 127

Step1 Determine mole fraction of all components, as oxides, based Appendix 5: Summary of deposit on chemical composition. In this calculation all oxides including Fe are cal- viscosity calculation culated as Fe2O3 but in the equations, for simplicity, are treated as FeO.

Step2 Calculate:

M = CaO+MgO+Na2O+K2O+FeO+2TiO2+3SO3 (A5.2)

Step3 Calculate: The model in this work was proposed by Urbain et al. (1981) and describes the fact that the viscosity of silicate melt and aluminosilicate melt is temperature dependent. It is also dependent on chemical modifiers and can be Alfa = M/(M+Al2O3+Fe2O3). (A5.3) described by.

As mentioned in Step 1, Fe2O3 is zero in this equation and is instead includ- Viscosity (η) = A*T*exp(1000*B/T) (A5.1) ed as FeO in M.

Step4 Calculate: Where T is the temperature in Kelvin, η the dynamic viscosity in Poise (10 Poise = 1 Pas), A and B are parameters which depend on composition. The 2 3 model has been shown to describe the ash viscosity at high temperatures, B = B0+B1*SiO2+B2*( SiO2) + B3*( SiO2) (A5.4) i.e. for viscosity less than 102–103 Pas (Urbain et al 1981, and 1982). The theoretical basis of the Urban model was the treatment of the slags 2 as complex silicates melts. Using the network former model of glasses, the B0 = 13,8+39,9355*alfa-44,049*(alfa) (A5.5) various oxide components within the melts are classified as network former, modifiers or amphoterics. Network formers are components that form the 2 long stable polymers responsible for high viscosity. The modifiers tend to B1 = 30,481-117,1505*alfa+129,9978*(alfa) (A5.6) reduce the stability of the network bye bridging oxygen bonds within the network former thus reducing the viscosity. Amphoteric oxides may have B2 = -40,9429+234,0486*alfa-300,04*(alfa)2 (A5.7) both network former and/or modifier characteristics, depending on the composition of the melt. For ash systems the classifications of the components are: 2 B3 = 60,7619-153,9276*alfa+211,1616*(alfa) (A5.8)

Network formers: SiO2, TiO2 Network modifiers: CaO, MgO, Na2O, K2O, FeO, P2O5, SO3 Step5 Calculate: Amphoterics: Al2O3, Fe2O3

The Urbain method follows 6 main steps: ln(A) = -0,2693*B+11,6725 (A5.9)

128 129

Step2 Calculate:

M = CaO+MgO+Na2O+K2O+FeO+2TiO2+3SO3 (A5.2)

As mentioned in Step 1, Fe2O3 is zero in this equation and is instead includ- Viscosity (η) = A*T*exp(1000*B/T) (A5.1) ed as FeO in M.

Network formers: SiO2, TiO2 Network modifiers: CaO, MgO, Na2O, K2O, FeO, P2O5, SO3 Step5 Calculate: Amphoterics: Al2O3, Fe2O3

The Urbain method follows 6 main steps: ln(A) = -0,2693*B+11,6725 (A5.9)

128 129

Step6 Calculate:

ln(viscosity) = ln(A)+ln(T)+(1000*B/T) (A5.10)

Summary Experimental determinations of glass viscosity including alkali metals is reported in many articles (e.g. Nicholls et al. 1940, Poole 1949, Taylor et al. 1966). These measurements have in some cases been done down to 450 ºC and up to very high viscosity e.g. 1012 Pas (nearly plastic creeping behaviour) and normally with a SiO2 value above 25 %. In 1981 El Badry et al. presented theories on how alkali oxides weaken the Si-O bonds and further more measurements down to 400 ºC were also presented. Several authors have used the Urbain model in order to analyse the mechanism for sintering or particle sticking, by studying the viscosity dependence. Figure46. The particle viscosity as a function of temperature, Lin et al. Kalmanovitch et al. (1988) conducted experiments on different kinds of and Urban model. silicate melts and modified the constants in the Urbain model. The authors presented ideas considering agglomeration of two particles covered by a The Urban model has only one slope and is probably not tested for as low liquid phase where the viscosity of the liquid phase is responsible for the temperatures as 500 ºC but gives reasonable results. sintering mechanism and the ash deposition. Senior et al. (1995) continue to develop the Urbain model for predicting particle sticking of coal ash. For Deposit viscosity coal ash particles, the adhesion efficiency (sticking probability) start from In the present work the Urbain model is applied on the measured deposit approximately zero at 800ºC and 105 Pas, rising to 1 for higher tempera- component analysis (Appendix 4) and the results are presented in figure 47. tures, approximately 1000 ºC and 103 Pas. Lin et al. (2003) have conducted a study on agglomeration in bio-fired fluidized bed combustors and especially considering utilization of straw as fuel. Their conclusion was that sand particles in the bed become coated with a layer of SiO2 and K2O. At approximately 700 ºC an initial sintering is started. The initial sintering temperature, Ts, was suggested to be 80–90 % of the corresponding melting temperature for amorphous materials. When the temperature is decreased to the level lower than Ts, the viscosity of ash increased more rapidly (change in slope) and the sintering effect may be less important. In the article, a viscosity for a SiO2 - K 2O system is presented with 35 wt % K2O and 65 % SiO 2 in the range of 1400–450 ºC. 35 wt % K2O correspond to 25,6 molar % K2O. As shown in figure 46 there is a change of the slope of the viscosity, in the log(viscosity)-1/T diagram, at 700 ºC. A comparison with the Urbain model is also conducted and presented in figure 46. The equivalent points on the x-axis of figure 46 are; from the left to right: 800, 700, 600 and 500 ºC.

130 131

Step6 Calculate: ln(viscosity) = ln(A)+ln(T)+(1000*B/T) (A5.10)

130 131

1,E+06 Fly ash, from textile filter Super-heater 2, half depth, nr. 1 1,E+05 Super-heater 2, half depth, nr. 2 Probe, at 570 °C 1,E+04 Super-heater 2, at the surface PAPER 1

1,E+03

Viscosity, [Pa s] 1,E+02

1,E+01

1,E+00 400 450 500 550 600 650 700 750 800 Temperature, [°C]

Figure 47. The particle viscosity as a function of temperature, calculated from the chemical analysis.

Conclusions The calculated deposit material viscosity is rather low, which is mainly an effect of the low amount of SiO2. For the two lines in figure 47 with the lowest viscosity (SH2/surface and Probe/570 ºC) the amount of SiO2 is below 10 %. Even if the mixture of the components are out of range compared to the mixtures used when the calibrated constants where settled in the Ur- bain model the model gives reasonably valid results. The fly ash has a high viscosity and will therefore not easily get stuck on the tubes. If the approximate value of 103 is used for the deposits viscosity, as suggested by Senior et al. (1995), for an adhesion efficiency of 1, it can been estimated that fly ash is not fully sticky until 600 ºC. However, at the surface of superheater 2 the value 1 is reached already at a temperature of approximately 450 ºC. The high level of iron oxide at the probe is an effect of that this analysis include the oxidation shell close to the probe surface.

132 133