Master Thesis CFD Modelling of a Biomass Incinerator for Prediction of Risk Areas for Corrosion Damages Svend Skovgaard Petersen July 2012 CFD Modelling of a Biomass Incinerator for Prediction of Risk Areas for Corrosion Damages 17/7-2012 This Thesis was produced by: Svend Skovgaard Petersen, s072374 DTU-MEK, Force Technology Supervisors Dr. Jens Honore Walther (DTU-MEK) Elisabeth Akoh Hove - Force Technology Jan Hein Jørgensen (DTU-CHEC) Danmarks Tekniske Universitet Preface This thesis was developed as part of a major project at Force Technology concerning damages on biomass incinerators. The project is part of the obligation that Force Technology has as being one of the nine GTS institutes in Denmark, which provide none profit research and development for the Danish industry. This thesis should be seen as a first step in this project for generating a model for predicting corrosion damages on biomass incinerators. The work was done as a cooperation between Force Technology, Danmarks Tekniske Universitet DTU and the Verdo heat and power plant in Randers - DK. The thesis focuses on the model development and the use of this through a CFD analysis, with the reader being Verdo and the Danish industry. Thus, quite a substantial academic work was moved to appendices, including specific CFD models and preliminary numerical analyses conducted on Verdo heat and power plant. When used in the report, short descriptions and references are made to these appendices. I would like to thank the production team at Verdo heat and power plant for providing the necessary informations on their geometry of the boiler, the running conditions and corrosion damages seen at Verdo. Also a big thank you to the department of Industrial Processes at Force Technology for providing the project, computational power and know how within simulations of incinerators. I also wish to thank my two main supervisors Dr. Jens Honore Walther from DTU-MEK and Elisabeth Akoh Hove from Force Technology. From DTUCHEC, Flemming Frandsen and Jan Hein Jørgensen should have a thanks for sharing their knowledge on this complex subject as well. Finally I wish to thank my wife for great moral support during the whole project. Abstract In this thesis a numerical model for predicting areas with high risk of corrosion in biomass fired boilers has been developed. The model was developed through a literature study and tested in a full scale CFD analysis of the second boiler at Verdo heat and power plant. STAR-CCM+ was used for the CFD calculations. The model is based on the metal temperature of heat transfer surfaces in the boiler and the concentrations of potassium chloride(KCl) and oxygen(O2 ). A series of preliminary analyses of the used models were conducted in order to validate the simulation of the combustion processes. The main simulation consist of: a wood chip grate firing simulated with a bed model, a biomass suspension firing simulated with combustion of Lagrangian particles and a fully spacial and physical resolved, integrated steam circuit of super heater 3(SH3). The fully resolved SH3 provided a precise load distribution of the super heater. The average outlet temperature of the steam in SH3 was within 10 % of the temperature measured by Verdo. The developed corrosion model does not predict precise corrosion rates but only high, medium and low levels of corrosion risk. The most severe spots of corrosion seen by Verdo were predicted by the model with good precision similar to the corrosion profile across SH3. A numerical model for coarse ash deposition was also developed, showing good agreement between the heaviest fouling areas in the boiler and the model. Decm was predicted in the bottom and mid section of SH3. position rates of 1 day The secondary air nozzles in the furnace had a poor configuration, as the jets pushed the freeboard combustion zone together instead of mixing it with oxygen rich air as intended. A result of this was an uneven load distribution in the boiler and in particular SH3. Resumé I dette speciale er der blevet udarbejdet en numerisk korrosionsmodel til forudsigelser af risikoområder i biomassefyrede kedler. Modellen blev udarbejdet gennem et litteraturstudie og testet ved en fuldskala CFD analyse af kedel nummer to ved Verdo kraft varmeværk. Til CFD beregningerne blev STAR-CCM+ anvendt. Korrosionsmodellen bygger på metaltemperaturen af hedefladerne i kedlen samt koncentrationer af kaliumklorid(KCl) og ilt(O2 ). En række indledende analyser af de brugte modeller blev udarbejdet for at validere forbrændingen. Hoved simuleringen består af: en bed-model for simulering af ristefyret flis, en simulering af suspensionsfyring af biomasse ved brug af Lagrangian partikler, og en fuld integreret, geometrisk og fysisk opløst simulering af overheder 3(OH3). Denne opløsning gav en meget nøjagtig fordeling af belastningen på OH3. Den gennemsnitlige udløbstemperatur for dampen i OH3 blev beregnet indenfor 10% af den målte temperatur hos Verdo. Den udviklede korrosionsmodel er ikke beregnet til forudsigelser af præcise korrosionsrater, men kun til forudsigelse af områder med høje, medium og lave korrosionsrisiko. Modellen giver et korrosionsprofil i form af risikogradueringer hen over OH3 meget lig det virkelige korrosionsprofil. En numerisk model til forudsigelse af grovkornet askeopbygninger blev sideløbende udarbejdet. Modellen viste gode overensstemmelser mellem de mest udprægede belægningsområder i kedlen og OH3. Belægningsraten blev bestem til cm ca 1 dag . Indstillingerne af de sekundære luftdyser i forbrændingskammeret viste sig at være uhensigtsmæssige. I stedet for a opnå den ønskelige blanding af ilt og brandbare gasser i forbrændingskammeret, skubber de bare de brandbare gasser sammen. Dette resulterer i en ujævn belastning af hele kedlen og især OH3. Contents Preface i Abstract iii Resumé v Nomenclature 1 Introduction xiii 1 1.1 Biomass as a fuel . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Verdo heat and power plant as a case study . . . . . . . . . . . . 2 1.2.1 Damages seen on Verdo . . . . . . . . . . . . . . . . . . . 2 Force Technology and "Damage seen on biomass power plants" . 5 1.3.1 5 1.3 Motivation for the study . . . . . . . . . . . . . . . . . . . 2 Operation conditions at Verdo 7 2.1 Firing methods used at Verdo . . . . . . . . . . . . . . . . . . . . 7 2.2 Fuel analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 3 Literature study 3.1 3.2 11 Critical species and corrosion processes . . . . . . . . . . . . . . . 11 3.1.1 Summary of the corrosion process . . . . . . . . . . . . . 14 Including the critical corrosion species in the bed model . . . . . 14 3.2.1 K release . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 3.2.2 Cl release . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 3.2.3 S release . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 viii CONTENTS 3.2.4 Minimum and maximum release of K . . . . . . . . . . . . 17 Corrosion risk evaluation . . . . . . . . . . . . . . . . . . . . . . . 17 3.3.1 First criteria - Metal temperature . . . . . . . . . . . . . 18 3.3.2 Second criteria - Oxygen . . . . . . . . . . . . . . . . . . . 19 3.3.3 Third criteria - Presence of Cl2 . . . . . . . . . . . . . . . 20 3.3.4 Summary on the model for risk of corrosion . . . . . . . . 20 Deposition of coarse ash particles . . . . . . . . . . . . . . . . . . 20 3.4.1 Deposition mechanism . . . . . . . . . . . . . . . . . . . . 20 3.5 Coarse ash deposition modelling . . . . . . . . . . . . . . . . . . 22 3.6 Simulating the grate firing using a bed model . . . . . . . . . . . 26 3.7 Previous work done on spreader simulation and suspension firing 28 3.3 3.4 4 Governing equations and numerical modelling 4.1 31 The governing equations . . . . . . . . . . . . . . . . . . . . . . . 31 4.1.1 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . 32 4.1.2 Conservation of momentum . . . . . . . . . . . . . . . . . 32 4.1.3 Turbulence modelling . . . . . . . . . . . . . . . . . . . . 33 4.1.4 The Energy equation . . . . . . . . . . . . . . . . . . . . . 33 4.1.5 Equation of state . . . . . . . . . . . . . . . . . . . . . . . 34 4.2 Concepts of CFD . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 4.3 Using STAR-CCM+ for CFD . . . . . . . . . . . . . . . . . . . . 35 4.3.1 36 Deposition of particles in STAR-CCM+ . . . . . . . . . . 5 Results 5.1 5.2 39 Results from preliminary analyses . . . . . . . . . . . . . . . . . . 39 5.1.1 Main and secondary air supply . . . . . . . . . . . . . . . 39 5.1.2 The distribution of wood chips on the grate . . . . . . . . 40 5.1.3 Suspension firing and bed model . . . . . . . . . . . . . . 40 5.1.4 Simulating the steam in SH3 . . . . . . . . . . . . . . . . 41 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 5.2.1 5.2.2 The mesh used for simulation with integrated steam circuit of SH3 . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Physics, models, BC’s and convergence . . . . . . . . . . . 44 CONTENTS ix 5.2.3 Results for the full domain . . . . . . . . . . . . . . . . . 46 5.2.4 Summary of the general freeboard . . . . . . . . . . . . . 55 5.2.5 The region near SH3 . . . . . . . . . . . . . . . . . . . . . 55 5.2.6 Summary of corrosion risk from surface temperature in the SH3 region . . . . . . . . . . . . . . . . . . . . . . . . 60 Coarse ash deposition . . . . . . . . . . . . . . . . . . . . 61 5.2.7 6 Discussion 65 6.1 Issues when simulating spreaders for wood chip . . . . . . . . . . 65 6.2 Problems and important parameters when simulating suspension firing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 6.3 Errors and uncertainties regarding the bed model . . . . . . . . . 67 6.4 Mesh limitations for boiler simulations . . . . . . . . . . . . . . . 68 6.5 Stabilities of combustion simulations with Lagrangian particles and multiple region interactions . . . . . . . . . . . . . . . . . . . 69 6.6 Deposition of particles . . . . . . . . . . . . . . . . . . . . . . . . 70 6.7 Risk assessment for high temperature corrosion in the Verdo boiler 72 6.7.1 Corrosion due to concentrations in the flue gas . . . . . . 72 6.7.2 Corrosion from a deposition point of view - including shedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 The missing SO2 simulation and corresponding sulphation of KCl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 Mitigation of high temperature corrosion . . . . . . . . . . . . . . 74 6.7.3 6.8 7 Conclusion and future work 77 7.1 Wood chip distribution on the bed . . . . . . . . . . . . . . . . . 77 7.2 Suspension firing using Lagrangian particles . . . . . . . . . . . . 77 7.3 Simulating combustion of spreader distributed wood chips on the grate with a bed model . . . . . . . . . . . . . . . . . . . . . . . 78 Full scale simulation with integrated steam region for SH3 . . . . 78 7.4.1 The general freeboard . . . . . . . . . . . . . . . . . . . . 78 7.4.2 The SH3 region with corrosion . . . . . . . . . . . . . . . 79 7.4.3 Deposition of coarse ash particles - fouling . . . . . . . . . 80 7.5 Recommendations for Verdo . . . . . . . . . . . . . . . . . . . . . 80 7.6 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 7.4 x CONTENTS List of Figures 81 List of Tables 89 Bibliography 92 A Models used in STAR-CCM+ 97 A.1 Modelling Lagrangian particles . . . . . . . . . . . . . . . . . . . 97 A.1.1 Momentum balance for particles . . . . . . . . . . . . . . 97 A.1.2 Lagrangian Energy Model . . . . . . . . . . . . . . . . . . 98 A.1.3 Turbulent Dispersion . . . . . . . . . . . . . . . . . . . . . 98 A.1.4 Two way coupling . . . . . . . . . . . . . . . . . . . . . . 98 A.1.5 Coal combustion of Lagrangian particles . . . . . . . . . . 98 A.1.6 Particle Radiation . . . . . . . . . . . . . . . . . . . . . . 104 A.2 Modelling radiation . . . . . . . . . . . . . . . . . . . . . . . . . . 104 A.3 Simulating heat exchangers with porous media regions . . . . . . 105 A.3.1 Energy extracted from the heat exchangers . . . . . . . . 106 A.3.2 Pressure drop over the heat exchangers . . . . . . . . . . 107 B Preliminary analyses 109 B.1 Primary and secondary combustion air . . . . . . . . . . . . . . . 109 B.1.1 Primary air . . . . . . . . . . . . . . . . . . . . . . . . . . 109 B.1.2 Secondary air . . . . . . . . . . . . . . . . . . . . . . . . . 109 B.2 Distibution of wood chips on the grate . . . . . . . . . . . . . . . 110 B.3 Simulating suspension firing . . . . . . . . . . . . . . . . . . . . . 118 B.3.1 Summary on suspension firing . . . . . . . . . . . . . . . . 124 B.4 Tuning in the bed model . . . . . . . . . . . . . . . . . . . . . . . 125 B.5 Simulating the steam in the SH3 tubes . . . . . . . . . . . . . . . 129 B.5.1 Mesh used to simulate the steam in the SH3 tubes . . . . 129 B.5.2 The physics inside the SH tubes . . . . . . . . . . . . . . 130 B.5.3 Results for flow in SH3 tubes separately . . . . . . . . . . 130 B.6 Average outlet temperature, residuals, pressure and heat flux for SH3 in full simulation . . . . . . . . . . . . . . . . . . . . . . . . 133 CONTENTS C Production values at Verdo during full load xi 135 C.1 Air monitor at Verdo . . . . . . . . . . . . . . . . . . . . . . . . . 136 C.2 Steam monitor at Verdo . . . . . . . . . . . . . . . . . . . . . . . 137 C.3 Wood chip monitor at Verdo . . . . . . . . . . . . . . . . . . . . 138 C.4 Biomass for suspension firing monitor at Verdo . . . . . . . . . . 139 C.5 Fuel analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140 C.6 Size distributions for biomass fuel . . . . . . . . . . . . . . . . . . 144 D Developed calculation codes 153 D.1 Matlab code for pipe flow . . . . . . . . . . . . . . . . . . . . . . 154 D.2 Matlab code for depositing particles . . . . . . . . . . . . . . . . 156 D.3 Maple code for calculating volatile mass fraction . . . . . . . . . 158 xii CONTENTS Nomenclature Ap Surface area of particle CD Drag force coefficient of particle Ec Activation energy for char Epn Activation energy for particle in devolatization H Height P Probability of an event Pa Probability of sticking due impact angle Pp Probability of sticking due to stickiness of particle Ps Probability of sticking due to stickiness of wall Pcorr Probability of corrosion Pstick Total probability of a particle sticking XCH4 Unknown mass fraction of CH4 XCO2 Unknown mass fraction of CO2 XCO Unknown mass fraction of CO XH2 Unknown mass fraction of H2 Yi,ult Mass fraction of C, O, H and KCl in the ultimate dry analysis Yi,vol Mass fraction of C, O, H and KCl in the volatiles Λ Ratio between air-excess and fuel Ω Solid angle of a sphere α Angle xiv ū Nomenclature Mean part of velocity in turbulence modelling βn,rest Normal restitution coefficient βt,rest Tangential restitution coefficient δij Kronecker’s delta Ṁ Mass flow Q̇ Effect ṁ Mass flux γcp Mass fraction of coal λ Wave length of radiation Ac Pre-exponential factor for char oxidation Apn Pre-exponential factor for devolatization Cp Heat capacity M Molar mass Nu Nusselt number Re Reynolds number R Universal gas constant T Temperature T15 Temperature for which the melt fraction is 15 % T70 Temperature for which the melt fraction is 70 % Tp Temperature of particle Tsurf Surface temperature of metal Y Mass fraction ep Erosivity of impacting particles hp Heat transfer coefficient for particles m Mass mp Mass of particle nK Molar quantity of K Nomenclature nCl Molar quantity of Cl p Pressure µ Dynamic viscosity ω Under relaxation factor ρ Density of the fluid ρu′i u′j Reynolds stresses σij Second order stress tensor τij′ The viscous stress Fb Body force on particle Fd Drag force on particle Fp Pressure force on particle from pressure gradients Fs Surface force on particle Fu User defined force on particle FV M Virtual mass of particle vp Velocity of particle vs Slip velocity between particle and fluid continuum cpn Reaction rate constant fmelt Melt fraction g Gravity h Enthalpy k Thermal conductivity of the fluid lp Distance from cell center to particle n Number of occurrences p Propensity of sticking ref Effective radius for a face cell rvpn Kinetic rate of volatile matter production rvpn Volatile matter production rate xv xvi Nomenclature t Time u′ Fluctuating part of velocity in turbulence modelling ui Velocity tensor volf rac Volatile mass fraction wt% Weight percent basis xi Spacial first order tensor Af ace Area of cell face Abbreviations CFD Computational Fluid Dynamics CHPP Combined Heat and Power Plant GHG Green House Gasses VM Volatile matter AMG Algebraic Multigrid solver BC Boundary condition BiCGStab Bi Conjugate Gradient Stabilizer CDS Central Difference Scheme DEM Discrete Element Model FC Fixed carbon GCV Gross Calorific Value NCV Net Calorific Value NS eq. Navier-Stokes equations RANS Reynolds Avereged Navier-Stokes equation SD Standard Deviation VM Volatile matter Tde Turbulent dissipation energy Tke Turbulent kinetic energy V Volume Chapter 1 Introduction 1.1 Biomass as a fuel Through the last century the use of fossil fuels such as coal and oil have been a key factor for the development of the modern society. In the use these fuels a large amount of green house gasses(GHG) have been released causing the concentrations of GHG in the atmosphere to rise significantly, [1]. These GHG have always been in the atmosphere, but the increase are now associated with an increase of the global temperature and melting of the ice poles. Here CO2 is the GHG contributing the most with an estimated share of 9 − 26 %, [1]. A great deal have been done to lower these emissions of GHG by increasing the use of natural sustainable resources such as wind, solar and biomass. All though emitting GHG, biomass is considered a green natural resource, as the release of GHG through burning of it, is no more than the plants have taken out of the atmosphere. Thus biomasses have a GHG cycle from one year in annual biomasses such as straw up to a couple of centuries for wood. However the burning of the fossil fuels release GHG stored million of years ago. Biomass in the form of wood have been used as a source for warming and light by man for millenniums. Today the biomass as a fuel includes straw, waste, wood and other waste products from the agricultural industry. In 2006, the global consumption of biomass was estimated to make up for 12 % of the global energy release, [2]. When including the share of water, wind and solar power, it makes up for 80 % of the total use of the sustainable energy. In 2006 in Denmark, the biomass held a share of the sustainable energy of approximately 70 %, where the main constitution of the biomass was straw, wood and biodegradable waste, [2]. The utilization of this biomass in Denmark and the western society is mainly 2 CHAPTER 1. INTRODUCTION done through large incinerators at combined heat and power plants(CHPP). In Denmark, the use of biomass is partly an effect of a government legislation from 1993 stating, that a use of 1.4 million ton of straw and wood a year should be used by the Danish energy sector by 2005. In 2008 this was expanded by the government with 700.000 tons by the year 2011. However, the use of biomass in CHPP’s is not without problems, as the inorganic content in the biomass cause fouling and corrosion damages in the boilers, in a scale far worse than coal fired CHPP’s, [3]. A lot of research have been conducted in hopes of understanding and reducing these problems. This goes from full scale experimental tests to very detailed academic work. E.g. the combustion of annual biomass has been investigated by Knudsen in a Ph.d. project, [4], and the deposition and corrosion in biomass incinerators have been investigated in the Ph.d. project by Nielsen in [5]. A large work was done by Frandsen in the last decade, leading to his doctoral thesis on the subject of ash formation, deposition and corrosion when utilizing straw in CHPP’s, [3]. Analyses of boilers have also been conducted by use of computational fluid dynamic(CFD) as a tool for quantifying the flow and temperature conditions in biomass incinerators, see references [6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]. In this work a numerical simulation of a boiler at a biomass CHPP will be conducted. This includes an effort of bringing some of the large amounts of material on biomass incinerators in the literature together in a numerical corrosion model for CFD codes. The model was developed through a literature study and applied on a full scale biomass fuelled CHPP, the Verdo heat and power plant. 1.2 Verdo heat and power plant as a case study The Verdo heat and power plant in Randers- Denmark changed in 2009 from being a coal fired plant to a pure biomass fired plant. This was done as a part of a strategy of putting Verdo in the lead of green energy production. Today the Verdo plant is one of the largest purely biomass fired plants in Denmark with a maximum production of 48 MW electrical power and 140 MW thermal power for district heating. However the change from a mix of coal and biomass to pure biomass has introduced problems with fouling and corrosion around their super heater 3, SH3, which is the first super heater the flue gas encounters. 1.2.1 Damages seen on Verdo In the fall of 2011 Verdo contacted the Korrosion og Metallurgi department in Force Technology in order to have the super heater investigated and estimate an expected time for replacement. This work showed severe corrosion of the tubes and Force Technology estimated less than 1 year before critical failure, 1.2. VERDO HEAT AND POWER PLANT AS A CASE STUDY 3 [17]. Measurements on the metal thickness of the SH3 tubes had also been conducted by Verdo them self since they changed from coal to biomass. These measurements showed severe accelerated corrosion rates, leading to a necessary replacement of SH3 after only 3 years of production with 100% biomass. In Figure 1.1 the location of the thickness measurements are indicated, and in Table 1.1 are the corresponding measured metal thickness’s of the tubes are listed. Figure 1.1: Sketch over location of known corrosion damages on SH3 at Verdo. The SH3 is located at approximately 14 m directly above the grate. In Table 1.2 the estimated corrosion rate and the minimum thickness before risk of rupture are listed. Table 1.1: The design and measured thickness at all 14 tube rows in the four positions indicated at Figure 1.1, adopted from [17]. Position A B C D Design [mm] 6.3 6.3 6.3 4.5 5.5 5.6 5.4 4.2 5.6 5.6 5.7 4.3 5.4 5.5 5.6 4.3 5.5 5.6 5.3 4.1 5.4 5.7 5.1 3.9 [mm] 5.7 5.6 5.2 3.9 5.5 5.5 5.1 3.7 5.4 5.3 5.1 3.9 5.6 5.6 5.5 3.7 5.5 5.4 5.1 4.2 5.8 6.0 5.4 3.8 6.2 5.7 5.9 3.8 The measurements show severe corrosion at particular the front side of the tubes facing towards the flue gas. This very defined location of corrosion gives a good reference for an eventual corrosion model. 6.0 5.8 5.3 3.7 5.6 5.5 5.6 3.9 4 CHAPTER 1. INTRODUCTION Table 1.2: Corrosion rates for the two tube thickness’s. The measured thickness is the minimum measured thickness by Verdo and Force. The min. thickness is the minimum thickness estimated by Force Technology - Korrosion og Metallurgi before risk of rupture. The Corrosion rate are the estimated future rates estimated by Force Technology - Korrosion og Metallurgi based on corrosion history and temperature, adopted from [17]. Design thickness [mm] Measured thickness [mm] Min. thickness [mm] 6.3 4.5 5.1 3.3 5.3 4 Corrosion rate 0.9 0.6 # mm $ 10.000 The information in Table 1.1 and Table 1.2 are plotted against each other in Figure 1.2. Figure 1.2: Plot of the measured thickness’s at location A, B, C and D. The design and minimum thickness for the two tube thickness are indicated by the four horizontal lines. Location A,B and C have a design and minimum thickness of 6.3 mm and 5 mm respectively. Location D has a design and minimum thickness of 4.5 mm and 4 mm respectively. It is clear that the largest corrosions are found in the middle rows. It is clear that for especially location C the corrosion is most severe in the middle rows. h 1.3. FORCE TECHNOLOGY AND "DAMAGE SEEN ON BIOMASS POWER PLANTS" 1.3 5 Force Technology and "Damage seen on biomass power plants" At the division for "Energi, Klima og Miljø" and department of "Industrial Processes" at Force Technology, a large project named "Damages on biomass fired power plants" is currently in the work. In connection to this project, an arrangement with Verdo was established, giving a relevant up-to-date case study for the project and a cutting edge analysis for Verdo. 1.3.1 Motivation for the study This work is the result of a job as student worker on the project "Damages on biomass fired power plants" at Force Technology and my master thesis at DTUMEK. The aim of this thesis is to provide a numerical model for predicting critical areas with a high risk of corrosion in biomass fired power plants. The study and final model are intended to be the first part of the Damages on biomass fired power plants project at Force Technology. The aim is not to generate an exact corrosion prediction model, but only a general assessment of the corrosion risk and the foundation for a future model for predicting damages in biomass incinerators. The potential for a successful model could be as a design tool in the design phase of incinerators. By this, gaining longer life time of super heaters and higher efficiencies of power plants, and thereby reducing some of the problems associated with biomass as a fuel in power plants. 6 CHAPTER 1. INTRODUCTION Chapter 2 Operation conditions at Verdo In this chapter the conditions at Verdo heat and power plant will be outlined. This includes the fuel and firing methods used. 2.1 Firing methods used at Verdo At Verdo they use a combination of wood chips and annual biomass as fuels. The wood chips are spread out on to the grate by the use of a spreader mechanism, where the chips are burned as grate firing. The annual biomass is burned in suspension above the grate, known as suspension firing. See Figure 2.1 for a schematic illustration of the two firing methods. For the combustion of the wood chips, air is fed in under the grate. This air is referred to as the primary air. On the walls, a series of air nozzles are located which supply additional air for both the combustion used by the suspension firing and combustion of volatile gasses from the grate. This is referred to as over-fire or secondary air. At the top of Figure 2.1 some piping are shown illustrating super heaters. It should be noted that the real configuration of these are quite different, see Figure 2.3. The fuel for suspension firing is injected through three feeding tubes onto spreader stones for distribution of the fuel, see Figure 2.2(a) for picture of the spreader stone. The wood chips fall down through a duct leading to the spreader shown in Figure 2.2(b). The spreading is controlled by the angle of the horizontal plate in the front and by the carrier air jet coming from the small nozzles above the plate. The two different firing methods can be adjusted separately in order to control the combustion. This is done primarily through the mass flow of fuel injected at each location and through the carrier air. The combustion in the furnace is also 8 CHAPTER 2. OPERATION CONDITIONS AT VERDO Figure 2.1: Schematic figure of the running conditions at the Verdo plant. Adapted from reference [15]. (a) Picture of the injection tube for the annual biomass fuel leading to a spreader stone. (b) Picture of the wood chip spreader seen from inside of the furnace. Figure 2.2: Pictures from inside of the furnace in Verdo at shut down. The pictures was taken at the end of this project, as the SH3 was about to be replaced. a) Picture of a spreader stone used for spreading suspension fired fuel. b) Picture of the spreader for wood chips at the top. At the bottom the spreader for the old coal firing is seen. 2.2. FUEL ANALYSES 9 controlled by regulating the primary and secondary air. The values for the fuel consumption and air are found from the information in Appendix C.1 on page 136 to Appendix C.4 on page 139, which corresponds to a full load production at Verdo. Figure 2.3: Geometry of the simulated second boiler at the Verdo heat and power plant. Fuel spreaders are located on the front wall in the lower left corner of the figure. SH3 is the fully resolved tube banks at the upper left part of the boiler. The orange blocks illustrates the SH2, SH1, ECO3, ECO2 and ECO1. 2.2 Fuel analyses A very important part of a combustion simulation is the fuel used for it. The contents of the fuel used at Verdo, consisting of the wood chips and annual biomass, is presented in this section. Verdo provided the fuel analyses which contains both a proximate, ultimate and size distribution analyses. These are found in Appendix C.5 on page 140 and Appendix C.6 on page 144. The most important analyses are presented here. The proximate analysis contains the distribution of water, volatile matter (VM), fixed carbon (FC) and ash on a weight percentage basis, wt%. The ultimate analysis, containing the atom distribu- 10 CHAPTER 2. OPERATION CONDITIONS AT VERDO Table 2.1: Proximate analysis of the fuel used at Verdo. 1) Wood chips. 2) Dark biopellets. 3) Light bio-pellets (oat peel). 4) Light bio-pellets 2. 4) Seed pellets. 5) Wood pellets. 6) Miscellaneous biomass dust. See Appendix C.6 on page 144 for analysis and pictures. 1 2 3 4 5 6 7 Ash, wt% 0.77 7.72 3.68 2.79 7.28 1.06 0.58 3.85 Water, wt% 42.6 17.02 8.42 13.18 14.23 7.90 6.17 11.15 FC, wt% 7 VM, wt% 49 Heating value [GJ/ton] 10.49 2-7 averaged 7 78 15.67 17.27 16.34 15.61 18.45 18.88 17.04 Table 2.2: Ultimate analysis of the fuel used at Verdo on dry basis with the Gross Calorific Value(GCV), and Net Calorific Value(NCV). 1) Wood chips. 2) Dark biopellets. 3) Light bio-pellets (oat peel). 4) Light bio-pellets 2. 4) Seed pellets. 5) Wood pellets. 6) Miscellaneous biomass dust. See Appendix C.6 on page 144 for pictures. 1 2 3 4 5 6 7 (2-7)averaged Ash, wt% 0.8 7.6 3.6 2.7 7.5 1 0.5 3.82 Cl, wt% 0.02 0.08 0.06 0.07 0.12 0.01 0.01 0.06 S, wt% 0.01 0.2 0.08 0.08 0.16 0.01 0.01 0.09 C, wt% 49.6 47.2 46.8 47.3 44.7 50.1 50 47.68 H, wt% 6 6.1 6.1 6.1 6 6 6 6.05 N, wt% 0.17 2.69 0.88 0.66 2.23 0.11 0.07 1.11 O, wt% 43 36 43 43 39 43 43 41.17 GCV 20.11 19.40 19.10 19.21 18.62 20.26 20.30 19.54 NCV 18.8 18.09 17.78 17.89 17.30 18.96 18.98 18.17 tion, is usually conducted on a dry, ash free wt% basis. The main elements in a ultimate analysis is C, O, H, N, S and Cl and sometimes heavier compounds such as K, Ca, Si, Al. The proximate and ultimate analyses are presented in Table 2.1 and Table 2.2. The fuel number 1 is the wood chips and fuel number 2-7 are the fuel used for suspension firing. As no information on the distribution of VM and FC was given, the value of 0.07 wt% for FC in reference [18] was used. This yields a VM value of 49 wt% for the wood chips and 78 wt% for the biomass. In order not to have seven different analyses, the average values for water, carbon, ash ect. for the six different biomass fuels used in the suspension firing are found. The six different biomass fuels are assumed to have an equally distribution of weight according to Verdo. This, combined with that the six fuels are not that different in contents, makes the averaging approximation reasonable. This average value is the last column in Table 2.1 and Table 2.2 named 2-7 averaged. Chapter 3 Literature study In the following chapter a model for high temperature corrosion damages on biomass incinerators on specifically SH tubes, will be developed from the literature. For this, the key chemical reactions and species on the tubes will have to be clarified. Once found, the deposition mechanism of these components can be investigated in accordance with the origin and concentration of the species. Thus one also has to look at the releases of the species from the burning of the biomass. The model therefore needs to address the release of critical corrosion species in order to quantify concentration levels, temperature and flow velocities of these species near the super heater tubes. These are all important parameters to estimate the risk of fouling and corrosion of the tubes, [19]. Thus the model contains three main categories: The release, deposition/fouling and corrosion. At the end of the chapter a short review of previous numerical work done on spreaders and bed models will be given. 3.1 Critical species and corrosion processes The most critical parameter in the corrosion of the super heater tubes in biomass fired boilers is the presence of chlorine, Cl2 , [3, 5, 18, 19]. The existence of Cl2 in the deposits is closely related to alkali metals and in particular potassium, K, as will be elaborated later in section section 3.2 on page 14. Combined with elevated temperatures, high temperature corrosion of the metal tubes can occur. From reference [19] the chemical reactions of the corrosion processes are described. In proximity to the tube, the environment can be oxidising or reducing, depending on the flow, gas mixture and others. An oxidising environment on the metal surface can occur when the flue gas is rich on oxygen, O2 , and 12 CHAPTER 3. LITERATURE STUDY thin layers of deposit and oxides exists. Reducing conditions can occur under thick deposit layers, if the flue gas is reducing or if burning particles stick to the surface and keep burning. In the latter, the particles use the oxygen for the remaining combustion. Normally (if the flue gas is oxygen rich) there will be an oxidising environment at the surface of the deposits and reducing environment at the metal surface, having a crossover in between. This decline in available oxygen towards the metal surface acts as a catalyst to some chemical processes. At the surface and outer layer of the deposit a sulphation of potassium chloride, KCl, can take place, in which hydrogen chloride, HCl, is formed, see reaction (3.1). 1 2KCl(s) + O2 (g) + SO2 (g) + H2 O(g) −→ K2 SO4 (s) + 2HCl(g) 2 1 2HCl(g) + O2 (g) −→ Cl2 (g) + H2 O(g) 2 (3.1) (3.2) The HCl can then form chlorine gas, Cl2 , see reaction (3.2). The chlorine gas will diffuse in all direction, including towards the metal surface all depending of the partial pressure. Here the chlorine can cause selective corrosion with Fe, Cr and Ni compounds in the metal, see reaction (3.3) to (3.6) Fe(s) + Cl2 (g) −→ FeCl2 (g) (3.3) Cr(s) + Cl2 (g) −→ CrCl2 (g) (3.4) 3 Cr(s) + Cl2 (g) −→ CrCl3 (g) 2 (3.5) Ni(s) + Cl2 (g) −→ NiCl2 (g) (3.6) The products in reaction (3.3) to (3.6) are the corrosion products, and will go from solid state to gaseous state where they are stable at the low oxygen partial pressure under the oxide layer, [5]. Further the relative high vapour pressure from these metal chlorides will cause an outwards diffusion. This diffusion will continue out to a point in the deposit with high enough oxygen content, where the metal chlorides will experience oxidation. Here the release of the chlorine gas, see reaction (3.7) to (3.8), closes an effective closed circuit, and the chlorine can diffuse in all directions again, including into the metal surface and repeat the corrosion process, see Figure 3.1. 3FeCl2 (g) + 2O2 (g) −→ Fe3 O4 (s) + 3Cl2 (g) (3.7) 3 2FeCl2 (g) + O2 (g) −→ Fe2 O3 (s) + 2Cl2 (g) 2 (3.8) The reactions in reaction (3.7) and (3.8) can similarly be written for Cr and Ni. 3.1. CRITICAL SPECIES AND CORROSION PROCESSES 13 Figure 3.1: Principle sketch of how high the temperature corrosion processes occur on the metal surface and deposit layer, adopted from [34]. For the reactions above to run, certain conditions need to be satisfied. First the chlorine partial pressure needs to be high enough to initiate the corrosion. Second, the oxygen partial pressure (reducing conditions) should be low, and third the metal chlorides partial pressure is high enough to diffuse outwards again, [19]. In Table 3.1, the minimum partial pressure of Cl2 for creating metal chlorides along with the temperature/partial pressure for four main metal chlorides are stated. In the table the two temperatures T4 and T6 are used. These refer to the temperatures at which the partial pressure of the formed metal chloride is high enough to drive the corrosion. Some uses the T4 as the limit while other more conservative work have used T6 . Table 3.1: Pressure for creation of metal chlorides, and the temperatures where the pressure is 10−4 atm (T4 ) and 10−6 (T6 ), adapted from [19]. Cl2 -partial pressure for T4 -Temperature - T6 -Temperature- generation of metal- partial pressure of metal- partial pressure of metal- chlorides chlorides is 10−4 atm chlorides is 10−6 atm [atm] [◦ C] [◦ C] FeCl2 10−14 536 418 CrCl2 10−18 741 588 CrCl3 10−8 611 489 NiCl2 10−10 607 494 It is clear that FeCl2 is the one most likely to be subjected to corrosion, as it has a very low minimum chloride pressure along with a low minimum temperature of T6 = 418 ◦ C to drive the corrosion. 14 3.1.1 CHAPTER 3. LITERATURE STUDY Summary of the corrosion process The main cause of corrosion is from a sulphation of KCl in the deposits on the tubes, in which chlorine gas is released and causes selective corrosion of the tubes. Reducing conditions are needed on the metal surface, as well as a minimum temperature of 418◦ C at the metal surface to drive the corrosion cycle. 3.2 Including the critical corrosion species in the bed model For the combustion of the fuel on the grate, a bed model is used. This is in short an external model in this case, that simulates the combustion by inputs from the CFD simulation and in return generates a profile of the temperature and release of species. This is explained in more detail in section 3.6 on page 26. However the main governing parameter in the bed model is the temperature. The following is with focus for final utilization in the bed model. The presence of Cl, K and S during combustion of biomass have a large influence on both the deposit build up, as well as the corrosion process, [3]. It is therefore natural to look at the release of these species. As actual measurements in the combustion zone of an incinerator are very difficult, controlled lab-scale test have been made by Johansen et al. in reference [20], and by Bjørkman et al. in reference [21]. These tests investigate the release of the critical species for ash formation at different isothermal conditions during pyrolysis and combustion. Although the tests are conducted on combustion of corn stover, the basic release mechanism and temperature range are presumed to be the same for wood chips and other biomasses for now, [21]. 3.2.1 K release In Figure 3.2 it can be seen that only a small amount of K is released in the low temperature range T < 700 ◦ C for combustion conditions, while this threshold is shifted approximately 100 ◦ C towards higher temperatures for pyrolysis. Beyond this point a high increase in the releases are observed, resulting in a release of approximately 50% of the total amount of K present in the fuel. The retained 50% of K in the ash is related to the amount of chlorine and the presence of silicate and aluminosilicate, which will be explained below. 3.2. INCLUDING THE CRITICAL CORROSION SPECIES IN THE BED MODEL 15 Figure 3.2: Relative release of K during pyrolysis and combustion. A considerable increase in the release is noticed at T > 700 ◦ C. Adopted from [20]. 3.2.2 Cl release In Figure 3.3 the relative release of chlorine is shown as a function of temperature. The graph shows a relative high amount of chlorine, 50 wt%, released at lower temperatures 400 − 700 ◦ C. Bjørkman et al. found similar tendencies in [21]. No clear explanation on how this chlorine is bound in the fuel and no exact release mechanism was found. The chlorine is probably released as HCl retained in the fuel. However, Bjørkman et al. concluded that the chlorine was not bound in the water content of the fuel as almost 100 % of the chlorine remained at 200 ◦ C. Above 700 ◦ C the remaining chlorine is released resulting in a complete dechlorination. A strong correlation between the release of chlorine and the release of potassium is found. This indicates that the release of potassium is limited by the amount of chlorine left after the first evaporation stage of chlorine, leaving chlorine to be the limiting factor on this stage. 16 CHAPTER 3. LITERATURE STUDY Figure 3.3: Relative release of Cl during pyrolysis and combustion, where the experimental running time before sampling are indicated with e.g. 20 min in the figure. Adopted from [20]. 3.2.3 S release Approximately 60 % of the sulphur release takes place in the temperature range below 900 ◦ C. An increase during combustion is seen from 900 ◦ C until complete desulfurization at 1150 ◦ C. During pyrolysis no additional release beyond 60 % is observed. Figure 3.4: Relative release of S during pyrolysis and combustion. Adopted from [20]. 3.3. CORROSION RISK EVALUATION 3.2.4 17 Minimum and maximum release of K A minimum and maximum expression is given in reference [20] to estimate the release of K into the flue gas. These are listed here as eq. (3.9) and (3.10). Kmin = n500 Cl nK (3.9) Kmax = ntot Cl nK (3.10) Here n is the molar quantity of Cl and K respectively and the superscript 500 is the molar quantity of Cl available at 500 ◦ C and tot is total amount of Cl in the fuel. See Figure 3.5 for the use of eq. (3.9) and (3.10). Figure 3.5: Upper and lower estimate of K release. Adopted from [20]. As the release of K is mainly associated with Cl, eq. (3.9) and (3.10) also express the release of KCl. As this project only needs the release amounts at the different temperatures for the bed model no further explanations will be given here. A more detailed discussion and explanation on the release path and mechanisms of K, Cl and S can be found in reference [20]. 3.3 Corrosion risk evaluation As mentioned a full model of the advanced deposition and corrosion mechanism is out of the scope of this work. A frame work for a more simple model for corrosion will therefore be given. Instead of modelling all the details, the main 18 CHAPTER 3. LITERATURE STUDY precursors for corrosions will instead be evaluated at the surface and adjacent fluid cells. Each contributing factor for corrosion is defined as an event. Each event is given a probability value between 0 and 1, depending on risk of that particular event to cause corrosion. E.g. a very low temperature will give a value of 0 and a high temperature gives a value of 1 for the temperature event, a zero concentration of KCl will give a value of 0 for the KCl event and so on. If the probability of corrosion due to each event is individually independent, the multiplication rule for independent events can be used, [24]: P (AB) = P (A)P (B) (3.11) where A and B are events. By multiplying the probability from each event the model probability of corrosion is found Ptotal = P1 · ... · Pi · ... · Pn (3.12) Here Ptotal is the total corrosion risk assessment by the model, Pi is the probability of event i to cause corrosion where i = [1; n] and n is the total number of events taking into account. First of all the different precursors for corrosion must be identified and assigned as an event. When calculating the probability as done in eq. (3.12), each event are able to completely dismiss the possibility of corrosion, as it can take the value 0 and thereby leading to a total probability of 0. By expanding the complexity of the model with more events the likelihood of independent events decrease, and the probability formula needed to describe the risk needs to be changed accordingly. In lack of depth of knowledge regarding the interdependencies only the most important parameters for corrosion are investigated in the following. When defining the important parameters the presence of Cl bound as KCl has already been identified, along with the metal surface temperature. Furthermore the oxygen was found to have significance. 3.3.1 First criteria - Metal temperature As a first criteria for high temperature corrosion the critical metal temperature of the tube surface, Tsurf , have to be high enough. In Table 3.1 the minimum temperature for the corrosion to occur was stated as 418 ◦ C. According to Nielsen et al. in reference [5], no or little corrosion occurs at Tsurf = 480 ◦ C. The probability of corrosion due to surface temperature event, Pcorr (Tsurf ), is set to zero between these two temperatures, 450 ◦ C. At Tsurf = 520 ◦ C, severe corrosions may become a problem according to Nielsen et al., and at 600 ◦ C rapid corrosion rates were found in reference [3]. A conservative upper boundary with Pcorr (Tsurf ) = 1 at 520 ◦ C is used. In between these two temperatures a linear 3.3. CORROSION RISK EVALUATION 19 function, as a first choice is chosen, see eq. (3.13). Tsurf < 450 ◦ C 1 Pcorr (Tsurf ) = 70 · Tsurf − 6.428 f or 450 ◦ C ≦ Tsurf ≦ 520 ◦ C 1 Tsurf > 520 ◦ C (3.13) The probability curve for the temperature event is illustrated in Figure 3.6. 0 Figure 3.6: Probability of corrosion due to metal surface temperature alone. 3.3.2 Second criteria - Oxygen The oxygen concentration was found to be important in the sulphation process of KCl. However at the surface of the metal tubes, on top of the corrosion front, an oxid layer exist. The corrosive species Cl2 (g) and the corrosion product FeCl2 (g) diffuse in and out across this oxide layer, [5]. A dense oxide layer will therefore lower the diffusion across and thereby the corrosion. According to Henriksen et al. in reference [19] and Hansen et al. in reference [25], a low oxygen concentration (reducing conditions) in the flue gas will result in a thinner and more porous oxide layer. This will allow more diffusion of the corrosion species across the oxide layer and allow high corrosion rates. Likewise will an oxidising environment yield a more dense and thick oxide layer, providing better protection. Reducing conditions will also prohibit the sulphation process in the freeboard, leaving KCl to be more stable. This gives KCl the possibility to deposit on ash particles or directly on cold wall and tube surfaces. Reducing conditions would therefore increase the KCl concentration in the deposits. If the conditions shifts or fluctuates between reducing and oxidising, the KCl will undergo sulphation in the deposits which will increase the risk of corrosion, [19]. 20 CHAPTER 3. LITERATURE STUDY A low oxygen concentration therefore yields a higher risk of corrosion, [19]. However there have not been established any specific thresholds. In stead of including oxygen in the corrosion model in eq. (3.12) as an contributing event, it will only be commented on through the overall oxygen profile in the freeboard. 3.3.3 Third criteria - Presence of Cl2 The presence of Cl2 on the surfaces in the boiler is closely related to the presence of KCl and sulphation by SO2 , eq. (3.1). KCl can be deposited in several ways as described in section 3.4.1 on page 20. Like the oxygen it have not been possible to establish certain corrosion thresholds for KCl and SO2 concentrations on the surfaces. The KCl and SO2 as events in eq. (3.12) must therefore also be discarded for now and only evaluated from the general concentration levels. 3.3.4 Summary on the model for risk of corrosion A general model for predicting the corrosion risk on surfaces was develop. Unfortunately it was only possible to establish valid thresholds for the temperature event. Other corrosion parameters that could be used as corrosion events, such as the concentration of O2 , KCl and SO2 have not be quantified in this work. However the model is not discarded as it could be a useful tool for further development. The model is used with the one parameter/event that was quantified, namely the temperature. As the model does not predict the total corrosion it will not be used as a final prediction model, but as an important tool in the work. The expression in eq. (3.13) is used to expressed probability from the metal surface temperature, Pcorr (Tsurf ), and will act as a base for the corrosion model. 3.4 Deposition of coarse ash particles In this section the different mechanisms of deposition is described followed by a development of a coarse ash deposition model. 3.4.1 Deposition mechanism Some studies on how deposition of ash particles, to especially the surface of super heaters occur, have been made, [3, 5, 10]. Most of them describe how deposits build up when firing with straw. Here several mechanisms are contributing, where the main mechanisms are impacts of coarse ash particles, condensation and thermal boundary layer diffusion(thermophoresis). Each mechanism has a different role, where the first two are the ones found most crucial. Coarse 3.4. DEPOSITION OF COARSE ASH PARTICLES 21 ash deposition depends on many parameters such as flue gas speed, direction in relation to the solid surfaces, particle size, temperature and stickiness of both particle and deposit/wall. Especially stickiness is an important factor or the particles will just bounce of and not stick. For the walls and particles to be sticky, initial condensation of alkali salts on the surfaces and temperature of both particles and surface are important. The condensation of the alkali salts is important, as it acts as an initiating film of glue for the impacting particles. The previously mentioned stickiness is a function of what compounds and species are on the surface and the temperature. Kær et al., [9, 10], developed a CFD model for modelling the deposition rates of impacting particles, thermophoresis and vapour deposition, stating good agreement between their model and experimental data. Simulating all of these deposition mechanism is out of the scope for this work. Instead a simulation of coarse ash impacting particles and their deposition contribution will be given, as this is the main deposition factor, [9]. A general assumption when burning wood chips rather than straw is the lesser amount of ash produced from wood chips [19]. This favours the condensation instead of the impactation mechanism regarding deposit build up. However, as the Verdo plant is firing both wood chips and annual biomass such as biopellets and waste products of seeds, the assumption of low ash in the freeboard cannot be made. Furthermore, the annual biomass is fired using suspension firing which further contributes to ash particles in the freeboard. The precursors for the deposition mechanism where ash particles cause large fouling problems therefore exist. As the most relevant species in the flue gas regarding fouling and corrosion problems is alkali salts, and predominantly KCl, the focus will be given to the deposition of this, [3, 5]. Condensation of KCl is primarily dependent on temperature, the partial pressure of KCl in the flue gas and the equilibrium partial pressure at the surface. According to reference [5] KCl will predominantly go directly from gas to solid phase called deposition or desublimation, and only for high concentration levels actually condense into liquid form. This can be shown using thermodynamical distribution calculations that minimizes the total Gibbs free energy of a system, [5]. See Figure 3.7 for a thermodynamical distribution of the species at a straw fired case. Here the ratio between air-excess and fuel, Λ = 1.3, was used. A value of 1 represent a stoichiometric mixture. A value below one represents a oxygen lean mixture and a value above one represent a oxygen rich mixture. 22 CHAPTER 3. LITERATURE STUDY Figure 3.7: Thermodynamical stable species of potassium under oxidizing conditions (Λ = 1.3), straw fired case. Adopted from [5]. From Figure 3.7 the concentration of KCl depends on other species such as sulphur, silicium and oxygen in the flue gas. In theory the distribution of the species therefore varies in the entire freeboard and no exact temperature or concentration level can be given to estimate condensation/deposition. However the temperature range of 650 − 750 ◦ C seems promising. 3.5 Coarse ash deposition modelling A deposition model for coarse ash particles developed by Kær in reference [9] will be explained and adapted in the following. The presence of alkali salt and especially KCl is an important factor for fouling. From reference [5] and Figure 3.7, the deposition temperature, when KCl goes from gas to solid, is found to approximately 750◦ C. However this one temperature maybe to general, as the deposition of KCl happens in a broader temperature range and depend on the the other specie concentrations, c.f. Figure 3.7. In reference [9], Kær derived an expression for the molten fraction, fmelt of flue gas particles, using KCl and silicate particles as the main species, see eq. 3.14. fmelt (T, Y) = fmelt,KCl (T) mKCl msilicate + fmelt,silicate (T) mtotal mtotal (3.14) Where T is the temperature, Y is the mass fraction, m is the mass of KCl and silicate, and fmelt (T) is found from Figure 3.8. 3.5. COARSE ASH DEPOSITION MODELLING 23 Figure 3.8: Approximated melting curves of potassium and silicate-rich particles. Adopted from reference [10]. In reference [6] and [16], the sticky and flow criteria, T15 and T70 was used to estimate whether particles will stick or not. The indices in T15 and T70 denote the temperatures for which the melt fraction of 15% and 70% occur. They argue that only particles with melt fraction above T15 can stick. If the melt fraction on the surface should come above T70 , the slag will melt off again reaching a steady state equilibrium. However, as the corrosion of the metal will proceed regarding a slag-layer thickness equilibrium at T70 , the upper limit of T70 is miss leading if only looking at potential fouling and corrosion problems and therefore not used. These melt criteria can be used on e.g. Lagrangian particles as done by Kær in [9]. Here Kær compare different methods of predicting the stickiness e.g. stickiness due to viscosity of the ash particles and the above mentioned melt criteria. The method that was adopted for the work in [9] was the meltcriteria, and is also chosen as the method for this work. Kær divided the sticking propensity from ash impactation into three categories: 1) Incoming sticky particles collide with a none sticky wall and stick depending on the melt fraction of the particle, first part of eq. (3.15). 2) Incoming none sticky particles that stick due to sticky surface, second part eq. (3.15). 3) non sticky particles that do not stick and erode material from the surface, last part of eq. (3.15). pstick = p(Tp ) + [1 − p(Tp )]ps (Tsurf ) − ep [1 − p(Tp )][1 − ps (Tsurf )] (3.15) Here p(Tp ) is the propensity of sticking for impacting particles at particle temperature. ps (Tsurf ) is the sticking propensity of particles due to stickiness of surface temperature. ep is erosivity of impacting particles on the wall. It is assumed that the contents of the coarse ash and the ash deposits are the same. p(Tp ) and ps (Tsurf ) can therefore be evaluated by the same melting curve. In Figure 3.9 the propensity for sticking, pmelt , due to T15 and the melt fraction 24 CHAPTER 3. LITERATURE STUDY of ash is illustrated by using the expression in eq. (3.14) and a distribution of 0.1 %wt KCl and 0.9 %wt silicate rich ash. Figure 3.9: Melt fraction and propensity of sticking due to T15 criteria. A ratio of 0.1 KCl and 0.9 Silcate particles was used. In Figure 3.9 the ratio of 0.1 wt% KCl and 0.9 wt% silicate particles gives a relative high T15 value of approximately 900 ◦ C. If a distribution of 0.15 and 0.85 between KCl and silica is presence the critical T15 temperature will be reached already at 675◦ C, see Figure 3.10. This lower temperature fits better to the predictions in reference[6] and [16]. Figure 3.10: Melt fraction and propensity of sticking due to T15 criteria. A ratio of 0.15 KCl and 0.85 Si-rich particles was used. As shown, the fraction of KCl in the ash particles has to be found as it influence 3.5. COARSE ASH DEPOSITION MODELLING 25 the model drastically. As discussed later in section 6.6 on page 70 the KCl fraction of the deposits could not be found numerically in the simulation. Instead, the KCl content of the ash located on the SH3 of approximately KCl = 36 %wt found in reference [23], was used. Allthough Kær postulate that the formula in eq. (3.15) provide good results in reference [9], some missing aspects can be pointed out. First off all, the formula will always predict particles sticking if the melt fraction exceed 0.7 from 700◦ C, p(Tp ) = pstick = 1. As the temperature in the first part of a boiler typically lies around 1100 − 1400◦ C, this will probably predict heavy fouling areas in the combustion zone on the convective walls. It is also contradicting to the statement, that both the particle and wall should be sticky in order to have coarse ash deposition, c.f. section 3.4.1 on page 20. Kær mention the effect of cooling of the particles in the cold boundary layer near walls. However this effect is most pronounced for very small particles. Larger particles coming perpendicular to the boundary layer from a hot zone would therefore often stick when using eq. (3.15). If one takes a more mathematical approach to eq. (3.15) and see the propensities as probabilities, as done in eq. (3.12), the rule of independent events can be used again. Assuming that the probability of sticking depends on both the stickiness of the particle and the wall stickiness it impinges, and these two are independent of each other, a different version of eq. (3.15) can be written: Pstick = Pp (Tp ) · Ps (Tsurf ) (3.16) Here neglecting the erosion part. Pstick is the total probability of sticking, Pp is the probability of sticking due to particle stickiness and Ps is probability of sticking due to stickiness of wall. The formula in eq. (3.15) also neglects the impact angle of the particles, αimpact . According to references [9, 19], particles sticking to a surface also depend on the direction of the flue gas, and thereby meant the direction of the impact angle in relation to the surface. Although a vague formulation, it make sense to include some kind of impact angle when predicting the probability of sticking. A simple guess provided here, for a probability of sticking as a function of impact angle, Pa (αimpact ), could be a Gaussian distribution curve as shown in Figure 3.11. 26 CHAPTER 3. LITERATURE STUDY Figure 3.11: Probability of sticking due to impact angle, αimpact [deg] This assumes that a particle is most likely to stick if it hits perpendicular to the surface. This conclusion can be drawn from both reference [9] and [19]. The impact angle can also be seen as independent of the temperature of the particle and wall temperature and thus an even more conservative guess of a sticking probability is found in eq. (3.17). Pstick = Pp (Tp ) · Ps (Tsurf ) · Pa (αimpact ) (3.17) This expression will be tested against the expression provided by Kær. As the simulations will be run as steady state, it only simulates one typical operational condition at a time. The boundary conditions(BC) of the walls are also dependent on how long the plant has been running, regarding deposit build up and the following insulating effect of these. Different boundary cases can be made simulating first clean tubes and then adding an ash layer corresponding to e.g. 24 hours off build up and changing the boundaries accordingly. The new BC will change the solution and deposition rates. This procedure can be done successively for a number of layers to achieve a semi transient solution, as done by Kær in [9] but will not be tried here. 3.6 Simulating the grate firing using a bed model A bed model models the burning of the fuel on the grate in a furnace, and is used as a tool in numerical simulations of furnaces. An extensive study on grate firing technology was done by Yin et al. in reference [18]. In references [7, 8, 10, 11, 13] the principle of bed models can be found. This study will not give any detailed description of the topic of bed modelling as the used bed model was provided by Force Technology. However, the main principles will be outlined in the following, as the modifications done to the provided bed model are explained. The technique used by Molcan et al. in reference [13] is adopted here. 3.6. SIMULATING THE GRATE FIRING USING A BED MODEL 27 Discretisation of the grate The provided bed model descretise the grate into four zones from the front furnace wall to the back, see Figure 3.12. Figure 3.12: Illustration of the grate discretisation. The blue area indicate the area used for the bed model. The discretisation is made from the duct system providing the primary air for the combustion. As seen, each zone goes from one furnace side to the other, with no discretisation in this transverse direction. For each zone, a uniform distribution of the calculated species and temperature is assumed. The four zones are, regarding the simulation, stationary, but as the grate moves in order to discard the ash in reality, the bed model must handle this. Distribution, transportation and burning of wood chips on the bed The used bed model was original developed for straw fired boilers. The feeding of fuel in such furnaces is usually done from the front by e.g. a stoker screw. Here the fuel burns in a certain order as it moves into the furnace. First by drying, then devolatization of volatile gasses and finally char burnout, leaving only ash as the component left on the bed before led into an ash pit. However, at Verdo the wood chips used as fuel on the bed are distributed by a spreader, giving a more even distribution of fresh fuel on the bed, see Figure B.17 in Appendix B.2 on page 110. This is not what the bed model was designed for. The solution to this was to define the zone for where the wood chips land as zone A. The zones, for which the chips would be transported into by the movement of the grate, is named zone B, C an D. For each of the zones A-D a given drying, devolatization and combustion is then specified. Thus wood chips landing on e.g. bed zone 3 will undergo the drying, devolatization and combustion associated to a zone A 28 CHAPTER 3. LITERATURE STUDY process on zone 3. The used % of the combustion process in zone A-D are listed in Table 3.2. Table 3.2: The used fractions for evaporation, devolatization, char burnout and release of KCl at the different zones. Zone A is the landing zone of the wood chips. Zone Evaporation, %H2 O Volatization, %V M C-burnout, %C Ash, %Ash KCl, %KCl A 70 % 75 % 20 % 0% 80 % B 30 % 25 % 55 % 10 % 20 % C 0% 0% 25 % 10 % 0% D 0% 0% 0% 80 % 0% Total 100 % 100 % 100 % 100 % 100 % The percentages stated in Table 3.2 are values based on experiences done in the department of Industrial Processes at Force Technology and the actual temperature profile at Verdo. This is explained further in Appendix B.4 on page 125. The inputs from zone A-D from the different wood chips are then summed up on zone 1-4 in the following way, where the evaporation of water is used as an example: Zone 1 Zone 2 Zone 3 Zone 4 q 1 − Zone2−4 ṁwood,Z4 · %H2 OZC + ṁwood,Z3 · %H2 OZB + ṁwood,Z2 · %H2 OZA + ṁwood,Z2 · %H2 OZA = ṁwood,Z4 · %H2 OZB + ṁwood,Z3 · %H2 OZA ṁwood,Z4 · %H2 OZA Here the notation of e.g. ṁwood,Z4 is the mass flux of wood chips landing on zone 4 and %H2 OZA is the percentage of H2 O released on zone A according to Table 3.2. The mass flux are found in Appendix B.4 on page 125. The values generated from the above are used as input to the bed model along with the irradiation on the zones. The irradiation is the radiation absorbed by the bed. A calculation sheet (bed model) then simulates the combustion processes and generates the input values for the CFD code. As the CFD code and the simulation are coupled through the radiation and release of species an iterative process is needed to find a steady state for the bed model. 3.7 Previous work done on spreader simulation and suspension firing Lagrangian particles have been used to simulate biomass suspension firing, see reference [7, 14]. In reference [14], Belosevic conducts a thorough literature re- 3.7. PREVIOUS WORK DONE ON SPREADER SIMULATION AND SUSPENSION FIRING 29 search on different CFD models and is a good overview on the subject. One of the conclusions is that a general acceptance of coal models used on biomass is found in the literature. Here Lagrangian particles often represents the biomass or fuel. Some of the main concerns about using a coal combustion model on biomass are the way volatiles are packed in the particles, the amount of moisture, the difference in surface morphology between coal and biomass particles, partitioning of particles, and the release rates of volatiles. However, as the focus for this work is the corrosion part and not detailed aspects of suspension firing, a coal combustion model is used for this. STAR-CCM+ provides a coal combustion model for Lagrangian particles and is readily used. This model also allows inorganic species such as KCl to be released with the combustible VM. A stoichiometric analysis of the fuel and combustion reactions must be done. This is done in Appendix A.1.5 on page 100. 30 CHAPTER 3. LITERATURE STUDY Chapter 4 Governing equations and numerical modelling This chapter will give a brief introduction fluid dynamics and to the field of Computational Fluid Dynamics(CFD). CFD is a numerical approach for describing and solving a fluid dynamical problem in a given domain. In general the fluid domain is descretized into small volumes of computational cells, where the governing equations are solved. The main subjects are therefore the discretization of the domain into what is called a mesh and the governing equations, as will be elaborated in the following. 4.1 The governing equations In fluid dynamics the three laws of conservation for a physical system applies, [26]: • Conservation of mass (Continuity) • Conservation of momentum (Newtons’s second law) • Conservation of energy (first law of thermodynamics) Different simplifications can then be applied to each, such as incompressible flow which leads to constant density and so on. The conservation of energy can be decoupled from the other two, and left out for cases where the energy part is neglectible. For fluids containing chemical reactions at least two extra laws must be satisfied as well, [26]: CHAPTER 4. GOVERNING EQUATIONS AND NUMERICAL MODELLING 32 • Conservation of species • Laws of chemical reactions No detailed descriptions will be given of these extra two conservation laws in this work, expect for relevant information when used in the work. 4.1.1 Continuity The equation for conservation of mass is often referred to as the continuity equation, eq. (4.1): ∂ρ ∂ρui =0 (4.1) + ∂t ∂xi Here ui is the velocity tensor on the compact form using Einstein notation, ρ is the density of the fluid, xi is a spacial first order tensor and t is the time. For an incompressible flow with constant ρ it takes the simpler form: ∂ui =0 ∂xi (4.2) which states that the volume should remain constant. 4.1.2 Conservation of momentum The conservation of momentum in a fluid volume can be expressed by the NavierStokes equations, NS eq., [27]: ∂ui ∂ui + ρuj = ρgi + üûúý ∂t ∂x j ü ûú ý ∂σij ∂xj ρ T ransient V olume ü ûú ý Convective (4.3) ü ûú ý Dif f usive where gi is the volume force(gravity) and σij is the stress on the fluid. On the left hand side the transient and convective terms are found and on the right hand side the volume and diffusive terms are found. σij is written as: A ∂uj ∂ui + σij = −pδij + µ ∂xj ∂xi B (4.4) Here p is the pressure, δij is Kronecker’s delta and µ is the dynamic viscosity. The first part on the right hand side is the static pressure in the fluid and the second part is the viscous stress. Eq. (4.3) and (4.4) are known as the constitutive equations for a Newtonian fluid and holds for a laminar flow. As most flows are turbulent, equations describing turbulent flows are desired. Using 4.1. THE GOVERNING EQUATIONS 33 the method of averaging and Reynold decomposition the Reynolds equation are obtained, [27]: A ∂ ūi ∂ ūi ρ + ūj ∂t ∂xj B = ρḡi + 2 ∂ 1 σij − ρu′i u′j ∂xj (4.5) Eq. (4.5) are also known as the Reynolds Averaged Navier-Stokes equation or in short, RANS. It expresses the mean of a turbulent flow and are similar to that of the laminar flow in eq. (4.3) except for the last part, ρu′i u′j . This part is known as the Reynolds stress and forms a symmetrical second order stress tensor with six unknowns. Thus the turbulent flow contains ten unknowns: six Reynolds stresses- ρu′i u′j , three velocities- ui and the pressure p. However the averaged form of eq. (4.1) and (4.5) only form 4 equations. Hence the system is not closed and this problem is referred to as the closure problem for turbulent flows. 4.1.3 Turbulence modelling As implied the turbulent flow can not be solved directly using only the governing equations. By the use of turbulence models, additional equations are used to close the system. A wide range of models exists with different levels of complexity and approximations. Usually the models are divided into Algebraic models, turbulence-energy equation models and simulation models, where algebraic models are the simplest and simulation models are the most advanced, [27]. Typically for the first two types, the Reynolds stresses are expressed through the product of a turbulence viscosity and mean strain rate. For the algebraic models, the turbulence viscosity is often calculated by a mixing length, and by turbulent kinetic energy in the turbulence-energy model. For simulation models like Large Eddy Simulation, LES, the larger turbulence eddies are sought to be resolved and the smallest eddies modelled. A more detailed explanation will not be given here. Books like [28] can be consulted for further detailed descriptions. The model used in this work is the well tested K −ǫ model which is a turbulenceenergy model. No particular focus have been given to this aspect of the CFDmodel, but kept in mind when resolving boundary layers. 4.1.4 The Energy equation The energy equation can be derived from the first law of thermodynamics as done in reference [26]. The result on compact form is: ρ 3 ∂h ∂h + ui ∂t ∂xi 4 = ∂p ∂ui ∂p ∂ + ui + (k · ∇T) + τij′ ∂t ∂xi ∂xj ∂xj (4.6) 34 CHAPTER 4. GOVERNING EQUATIONS AND NUMERICAL MODELLING where h is the enthalpy, k is the thermal conductivity of the fluid and τij′ is the viscous stress also found in eq. (4.4). The thermal conductivity originates from the assumption that the fluid heat conduction follows Fourier’s law. The variables h, k, ρ and µ in the governing equations (4.1), (4.3)-(4.4) and (4.6) are all a function of the primary variables (p, T) and relations is typically found from experimental data. 4.1.5 Equation of state The equation of state correlates the density with temperature and pressure. The famous ideal gas law can be used for such a correlation. This is usually a fair approximation but fails to predict condensation and evaporation. The model used in the fluid continuum for this work is an ideal gas, where the evaporation of water from Lagrangian particles is done with a dedicated model described in Appendix A.1 on page 97. 4.2 Concepts of CFD This section will give a brief introduction into how the governing equations of the fluid is solved using numerical calculation. Once the fluid domain is defined, a computational mesh can be generated. This mesh is a spatial discretisation of the full domain into finite volume cells. The level of accuracy of the final solution strongly depends on the refinement of this descretisation. Different methods of generating such a mesh exist, where the most common cell types are structured and unstructured rectangular cells, tetrahedral cells and polyhedral cells. Typical simple volume shapes are preferred, as the computational grid is derived from the cell faces. Having complex cell shapes increases the number of faces and thereby equations and computational time. However, for complex shapes with e.g. high curvature and/or sharp edges, cells like polyhedral cells are necessary. The governing equations are non-linear partial differential equations. In order to solve these equations the continuous derivatives are replaced with discrete approximation by the Finite Difference and Finite Volume Method applied on the computational grid. One of the most commonly used approximation schemes are the Central Difference Scheme, CDS. For the finite volume method, the scheme uses two neighbour values to approximate the derivative/flux across the face into the cell. As the flux approximation uses neighbour values all the computational cells are coupled into a system of linear equations of the form Ax = b, [29]. For very large simulations the computational requirements to solve the system 4.3. USING STAR-CCM+ FOR CFD 35 explodes, e.g. a direct Gaussian elimination scales O(n3 ) where n is the number of cells, [29]. Efficient iteration schemes such as multi-grid solvers have been developed and can reduce the scaling to O(n), but the computational cost is still substantial, when the number of computational cells increase, [29]. Especially when working in 3D, the number of cells can explode if not careful. E.g. if the refinement of the discretisation in all three directions is doubled the number of cells growth n = 23 = 8 times. Thus, there is a trade off between spatial discretisation accuracy and the computational effort, which also sets a limit for the size of the domain and the level of details that can be resolved in that domain. Different tools, e.g. the wake refinement tool in STAR-CCM+, are available to refine the mesh in local areas. These are useful when resolving the gradients of the continuum with the discretisation. One of the main focus when building a mesh should be to refine the mesh, such as the gradients are properly resolved. High gradients are typically found in areas with high velocities near a wall due to a no slip condition. This induce velocity gradients. Combustion zones generate high local temperatures and thus high temperature gradients. If not resolved the numerical diffusion of the scalars will not produce correct results. Likewise areas with low gradients allows the mesh to be coarse. 4.3 Using STAR-CCM+ for CFD For this work the commercial CFD software STAR-CCM+ has been used. For simulating physics like radiation and particles, the appropriate models have to be selected and set up in STAR-CCM+. See Table 4.1 for an overview of the used models for simulating of the Lagrangian particles and Appendix A.1 on page 97 for an explanation these models. Other models used for describing the physics in the fluid continuum such as radiation is explained in Appendix A.2 on page 104 and heat exchangers through a porous media is described in Appendix A.3 on page 105. CHAPTER 4. GOVERNING EQUATIONS AND NUMERICAL MODELLING 36 Table 4.1: Table of used Lagrangian models to simulate the Lagrangian phase for coal combustion Group Model(s) used Particle type Material particles Material Multi-Component Coal Equation of State Constant Density Mass Transfer Coal Combustion Energy Lagrangian Energy Model Tracking Track file, Boundary sampling Species Lagrangian Species Moisture Evaporation Coal Moister Evaporation Char Oxidation First-Order Char Oxidation Raw Coal Devolatilization Two-Step Devolatization Optional Models Particle Radiation, Turbulent Dispersion, Two-Way coupling, Drag force 4.3.1 Deposition of particles in STAR-CCM+ The models in section 3.5 on page 22 needed to be implemented in STARCCM+. This was done through field functions describing the sticking probabilities, and a composite BC for particles when hitting a wall or interface boundary. The composite BC was set to rebound the particles as default, and escaping described by the field function for sticking. Thus, if the probability for a particle to stick is zero it rebounds. If not zero it will take the probability calculated for escaping. However through dialogue with the developer of STAR-CCM+, CD-adapco, it was concluded that STAR-CCM+ could not monitor the mass flux of particles leaving the domain(deposits) through wall boundaries. It can only monitor the incident mass flux, which is the total mass flux hitting the boundary. To solve the monitor problem of particle mass flux leaving the system, a somewhat laborious method needed to be used. A method, involving extracting the parcel mass flow escaping and face areas, for post-processing in the programming software Matlab, was developed. A parcel is a Lagrangian particle holding the physical particles. Thus a parcel can, and usually does, contain multi particles. In STAR-CCM+, the parcel mass flow escaping through a wall can be calculated by field functions, using the parcel mass flow rate and sticking probability. Ṁescape = Ṁ · Pstick (4.7) kg where Ṁescape is the mass flow escaping with units [ kg s ], Ṁ [ s ] is the mass flow rate and Pstick [−] is the probability to stick. This value is as mentioned in the kg units [ kg s ] and needs to be in [ m2 s ], which is where the problem in STAR-CCM+ 4.3. USING STAR-CCM+ FOR CFD 37 comes in. No solution was found to extract the face area for which the particles hit, and then use this information in another field function to get the wanted value. Furthermore, the contribution from multiple hits on a face cell needs to be summed which also was an issue. This is due to the way tracked values for particles are stored, which results in, that the information can not be transferred to the boundaries for post-processing. The Matlab code was therefore written to calculate the mass flux escaping from a parcel through the hit face cell, and then sum multiple hits if any. The code needs a Xyz internal table from STAR-CCM+ with the cell face areas of the boundary and corresponding positions along with a Xyz table containing the Ṁescape values for the parcels, and the position for where they hit. For finding the hit cell faces, the effective radius, ref , of the faces are calculated by: ó Af ace 2 Af ace = πref =⇒ ref = (4.8) π where Af ace is the area of the face. If a parcel hit is within this effective radius for any face, the mass flux is calculated and added to the face with the closest face center. The value is stored in a table similar to the area table extracted. ref will always under predict the distance from the face center to the nodes expanding the boundary faces to the neighbour cells. A chance of a particle hit, with the distance to face center, lp , that does not fall under any effective radius is therefore presence. This holds especially true for very long faces, see Figure 4.1. When this occur the ref is multiplied by the ratio of lp and ref to include it. Figure 4.1: Schematic figure of ref and parcels hitting the face but outside the ref . Red dots are cell nodes, green are boundary nodes and black is the impacting particle. The code produce an output .csv file with the total mass flux escaping for the positions giving by the face areas. The produced table can then be imported into STAR-CCM+ where the values can be interpolated onto the boundaries using the interpolatePositionTable function. The actual code is found in Appendix D.2 on page 156. 38 CHAPTER 4. GOVERNING EQUATIONS AND NUMERICAL MODELLING Chapter 5 Results The results from the conducted analyses will be presented in this chapter. The results are divided into two sections. One for the preliminary analyses conducted in order to validate some of the used models and assumptions, and one for the main simulation of the boiler with an integrated steam circuit in SH3. 5.1 Results from preliminary analyses As the final simulation was suspected to be very large and complex, it was decided to investigate some of the most crucial models and parameters in different preliminary simulations. In this way important informations on several parameters could be obtained relatively quickly, as the computational time was much faster for each simulation. A short description along with main results from the analyses is given in this section. The produced results can be found along with an detailed description of these in Appendix B on page 109. This section can be seen as an introduction to the main results of this thesis in section 5.2 on page 42, as these lay the foundations for the parameters used in the full simulation of the boiler. 5.1.1 Main and secondary air supply The primary and secondary air flow was not given for each zone and nozzles, but only as a total flow for the total primary and secondary air supply. The assumption made and calculations done can be found in Appendix B.1 on page 109. A Matlab script was made to calculate the secondary air and can be found in Appendix D.1 on page 154. Reasonable values for the velocities were found. 40 5.1.2 CHAPTER 5. RESULTS The distribution of wood chips on the grate The wood chip distribution on the bed was analysed using Lagrangian particles. This was done to find the correct distribution of wood chips for the bed model and evaluate the momentum from the carrier air jet. The exact geometry of the air nozzles was not given, so the momentum of the jet was evaluated according to the wood chip distribution. To do this, a parameter study of the Lagrangian particles regarding size and restitution coefficient was also done. The whole analysis can be found in Appendix B.2 on page 110. The main conclusion from this study was as following. The values of βn,rest = 0.15 for the normal restitution coefficient and βt,rest = 1 for the tangential restitution coefficient of the Lagrangian particles were adopted for the rest of the work. An rectangle inlet geometry was found for the air nozzles for the carrier jet. This produced a satisfactory momentum and thereby wood chip distribution. The most important parameters for the numerical simulation, besides the momentum of the jet, was found to be the restitution coefficients. 5.1.3 Suspension firing and bed model To simulate the suspension firing, the coal combustion model for Lagrangian particles in STAR-CCM+ was used. The simulation is described in Appendix B.3 on page 118. The simulation produced a combustion in the same areas described by Verdo and was concluded to be satisfactory for this work. The most important parameter for tuning was found to be the particle size of the simulated fuel. A more precise size analysis was desired, as fragmentation during transport of the fuel could not be included. As the suspension firing and grate firing influence each other significantly, the bed model was tuned simultaneous with the suspension firing. This entire work done for bed convergence and calibration can be found in Appendix B.4 on page 125. The developed bed model and the produced temperature profile was calibrated after a real thermal picture of the grate. The model predicted a rough profile matching the thermal picture. The main combustion was predicted in the half closest to the front wall. Unfortunately, a misunderstanding between Verdo and Force Technology regarding the front and back wall in the thermal image, resulted in a reverse profile of the bed model. Thus the main combustion should have been nearer the back wall. This misunderstanding was found to late in the project to change and effects the entire work. However, due to the configurations of the secondary air nozzles, the main flow was estimated in consultancy with the supervisors of the project, not to change drastically above these. The error should not discard the corrosion model, as the net release of energy and species would be the same if the main combustion was moved further to towards the back wall. 5.1. RESULTS FROM PRELIMINARY ANALYSES 5.1.4 41 Simulating the steam in SH3 Due to numerical instabilities caused by an initial bad mesh of the SH3 tubes, the tubes were meshed and simulated in a separate simulation. The work done in the separate simulation produced a far better mesh and an initiation solution for the steam region in the full scale simulation. The better mesh was produced with the generalized cylinder meshing tool in STAR-CCM+. The analysis showed that the Cp value for the superheated steam needed to be altered in STARCCM+. The default value under predicted the value by almost a factor 2, as it was based on the atmospheric pressure. The actual pressure in the tubes is 109 bar. Using a constant uniform heat flux as a BC to match the total energy extracted by SH3, the average outlet temperature of the steam was found only 1.5◦ C from the actual value. This corresponds to a deviation of the numerical value of approximately 3 % from the real value. 42 5.2 CHAPTER 5. RESULTS Main results The work produced a full simulation of the second boiler at Verdo, with a fully resolved and integrated co-simulation of the steam circuit in SH3. The conditions simulated was a full load condition. In the following the main results of this will be presented. First a description of the mesh and physics used for the simulation is presented. Next, the main results for the general freeboard in the full simulated domain is described. This leads to a more detailed investigation of the region where the SH3 is located, which includes the application of the corrosion model and results for the coarse ash deposition model. 5.2.1 The mesh used for simulation with integrated steam circuit of SH3 The mesh was generated using polyhedral cells. On the furnace walls only one prism layer was used as the main flow in a furnace is not very boundary sensitive. On the SH3 tubes two prism layer cells was used in order predict the flow and heat transfer better. For the mesh inside the tubes in SH3 the mesh produced in Appendix B.5.1 on page 129 was used. STAR-CCM+ automatically makes the interface between the main freeboard mesh and the imported mesh for the SH3. The mesh used for the simulation contained in total 11.557.129 volume cells, see Figure 5.1 for global mesh. Volumetric control volumes were used to make the mesh coarse between the tube rows, as the mesh could not grow in size in between otherwise. This saved in the order of 3 million computational cells. In Figure 5.2 the mesh around the SH3 is shown. 5.2. MAIN RESULTS 43 Figure 5.1: Global mesh used for the main simulation. 11.557.129 volume cells with one prism layer on the furnace walls, and two on the SH3 tubes. The grey cells are wall boundary cells, tan is volume cells in the mid plane and yellow is in-place interface cells towards SH2. Figure 5.2: Mesh around SH3 tubes. Two prism layer cells on the SH3 tubes with a stretching of 1.5 and total thickness of 1.5 mm. 44 5.2.2 CHAPTER 5. RESULTS Physics, models, BC’s and convergence All the models described previously in section 4.3 on page 35 and Appendix A on page 97 were used to generate the results in this section. Beside these, the altered Cp value for the steam circuit described in Appendix B.5 on page 129 was used. Instabilities in the Bi Conjugate Gradient Stabilizer(BiCGStab) in the Algebraic Multigrid solver, AMG solver, for the pressure solver was seen. A W cycle was there fore used for the AMG pressure solver. The W cycle has more relaxation sweeps between grid levels than the V cycle and thus more robust for stiff systems, [30]. As the surfaces of the physical tubes in SH3 are subjected to condensation, deposition and fouling of ash and alkali metals, the thermal properties must be expected to be different as well. Two of the most important properties that are expected to influence the simulation are the radiation emissivity of the surface and thermal resistance of the ash layer. The default value in STAR-CCM+ for emissivity of 0.8 fits the overall surfaces in the furnace consisting of bare metal, but not a surface with ash depositions. An estimated mean value of 0.55 fore the SH3 BC is found in reference [39] and used. The thermal resistance of the fouled surface must also be considered, as ash deposits have a very low heat W conductivity. In reference [40] the value of 0.14 m·K is found, and adopted here. 5.2. MAIN RESULTS 45 Convergence of the full scale simulation In Figure 5.3 the residuals for the simulation are presented. It shows relatively high residuals in the order of 1 − 20% for the most parameters. The two with really low residuals are the steam and coal volatile species. The spikes seen in especially the turbulent dissipation, Tdr, is caused by Lagrangian particles injected. However the residuals are relatively stable and no indications of further convergence can be seen. To better evaluate convergence, monitors of the temperature, heat flux through the tube walls and pressure of the steam in SH3 were set up. In Figure 5.4 the average outlet temperature is plotted. It is seen that minor fluctuations occur but is converged towards a temperature of 494 ◦ C. The monitored outlet temperature from Verdo was found as 490 ◦ C in Appendix C.2 on page 137. The pressure and heat flux monitors can be found in Appendix B.6 on page 133. These also show a stable converged solution in the SH3. The stable values in the tubes indicates converged flow and thermal conditions in the freeboard, as the two regions are coupled. It was therefore concluded that the simulation had converged towards a reasonable solution. Figure 5.3: Residuals for the full scale simulation with the steam in SH3 simulated as an integrated circuit. The large fluctuations in the turbulent kinetic energy, Tke, is caused by injection of Lagrangian particles. 46 CHAPTER 5. RESULTS Figure 5.4: Monitor plot of the average outlet temperature for the steam in SH3, in the the full scale simulation with the steam in SH3 simulated as an integrated circuit. Minor oscillations at are seen. A further discussion of the residuals can be found in section 6.5 on page 69. 5.2.3 Results for the full domain In Figure 5.5 the mid plane temperature profile are shown. The lower part of the furnace are much the same as in the analysis of the suspension firing in Appendix B.3 on page 118. The temperature in the region where the SH3 is located, top left corner, are in the range of approximately 700◦ C to 1100◦ C. The highest temperatures are found in the center of the hot column. The temperature just before SH3 is around 1000 ◦ C, which is the melting range for silicate, see Figure 3.8. In the straight section before the outlet of the domain the effect of Eco1-3 and SH1-2 can be seen as the temperature drops to around 500◦ C. The porous regions simulating these are outlined in Figure 5.6. Here the temperature through the furnace and up through the super heaters and economisers are shown by horizontal profiles. 5.2. MAIN RESULTS 47 Figure 5.5: Temperature profile of the mid plane. SH3 are located in the top left corner just above the narrow passage. The high temperatures indicates the grate and suspension firing. The temperature just before SH3 is around 1000 ◦ C, which is the melting range for silicate, see Figure 3.8. Figure 5.6: Horizontal temperature profiles at y= 0.1 m, 2.5 m, 4 m, 6 m, 8 m, 10 m, 12 m, 14.5 m, and 18 m above the grate. Notice the effect of the secondary nozzles, in how they push the main flow together. 48 CHAPTER 5. RESULTS In Figure 5.6 the narrow band making up the main flow are clearly shown as the secondary jets push the main flow together. The high concentrated temperatures induce a convectional driven current with high velocities in the centre. The velocities can be seen in Figure 5.7. The colorbar is limited at 15 ms as showing the velocities of the secondary and carrier jets would cause a poor resolution of the interesting velocities in the freeboard. Some of the jets are seen as the large red vectors. The other jets are not visible in the figure as the mid plane does not cut through them. Figure 5.7: Vector plot of the velocities at the mid plane. Due to very high inlet m velocities of 43 m s at the carrier jets, the velocity field is limited at 15 s to more clearly see relevant velocities. A large low velocity eddy is seen at the upper back side of the furnace. In the right upper corner of the furnace just before the wall contraction a large low velocity eddy is located. This part of the furnace also have a low temperature as shown in the earlier figures. The low temperatures are caused by these eddies recirculating the same flue gas. This is a ineffective region for extracting heat. In the last part before the outlet, the pressure drop over the porous regions evens out the flow to a more homogeneous velocity profile. The flue gas is accelerated as it is forced through the narrow passage. Right after the passage, the flow separates due to the sharp expansion of the freeboard, causing a small eddy between SH3 and SH2. The black vertical line in the furnace to the left, is the 5.2. MAIN RESULTS 49 Figure 5.8: Vector plot of the velocities at the mid plane near SH3. Just after the expansion an eddy is seen, as the flow separates from the wall. The maximum velocity near the SH3 is around 12 m s. plane used in Figure 5.9. In Figure 5.8 the velocities in the mid plane in the region where the SH3 is located are shown. The highest velocities are around 12 ms right after the contraction. Including Figure 5.9, which shows a crosssectional velocity profile around the SH3, the 12 ms are the highest velocities and found near the mid plane. Figure 5.9 also shows to large eddies near the loft, one in each side. These are probably caused by the main flow hitting the loft and acting as an impinging jet. The tubes are also outlined in this figure showing the 14 rows. The vertical black line is the mid plane used in the other figures as well. The large eddies cause the flow to pass the tubes almost perpendicular to the tubes in some areas, instead off parallel as in the center of the SH3. This could be a factor for fouling, c.f. Figure 3.11 in section 3. Figure 5.9 also indicates that the solution is not completely symmetric. This originates from the random Lagrangian particles. It could also be caused by premature injection of the first Lagrangian particles before symmetry was obtained or too few iterations between each particle injection. This will be elaborated and looked further into in section 6.6 on page 70. A more evenly distributed flow would give a better heat transfer to the SH3 and thus a higher efficiency for the power plant. 50 CHAPTER 5. RESULTS Figure 5.9: Vector plot of the velocities across the center of SH3. Two large eddies are seen in the top corners. The highest velocities are seen in the center of SH3. The oxygen concentration at the mid plane is shown in Figure 5.10 plotted as the mole fraction. The colorbar was limited at 0.1 to highlight the oxygen lean areas. The blue/green areas represent the areas where a combustion have used the oxygen. It is however not clear if there is a combustion in the flue gas all the way to SH3, or the lean zone is an effect of poor mixing and thus originates from further back towards the bed. Using the same cross-sectional plane as earlier, an oxygen profile across the furnace is produced in Figure 5.11. It is clear that the main combustion takes place in the middle of the furnace as there is relatively much oxygen near the side walls in the top. At the side walls near the bottom, a combustion of the larger particles hitting the side walls takes place and use the oxygen. The oxygen lean areas will cause a reducing environment near the metal surface, which was one of the prerequisites for the corrosion, c.f. section 3.3.2 on page 19. Thus the lean oxygen region in the middle of SH3 will contribute to the corrosion process. The U-shaped yellow contours in Figure 5.11 at near the suspension firing are caused by a combustion between the CO released from the bed and the oxygen in the carrier jet for the wood chips. The high CO concentration from the bed can be seen at the bottom part of the furnace in Figure 5.12. 5.2. MAIN RESULTS 51 Figure 5.10: O2 distribution at the mid plane. Concentration plotted as mole fraction. Cutoff at 0.1 to highlight oxygen lean band. The suspension firing is clearly seen as the oxygen lean area. Figure 5.11: O2 distribution at the transverse plane used previously in the middle of narrow passage. Concentration plotted as mole fraction, cut off at 0.1 mol to highlight oxygen lean band. 52 CHAPTER 5. RESULTS Figure 5.12: CO distribution at the mid plane. Concentration plotted as mole fraction. This CO profile is in the mid plane and the combustion process with the carrier jet are therefore not visible as the plane does not intersect these jets. Again, the area where the main suspension firing takes place is visible as the CO rich area in the middle of the furnace. The combustion seems to continue all the way to the SH3 as the concentration diminish. The CO is one of the parameters power plants often measure further down in the system, as this is an indication of how good their combustion is. In Figure 5.13 a horizontal profile of the CO just before the outlet is shown. The outlet section of the simulated domain is the blue and yellow area to the right. The red area is the equivalent profile in the furnace. The colorbar is limited at a highest of 2 · 10−4 and lowest of 2 · 10−5 to focus on the outlet section. It can be seen from this that the profile is not homogeneous. A surface average was calculated on the outlet witch showed a concentration of 61.4 ppm CO. 5.2. MAIN RESULTS 53 Figure 5.13: CO distribution horizontal plane right before outlet at the left. Concentration plotted as mole fraction. Surface average concentration at outlet calculated as 61.4 ppm. An important parameter for the corrosion process was the presence of KCl. In Figure 5.14 the concentration at the mid plane are shown. The release of KCl is done both from the bed and suspension firing, where the amount is found from the ultimate analysis and eq. (3.10). The main concentration follows the same flow patterns as for the oxygen and CO just described. Sadly the sulphation reaction in the flue gas between KCl and SO2 described in eq. (3.1) was not done. Thus the concentration levels are to high. However as the SO2 is released under almost same conditions and therefore place, the reaction only decrease the concentration of KCl and not distributes it. A cautious estimation of the critical areas for corrosion from a KCl concentration point of view, can therefore still be done. In Figure 5.15 a transverse KCl profile in the previously used plane are shown. A main concentration is seen in the center of the furnace but also some high concentrations near the side walls. These originates from the suspension firing. An estimation from this combined with Figure 5.14, is that the most exposed parts of the SH3 for KCl are the middle rows lower half. 54 CHAPTER 5. RESULTS Figure 5.14: KCl distribution at the mid plane. Concentration plotted as mole fraction, cut off at 0.0004 to highlight KCl rich band. Figure 5.15: KCl distribution at previusly used transverse plane in the middle of the narrow passage. Concentration plotted as mole fraction, cut off at 0.0004 mol to highlight KCl rich band. 5.2. MAIN RESULTS 5.2.4 55 Summary of the general freeboard The simulation showed that the main flow is pushed together as a narrow band by the secondary air jets. This generates high velocities and relative high concentrations of all the flue gas species except for O2 . The oxygen had a lower concentration in this band due to the combustion, which could accelerate the corrosion. Two large eddies are seen in the SH3 region. The eddies are probably caused by the main flow acting as an impinging jet on the sealing of the furnace. The eddies cause large temperature differences in the regions. The most exposed parts of the SH3 for KCl are the middle rows lower half. 5.2.5 The region near SH3 The temperature of the tube surfaces of SH3 are shown in Figure 5.16 along with a temperature profile of the flue gas just after SH3. Unfortunately the resolution of the plotting feature in STAR-CCM+ is not good enough to see details when plotting all the tubes together. However the overall tendency is clear and relevant areas will be highlighted later. Figure 5.16: Surface temperature of SH3. Inlet temperature of steam, Tin = 448◦ C. Average outlet temperature of steam, Tout = 492◦ C. Left colorbar represent the scale of the SH3 surface temperature. Colorbar at right represents the temperature of the flue gas in the plane just after the SH3. The highest temperatures are found in the upper part of the middle tube rows near their outlets. Naturally the longer the total length of a pipe, the higher 56 CHAPTER 5. RESULTS the outlet temperature. Thus the front tube, which does not have an extra loop have a relative low outlet temperature. The high temperature of 562◦ C was a bit surprising as the mean outlet temperature is 490◦ C according to Appendix C.2 on page 137. The calculated average temperature was found to 494◦ C, see Appendix B.6 on page 133 for monitor plot of outlet temperature, residuals for simulation, monitor of max. pressure in tubes and total heat flux through SH3’s surface. The upper threshold for corrosion was found to be 520◦ C in section 3.3 on page 17. The plot in Figure 5.17 are therefore generated with this being the upper limit for the colorbar. This also highlights the great temperature differences of the tube surfaces. Figure 5.17: Surface temperature of OH3. Cutoff at T = 520◦ C in left colorbar representing the scale of the SH3 surface temperature. Colorbar at the right represents the temperature of the flue gas in the plane just after the SH3. It is clear how the uneven temperature field effects the load distribution of SH3 The large eddies shown in Figure 5.9 actually cause a slight cooling of the outlet steam in the tubes near the side walls, although not obvious from the presented results. This is a result of the eddies and the relative cool side wall temperatures of 330◦ C that produce a cold current in near the side walls. The steam temperature in one of the middle rows are seen in Figure 5.18. This illustrates the relative cold front tube compared to the other tubes with an extra loop, and a general idea of the temperatures in the different tubes of a single row. Notice the outlet temperature of near 515 ◦ C for the inner tubes, which is 25 ◦ C above the mean. 5.2. MAIN RESULTS 57 Figure 5.18: Temperature inside a middle row of tubes. It can be seen how the outer tube has a relative cold outlet temperature compared to the other tubes. An outlet temperature of nearly 515 ◦ C is seen for some tubes. Figure 5.19: Velocity inside a middle row of tubes. Higher velocities are seen near the bending of the tubes. Likewise the magnitude of the velocities are shown in Figure 5.19. The inlet velocity was set to 13.59 ms , meaning that the maximum velocity reaches approximately 50% higher values at corners than the mean. The steam flow direction is down from the left and up at the left, which is also indicated in the temperature plot as well, since the steam is heated. The inner five tubes have an extra loop going up and down compared to the outer tube. 58 CHAPTER 5. RESULTS The literature study produced a model for prediction of corrosion risks from the surface temperature of metals. This was applied on the SH3 tubes and generated the plots in Figure 5.20 and Figure 5.21. Here it must be stressed that this is only a risk assessment from the metal temperature. The model divide the risk into three categories: Low, Medium and High. Here a low risk is areas with little or no risk of high temperature corrosion. Medium risk areas are areas where high temperature corrosion is likely to occur if other necessities are fulfilled, described in section 3.3 on page 17. High risk areas are areas, where high temperature corrosion will occur if the other necessities are fulfilled. Figure 5.20: Risk of corrosion due to surface temperature alone. It is seen that the middle rows are the ones most prone to exhibit high temperature corrosion from this plot. In particular the back tubes have an elevated risk. The corrosion risk on the tubes is low at the front tubes, near the steam inlet for all tubes. The risk is highest for the tubes in the middle, especially near the outlet of the steam. 5.2. MAIN RESULTS 59 Figure 5.21: Detailed view of the risk of corrosion due to surface temperature alone in the lower middle part of the tube bank. The spacing between the rows in the middle, that is just a bit larger than between the other rows, is the middle of the SH3 and boiler. The small high risk areas near the bending indicates an elevated risk of corrosion on the center front side tubes near the tip. As shown in Figure 5.21 there are some high risk regions at the lower tip of the tubes as well. This is in the region where Verdo have experienced high corrosion levels as well. The potassium chloride concentration in the boundary cells can be plotted on the surface of the tubes. This is done in Figure 5.22. The uneven distribution of the KCl is caused by an uneven distribution of the Lagrangian particles after injection. The three main areas originates from the three suspension firing areas. It should be noticed that the maximum concentration only is approximately 50% higher than the minimum. Thus, there is a relatively even distribution. 60 CHAPTER 5. RESULTS Figure 5.22: The KCl mole fraction in the boundary cells of the SH3 plotted on the surface of SH3. A relative even distribution when looking at the scale. All though three main areas can be pointed out near the side walls and in the center. 5.2.6 Summary of corrosion risk from surface temperature in the SH3 region The model generally predicts that there is a medium or high risk from the surface temperature event in the last section of the tubes, and low risk on the front tubes with the relatively cold steam. In Figure 5.21 the model predicts high risk areas on the outer tube in the area facing the flow and in the middle of the free board. This fits the corrosion profile over the tube banks seen at Verdo. On the tubes coming down again from their second loop the model also predicts high risk areas. No corrosion information exist at the presence for this section. 5.2. MAIN RESULTS 5.2.7 61 Coarse ash deposition To produce a final risk assessment for high temperature corrosion the deposition of KCl was found to be important. One of the main cause of deposition of KCl was found to be through impacting particle deposition. An analysis of impacting particles was conducted according to the models discussed in section 3.5 on page 22. Both the model used by Kær in reference [9] and the model developed by the author of this work were tested. Each model was tested for 25.000 and 100.000 parcels, but only the results for the 100.000 parcel analysis is shown. For the analysis, the particles from the suspension firing along with particles released from the bed was used. The particle diameter of the particles from the bed was set as a uniform distribution from 10 µm to 1 mm, [9]. All of the analysis were produced from a baseline simulation ensuring the same velocity field for each analysis. In Figure 5.23 the two models are compared. A limitation of kg 1 h·kg on the colorbar is used to highlight the important areas. This limitation kg cm corresponds to 0.96 day when using a density of 2500 m 3 for the ash. Some cells kg experienced deposition rates as high as 200 h·kg . This is unnatural high and was caused by very small cell face areas in combination with impacts of parcels carrying large mass. The value of 1 represented a reasonable upper limitation of the colorbar, removing the abnormalities and focusing on the main tendencies. 62 CHAPTER 5. RESULTS (a) Deposition on the side walls predicted by the model developed in this work. (b) Deposition predicted by the model developed by Kær in reference [9]. Figure 5.23: Comparison of the two models used to predict fouling a) The model developed in this work b) The model used by Kær in [9] It is clear that the model of Kær predicts way more deposits on the furnace walls compared to the model developed in section 3.4 on page 20. As discussed 5.2. MAIN RESULTS 63 later in section 6.6 on page 70 the modified model predicts the actual conditions at Verdo better, and is therefore the one used when looking at deposition on the SH3 in Figure 5.24 and Figure 5.25. Figure 5.24: Deposition of ash particles at SH3. The units are hkg and the value of ·m 2 kg cm 1 equals 0.96 day when using an ash density of 2500 m3 . Cutoff at 1 due to insufficient particle resolvement and to fine mesh, and thereby some unnatural high deposition rates locally. 100.000 parcels was used. The numerical predicted deposition on SH3 is illustrated in Figure 5.24. It shows two main areas of deposition. One being the middle of the tubes in the top third, and the other being the front side of the tubes near the bottom. As the bottom is where the corrosions are measured as well a more detailed plot is showed in Figure 5.25. The main reason for high deposition rates in these areas in particular are the part of the model that includes the impact angel. The areas with the highest deposition rates are therefore typically where main flow is not parallel to the pipes. This corresponds well to the conditions seen at Verdo which will be elaborated in section 6.6 on page 70. 64 CHAPTER 5. RESULTS Figure 5.25: Deposition of coarse ash particles at lower right part of SH3. The units are kg kg cm h·m2 and the value of 1 equals 0.96 day when using an ash density of 2500 m3 . Cutoff at 1 due to insufficient particle resolvement, and thereby some unnatural high deposition rates locally. 100.000 parcels was used. Heavy depositions are seen at the front of the tubes facing the flow direction. The deposition on the tubes was also investigated with the model by Kær. However as there is no certain growth rate for the deposits at Verdo it was not possible determine if one model is more valid than the other on this background. cm The developed model predictions of approximately 0.96 day fits measurements conducted at the biomass fired power stations used in reference [5]. These show cm deposition rates between 0.2 to 4 day in locations similar the one of SH3. Summary of coarse ash deposition A new numerical model for coarse ash deposition have been develop and tested against an existing model developed by Kær in [9]. The new model fits the fouling conditions better at the side walls. Both models gave abnormal local deposition rates on the tubes due to too fine a mesh on the tubes. A rough cm mean value of the deposition areas showed deposition rates in the order of 1 day on faces towards the flow direction. Chapter 6 Discussion A discussion of the produced results and assumptions behind will be given in this chapter. As the found values, used parameters and models in the preliminary results were of such importance for the final full simulation these are discussed here as well. 6.1 Issues when simulating spreaders for wood chip During simulation of the wood chip distribution in Appendix B.2 on page 110 the carrier jet was approximated with a rectangular inlet boundary. Due to lack of information at this stage of the project it was concluded to be a reasonable approximation. A visit to Verdo, during shut down of one the boilers gave the opportunity to see the actual geometry, see Figure 2.2 on page 7. The jet nozzles are the small holes below the horizontal spreader plate. From this, it seems that the assumption of a rectangular shape was OK, as there is less than one hole diameter between the holes. In the feeding duct there are some dust deposited in small banks. This was the same tendency that the Lagrangian particles showed when βt,rest was not 1. The assumption made of setting it to 1 could therefore be wrong. However in the real duct the wood chips bounce together and push each other towards the jet. This is not the case for Lagrangian particles as they do not have any interactions. The dust banks are also much smaller in particle size than the wood chips which could be a contributing factor for these formations. According to Verdo some wood chips fell on the first half of the bed as well, which the analysis did not show. The prediction of a more concentrated distribution in one area by the model indicates that more work could be done on the restitution coefficients. A quick fix to this could be to have the coefficients over an interval 66 CHAPTER 6. DISCUSSION instead of a fixed value. However more scientific analyses for supporting the models are desired. Also the shapes of the wood chips are expected to have an important influence on the trajectories and distribution. Lu et al. in reference [41] and Rosendahl et al. in reference [42] also points out the importance of the shape deviations from spherical. Rectangular shapes will typically have a higher Cp value and a different cross-sectional area which typically both increase the drag and thus the potential travel distance in suspension. In Appendix B.2 on page 110 the Discrete Element Model (DEM) in STAR-CCM+ is mentioned as a possible solution to resolve some of these issues. The DEM model does not include the coal combustion model. However, the analysis in Appendix B.3 on page 118 shows that the evaporation and thus the combustion can be neglected for large wood chip particles in suspension. For a pure analysis of the distribution this could therefore be a better tool. More information about the wood chip mass flux distribution over the injector surface could also improve the accuracy. For this study an uniform mass flux over the wood chip inlets was used. 6.2 Problems and important parameters when simulating suspension firing Some of the same issues as for the wood chips, are found regarding simulations of suspension firing. The shape however may not be as critical at Verdo, as much of the fuel(the pellets) gets broken up into smaller more spherical shapes in the feeding ducts and conveyor belts. However, this gave rise to another problem as there is no way of predicting the fragmentation of these pellets. The size distribution analysis was conducted on fresh unfragmented fuel samples, see Appendix C.6 on page 144. The actual size distribution of the fuel when entering the furnace are therefore very different. At a visit to Verdo, fragmentation of the fuel for suspension firing on the conveyor belt was evident. Unfortunately no size analysis was conducted on this part. For this reason a distribution of 50 wt% for small dust particles with 1 mm diameter and 50 wt% for pellet particles with 5 mm diameter was used. At the late stage of the project, Verdo suggested a distribution of 75/25 towards the smaller particles. As the used distribution lie in the middle of the fuel analysis and the estimations of Verdo, the used sizes are found reasonable. However, the simulation analysis for the suspension firing showed a dependency of the particle size. Thus for future work a fuel size distribution analysis as close to the injection point would be desirable. Fragmentations of the particles during combustion was not included in the simulations. The primary reason being that STAR-CCM+ can not simulate this for solid particles, only for droplets impinging walls. Work done by Syred et 6.3. ERRORS AND UNCERTAINTIES REGARDING THE BED MODEL 67 al. in reference [43] shows the effect of fragmentations and conclude, that for an accurate prediction of the combustion, fragmentation models must be used. As the combustion process in it self is not the main focus of this work, the use of no fragmentation models are found OK. Naturally, the combustion must not be completely off, but as the main combustion of the suspension firing was found to match the flame seen by Verdo, the model is found valid. For more precise tuning of the model, more precise experimental results must be used as a reference. This is also out of the scope for this thesis and the project at Force Technology. 6.3 Errors and uncertainties regarding the bed model The bed model was adapted to handle a spreader distributed fuel supply. The technique predicted the temperature profile within 50◦ C for the main combustion areas. However, at a presentation of the results for Verdo in the late stage of the project, a misunderstanding of the thermal image was exposed. The lack of information on the thermal image regarding physical orientation, and a misunderstanding between Verdo and the author resulted in an inverted bed model of the grate profile. The front wall was mistaken for the back wall, thus resulting in the inverted bed profile. The misunderstanding arose from the assumption made in Force Technology that the grate was moving. The grate only have one revolution a day(≈ 0.5 m h ), which makes it almost stationary regarding the combustion. The main area of grate fired combustion and heat release is therefore in the back of the furnace where the wood chips land and not at the front. The combustion process on a grate for a biomass incinerator is typically divided in to four zones starting with an evaporation zone from the injection side, then a pyrolysis and combustion zone, a char combustion zone and a char gasification zone at the back of the grate. The fuel undergoes these four steps as it is transported across by the movement of the grate. An important cause for this profile is the way the fuel is feed in as a relative thick layer from the front wall. The bias regarding the combustion areas obtained from waste incinerators was therefore a contributing factor. This is of course very unfortunate, but the error was discovered too late in the project to change it. The error clearly effects the combustion at the lowest part of the furnace. The question is how much it effects the corrosion model and conditions in the SH3 region. It can be argued, that due to the settings of the secondary jets, the main flow above these is almost independent of the profile of the bed. The four rows of secondary jets at the back wall push the main flow towards the front wall, as explained in section 5.2 on page 42. Assuming that this is the case, the conditions near the SH3 should not change drastically by moving the location of the combustion zone of the bed. This assumption was shared by the supervisors of the project, and much of the flow conditions was 68 CHAPTER 6. DISCUSSION still recognizable by Verdo in-spite of this error. Thus the error is not expected to dismiss the corrosion model, but should be kept in mind. For the interest of Verdo the error could be undone relatively quickly, but due to the post-processing effort needed for this project it was not done here. 6.4 Mesh limitations for boiler simulations The large size of incinerator boilers in combination with very small details, in especially the curvature of the heat exchanger tubes, makes it very difficult to fully simulate using CFD. To the authors knowledge the discretisation of superheaters like SH3 in combination with a full incinerator simulation have never been attempted before. In reference [44] Saripalli et al. attempts with a semi-couple model, where sub-simulations are coupled. The normal approach for simulating heat exchangers are normal done through porous media. Either as a whole block like the ones used in this work for the rest of the heat exchangers, or by slap approximations as done by Kær in reference [9, 10]. In the light of this, the approach made in this work seems very ambitious. However, detailed information of the specific load on the SH3 tubes seems to justify this. By fully simulating the tubes, the computational mesh size increase drastically. The simulations done by Kær was done on 600.000 computational cells while an excess of 11.500.000 cells was used in the simulation here. It should be mentioned, that major progress have happened in the available computational power in the decade between. The minimum requirements for the main simulation was around 24 Gb of memory and preferable 36 Gb corresponding to 2 and 3 servers with 8 CPU cores at each server at Force Technology. Besides from the computational requirements, a heavy graphic card must be available in order to render the plots of the tubes in post-processing. As the calculations was at the limit of the available computer power, a finer mesh was not attempted. Likewise, no mesh convergence analysis was done, as it was not possible to increase the mesh size any further. Instead, great effort was put into bringing the cell count down in less important areas and refinement in important areas. The fine mesh of the tubes presented another problem when simulating fouling. The amount of parcels needed for convergence of deposition was not found due to too many face cells on the tubes, and not enough computational power for enough Lagrangian parcels. That said, it did not seem impossible to produce reasonable results using roughly 100.000 parcels. An estimation of the needed number of parcels can be found by dividing the cross-sectional area of the freeboard in the region of the SH3, by the area of an average cell face of the tubes. By this, a total of approximately 250.000 parcels are needed. One solution to lower the number of needed parcels, was to use a fictive coarse surface mesh of 6.5. STABILITIES OF COMBUSTION SIMULATIONS WITH LAGRANGIAN PARTICLES AND MULTIPLE REGION INTERACTIONS69 the tubes. The deposition flux from the smaller cells could then be projected to the coarse area to get the mass flux on the fictive surface. This mass flux can then be projected back onto the fine mesh for post-processing. By doing this, the flux is evened out and convergence of the parcel numbers could be reached earlier. This should be a feasible way, as the number of physical particles in each parcel is increased as the number of parcel decreased. Also, the total deposition mass flux is found in the external code. Changing the used area here to the fictive coarse surfaces is easy, and the projection back is done through the interpolation function in STAR-CCM+. 6.5 Stabilities of combustion simulations with Lagrangian particles and multiple region interactions It proved difficult to get low residuals on especially the combustion species and energy, c.f. Figure 5.3 on page 45. However, it was found from monitors of the outlet temperature of the steam, maximum pressure in the tubes and heat flux through the tube walls, that the simulation had converged and stabilised at a reasonable solution. One could argue that the residuals in the simulation could be reduced even more. This could maybe be done through lowering the under relaxation factors. But, as the flow is coupled through so many elements such as randomness in Lagrangian particles, the combustion reactions, the momentum and energy conservation, radiation and last but not least the interface between the SH3 region and the general freeboard, lowering the relaxation factor could also have a negative effect. In practise there are two coupled solutions that is sought at the same time. Thus when an change in the solution in freeboard happens near the inlet of the steam in SH3, this effects the entire SH3 region. While this change converge it effects the freeboard near the steam outlet and so on. This extra coupling is expected to increase the residuals and can introduce some fluctuations in the solution and makes it a form of transient. The small fluctuation seen on e.g. the outlet temperature in Figure 5.4 on page 46 could be an artefact of this. When combining these fluctuations with the random walk of the Lagrangian particles it is not surprising that the residuals are at a relatively high level. Further more, the experience at Industrial Processes at Force Technology shows, that residuals for a combustion in STAR-CCM+ are generally high. The solution found for the presented results are therefore found to be valid for both a general picture of the simulated condition and to apply the corrosion model on. 70 6.6 CHAPTER 6. DISCUSSION Deposition of particles The model for predicting depositions of coarse ash particles developed by Kær in reference [9] was adopted and modified as well. The modifications done to the model of Kær was done as it was expected to over predict depositions. The two models was run under the same conditions and produced the results back in Figure 5.23. The model by Kær predicted heavy fouling on the side walls as expected, where the modified version only showed minor depositions. At Verdo there were almost no fouling on the side walls of the furnace, which supports the suspicion, see Figure 6.1. (a) Picture of the furnace walls. Little fouling are seen on the walls. (b) Picture of the a deposit build up at the bottom of SH3 at the 6th row from the left when seen from the front. Furnace front wall in background. Figure 6.1: Pictures from inside of the furnace in Verdo at shut down. a) Little depositions are seen on the side walls. b) Heavy fouling is seen on especially the front side(bottom) of the SH3 tubes. However, that the modified model produced areas with depositions on the side walls at all was unexpected. The model should produce no depositions with the surface temperatures found at the side walls. The temperature was set as a fixed temperature according to Verdo. The explanation should be found in the way STAR-CCM+ extracts the temperature of the impact surface. The field function, for describing the probability of deposition due to surface temperature, is based on the temperature on impact. This was thought to be the boundary value of the surface, but is instead the value of the volume cell. Thus instead of extracting the surface temperature it extracts the temperature of the boundary layer. The results for the developed numerical model for predicting fouling areas was presented in Figure 5.24 and Figure 5.25 on page 63. It was clear that the model predicted heavy deposit build up on the front side at the bottom of the tubes and in the top half of the middle tubes. The deposit at Verdo at the bottom of 6.6. DEPOSITION OF PARTICLES 71 SH3 is shown in Figure 6.1(b). It is seen that there is a good correlation between the numerical predictions and the conditions at Verdo. The large eddies seen in the region can explain a phenomena seen at Verdo shown in Figure 6.2(b). Here the deposits builds up on the side and not front. This is probably due to the eddies. In Figure 6.2(a) the deposits seen on SH2 are shown. These deposits are located one meter behind SH3. No attempt to predict the fouling on SH2 was done, but it is fair to believe that the large region with depositions on SH3 just in front predicted by the model, can be correlated to these formations. (a) Picture of the deposit build up at the middle of SH2 facing the flow direction. (b) Picture of row 10 in SH3 from the right. The deposits have a growth direction towards the center of the furnace indicating the flow direction. Figure 6.2: Pictures from inside of the furnace at Verdo. a) Ash deposit at the top and front of the middle part of SH2. b) Ash deposits on the side of row 10 in SH3. Overall, there is a very good correlation between the numerical predictions and the deposit formations at Verdo. However, the issue with STAR-CCM+ not extracting the actual wall temperature have to be resolved. Furthermore, Kær mention that the cooling in the boundary layer is an important factor for the stickiness of especially the small particles. As there was only used one prism layer cell to resolve the boundary layer of the furnace walls and two at the SH3 tubes this could be a source of error. This could be a subject to investigate further in a test case, with the effect of particle size and boundary layer resolution on the particle stickiness being the main topics. Also a more scientific base for the stickiness due to impact angle could be desired all though the one used seems to behave OK. In section 3.4 on page 20 it was also shown how the content of KCl influence the stickiness of the particles. As the sulphation reaction between the fluid continuum and the particles was not succeeded, it was assumed that the particles 72 CHAPTER 6. DISCUSSION would have a minimum of 15 wt% KCl in order to satisfy the probability curve in Figure 3.10. In the department of Korrosion og Metallurgi at Force Technology, the chemical composition of the depositions at the SH3 tubes was analysed, [23]. This showed that the major contents of: K = 33 wt%, SO4 = 19 wt% and Cl = 18 wt%. Here Cl is likely to be bound as KCl, and SO4 as K2 SO4 . The molecular weight is almost the same for the two components giving an approximately content of KCl = 36 wt%. Assuming that this is the content of the deposited coarse particles the criteria of 15 wt% is well met. However for a general fuel composition the chemical reactions in the flue gas reactions must be included in order to predict fouling before the boiler is build. 6.7 Risk assessment for high temperature corrosion in the Verdo boiler The main model for evaluating the risk of high temperature is the model deduced from the surface temperature. This is because all the corrosion reactions do not occur if this is not satisfied. Other parameters such as the concentration of KCl is more diffuse as it is hard to define a general concentration level for which the process will e.g. be accelerated. Thus when evaluating the risk, the base will be the risk from surface temperature. The results in Figure 5.20 and Figure 5.21 therefore represent some of the most important results for this work. By comparing Figure 5.21 and the measurements used for the motivation for the work in Figure 1.1 and Table 1.1 there is a good correlation. The model predicts a high risk area at the tip of the SH3 facing towards the flow, especially in the middle rows. 6.7.1 Corrosion due to concentrations in the flue gas By comparing the risk from surface temperature with the KCl concentration in Figure 5.14 and Figure 5.15 one sees that the areas with high surface temperature risk are also the areas exposed the highest concentrations of KCl. This may not be surprising as the KCl is released during the devolatilisation process, where a lot of heat is released in the following burning of the combustible volatiles. The potential for depositing KCl and thereby Cl2 in the high surface temperature areas are therefore also present. This increases the risk of corrosion substantial. The O2 concentration was found to be very low in the center of the SH3 region. This favours the deposition of KCl onto the cool surface as it is more stable under reducing conditions. Furthermore it reduce the protecting oxide layer on the metal tubes. Thus, the oxygen lean area in the center of SH3 will probably accelerate the corrosion here. 6.7. RISK ASSESSMENT FOR HIGH TEMPERATURE CORROSION IN THE VERDO BOILER 73 6.7.2 Corrosion from a deposition point of view - including shedding The measurements done in reference [17] showed that the highest corrosion was on the tube side facing the flow direction. Comparing the fouling analysis in Figure 5.25, the picture in Figure 6.1(b), the risk from high temperatures in Figure 5.21 and the corrosion measured, the following is worth considering. The deposition of ash particles occur mainly on the front of the tubes facing the flow. As mentioned this ash deposited contained high amounts of the corrosive species Cl, c.f. the discussion above and reference [23]. The deposition of Cl clearly increase the corrosion risk in these areas in this point of view. However as the ash is also very insulating a thick layer will lower the surface temperature. Thereby, it changes the plot in Figure 5.25 to a lower risk rating in these areas. This is especially true for areas where there are a large temperature gradient between the flue gas and steam circuit when the tubes are clean, e.g. in the first half of the steam circuit due to relative cold steam. This means, that either is the plot in Figure 5.21 not as saying as intended, and/or something else is causing or contributing to the high corrosions measured. A theory for corrosion of the front side was therefore developed. In reference [3], Frandsen discusses the shedding of large deposits. When ash deposits grow too large, their own weight can cause large chunks to fall of known as shedding. Sometimes this exposes the bare metal tubes. This adds a whole new dimension to the corrosion problem as it is difficult to estimate what, when and where this will happen. However, the shedding increases the metal surface temperature, which again increases the corrosion rate, [3]. The shedding therefore allows the temperature and presence of Cl to drive a fast corrosion rate in-spite of the insulating layers in some periods. The risk model from high surface temperatures therefore still holds. Although there is a very high metal temperature at the tubes near the steam outlet, the temperature gradient is not as high here as the steam is very hot. Thus near the steam outlet the surface temperature will allmost allways be very high. The deposition analysis and pictures from Verdo near the steam outlet in the middles rows, show very high fouling tendencies. This should give this area a high risk of corrosion as well. However no damages on the tube in form of tube ruptures have been experienced here. No measurements of the corrosion have be conducted either. There is therefore no data to compare this prediction with. If no corrosion occur in this region near the steam outlet, an explanation is needed to distinguish between regions like the ones near the steam outlet and the region near the bottom. A theory could include the shedding. One can imagine that the deposited Cl will eventually diffuse out of the deposit during the described corrosion process over time. If no shedding, and thereby fresh amounts of Cl to the inner layers of the deposit occur, the corrosion will grad- 74 CHAPTER 6. DISCUSSION ually decrease. If the deposit formations occur on the side of the tubes or in between, they will likely grow together, see Figure 6.2. By doing this the attachment area for the total formation will increase. A formation on the side of a tube bank will induce a different kind of load on it self. This will probably also cause less sheddings. The two situations, where one being the deposits suspended below and the other attached to the side, could therefore have a difference in the corrosion process. Measurements of the corrosion on the tubes and ash contents in the region before the steam outlet is necessary to confirm or dismiss this theory. 6.7.3 The missing SO2 simulation and corresponding sulphation of KCl The corrosion risk was according to Flemming Frandsen at DTU, the author of reference [3], very dependent on the ratio of SO2 and KCl. This is because of the sulphation process in reaction (3.1) in section 3.1 on page 11. The ratio is one of the driving forces for the corrosion rate. The SO2 was not included in the simulation as not enough information was found regarding the reaction between SO2 and KCl. The reaction will happen both on the surfaces of the particles and tubes and in the fluid continuum. Chemical reactions between solids and the continuum are also not possible to simulate in STAR-CCM+ directly. Additional deposition models as the ones in reference [9], should be included before having an accurate estimation of the ratio. These would cause a drain of the species in the continuum once deposition occur and thereby changing the concentrations. In order to estimate the corrosion rate from the ratio between SO2 and KCl, considerable additional work would therefore have to be done in describing the chemistry and model development. The missing work in the complex chemistry of the sulphation process and deposition could off course refine the corrosion risk assessment. 6.8 Mitigation of high temperature corrosion Several methods for mitigating high temperature corrosion in biomass incinerators have been attempted in the industry, [19]. The PSO project in reference [19] is a good overview of the most promising methods. Methods such as washing of the fuel to extract Cl, material choice, additives and flow condition optimisation are evaluated from full scale tests conducted. This work will not give a deeper description of the subject, although highly relevant, as it is out of the scope of this work. However, the most promising methods for Verdo based on the analysis conducted, would be flow optimisation and maybe additives if the oxygen lean band can be spread out, and will be briefly outlined. The additive solution is based on adding species such as SO2 to the flue gas in 6.8. MITIGATION OF HIGH TEMPERATURE CORROSION 75 order to release K from KCl in the flue gas and bind it into K2 SO4 . This will, if implemented correctly, remove some of the Cl from the deposits and have it as HCl in the flue gas, where it is far less corrosive in the convective part of the boiler. This could be an interesting perspective to include in the CFD analysis, once the sulphation reactions have been included. However, the oxygen lean column would have to be addressed first for an effective implementation. The flow optimisation solution is more broad and free for interpretation. The main goal of such an optimisation is to have a better mixing of the flue gas before reaching the super heaters, and lowering the velocities in the secondary combustion zone. The mixing insures a better combustion, more even distribution of the critical corrosion species and temperatures. Thus a more even load on the super heaters. By having a better mixing, the high velocities could also be reduced. With lower peak velocities, than the ones seen in the narrow band described, it is possible to have some particle settling. This would reduce the amount of ash convected up to the super heaters and thereby reduce fouling and corrosion elements. The optimisation would therefore not only mitigate the fouling and corrosion but also optimise the combustion and the efficiency of the super heaters. An obvious optimisation parameter is the secondary air jets. The configuration of the jets during this investigated was very unfortunately, as it push the main flow together instead of mixing it. All tough the bed model produced a wrong picture of the lower part of the combustion, it is not expected to influence the negative effect of the secondary jets. An optimisation of these jets is therefore highly recommendable to Verdo. The jets could easily be adjusted in the number of active jets along with the magnitude and possible direction. For large tests and possible alternations of the boiler, a CFD analysis as conducted here could save a lot of resources in the process, as alternations is more easily done in a CFD analysis. 76 CHAPTER 6. DISCUSSION Chapter 7 Conclusion and future work The second boiler at Verdo has been analysed using a CFD analysis conducted in the commercial program STAR-CCM+. Preliminary analyses were conducted prior to a full scale simulation of the boiler with fully integrated steam circuit of SH3. A developed model for predicting high risk areas of corrosion has been applied on the SH3 tubes. The main conclusions from these analyses are presented in the following. The developed corrosion model was only applied on the SH3 in the full scale simulation. 7.1 Wood chip distribution on the bed A CFD analysis of the wood chip spreaders was conducted using spherical Lagrangian particles. The distribution found numerically fitted the main distribution at Verdo. The restitution coefficient, shape and size of the Lagrangian particles was found to have most influence on the results. The evaporation of water from the particles during suspension was found to be neglectable. The element contents of the wood chips found from the fuel analysis could thus be used directly in the bed model. The mass flux distribution of the wood chips on the grate was found and used in the bed model. 7.2 Suspension firing using Lagrangian particles The suspension firing of biomass was simulated using Lagrangian particles and the coal combustion model included in STAR-CCM+. The analysis showed that particularly the size of the particles was of importance for the rate of combustion. The general properties effecting the Lagrangian particles was fund 78 CHAPTER 7. CONCLUSION AND FUTURE WORK to be the same as for the wood chip analysis. The main combustion of the fuel was found in the areas seen by Verdo. The larger particles used for the simulation ended up on the bed where the main combustion proceeded. The coal combustion model seems to predict a reasonable combustion for the used biomass. For final validation, more experimental results are needed. The model was found sufficient for this work. 7.3 Simulating combustion of spreader distributed wood chips on the grate with a bed model Due to a misunderstanding between Force Technology and Verdo, the produced temperature and species profile of the grate by the bed model was reversed regarding the front and back of the grate. The overall technique used to adapt the bed model for predicting the profile was found promising. Further work would have to be done to establish and validate the bed model for use on spreader distributed biomass. The net release of energy and species by the reversed profile should not be different. Due to the configurations of the secondary air jets the overall picture of the boiler, regarding temperature and species concentrations, is not expected to suffer critically from the reversed profile. The error therefore effects the presented results, in especially the lower part of the furnace, but does not discard the work. 7.4 Full scale simulation with integrated steam region for SH3 The main simulation of the boiler, with the steam of SH3 as an integrated circuit, was found using 11.5 million volume cells. The average outlet temperature of the simulated steam was found approximately 4 degrees higher than the measured value at Verdo. The deviation is from a heating of the steam from 448 ◦ C to 490 ◦ C giving over prediction of about 10%. A better mesh resolution of the boundary layer was desired especially for for the coarse ash deposition model, but not done due to available computational power. The overall solution was found to be credible besides from the bed model. The results consisted of the three main categories: general freeboard, the SH3 region with corrosion and fouling. 7.4.1 The general freeboard The results show that the secondary air jets push the main flow into a narrow band in the front to back direction. Near the secondary air jets the band spans 7.4. FULL SCALE SIMULATION WITH INTEGRATED STEAM REGION FOR SH3 79 from side wall to side wall but narrows in until reaching the SH3. In this narrow band higher velocities and concentrations are experienced for all species except for oxygen. A combustion of CO is seen all the way up to the SH3 which is a contributing factor to the narrow band. A large eddy is seen in the top and back part of the furnace. This eddy consist of slow moving recirculating air cooled by the back wall. This lowers the efficiency of this part of the furnace. 7.4.2 The SH3 region with corrosion Two large eddies were found near the sealing of the boiler in between the SH3 tubes. These were suspected to originate from the main flow acting as an impinging jet on the sealing. The eddies pull cooled flue gas into the SH3 area due to the circulation near the sealing and side walls. The impinging jet effect is suspected to be an artefact of the high velocities in the the narrow main flow core. It was found that the main flow continues up through the SH3 as a rather concentrated core with high velocities, temperatures and low oxygen concentration. This cause a very uneven load on the SH3. The outlet temperatures of the middle rows were found in Figure 5.18 to be over 25 ◦ C above the mean outlet temperature for the SH3. This corresponds a load of 1.6 times higher on the middle tubes than the average. A corrosion model was developed. The model was based on metal surface temperature and the concentration levels of KCl, SO2 and O2 . The model needs additional information regarding general critical levels of the species, to be applicable for a general case. The base of the model is thus the surface temperature of the tube metal with a general evaluation of the corrosion species. The model was applied on the SH3 in the second boiler at Verdo. The risk of corrosion due to high metal temperature revealed high risk areas on the front side facing the flow direction at the bottom of SH3. The risk level in these areas corresponded well with the corrosion experienced and measured by Verdo and Force Technology. Furthermore, the model predicted high risk areas of the middle row tubes in the back half of SH3. No measurements were available for comparison in this area. The KCl concentration was found to be relatively even near the tubes in SH3, and thus no particular corrosion risk could be deduced from this. A lean oxygen level was found to be present in the core of the main flow and thus the middle of the SH3. This will generate a favourable environment for KCl condensation on the tubes and thus a higher level of corrosion risk. The reducing environment of the lean oxygen flue gas will cause a reduction of the protecting oxide layer on the metal surface of the tubes. This also increase the risk level in the center of the SH3. The overall corrosion risk was found to be the highest at the middle tubes 80 CHAPTER 7. CONCLUSION AND FUTURE WORK near the bottom and back of SH3. The model fits the measurements at Verdo. Measurements of the back tubes is needed for more full comparison. The model is considered as a good initial model able of predicting the most critical areas with room for further development. 7.4.3 Deposition of coarse ash particles - fouling A model for coarse ash deposition was developed showing good correlations between the extreme fouling areas at SH3 and the numerical predictions by the model. The main fouling areas were found to be the front side of the outer tube facing the main flow at the bottom, and in the middle section of the upper back half in SH3. The level of deposition in these areas was found to be in the order cm of 1 day . No values for the deposition rate at Verdo were available, but the model predictions fits measurements conducted at other biomass power plants. The deposition of coarse ash is found to be related to the corrosion areas. The heavy fouling areas are therefore estimated as a contributing factor for the corrosion process at Verdo. 7.5 Recommendations for Verdo It is recommended that Verdo seeks a better solution to their use of the secondary air jets. The jets push the main flow together instead of mixing it, and optimisation possibilities here are therefore evident. This would lead to a better more complete combustion and even the load on their SH3, and thereby gaining a more efficient daily production and a reduction in the corrosion rates. The preliminary investigations of such optimisations could be done through a CFD analysis as done in this work. The use of additives in form of sulphates could increase the corrosion in the lean oxygen regions. 7.6 Future work The inclusion of the critical levels for the O2 , KCl and SO2 would be next step in developing the corrosion model. A deeper literature study will therefore have to be made on this subject. By doing this, the sulphation process should be included. The core of the coarse ash deposition was found to be the stickiness of both particle and wall, originating from the content of alkali salts. A solution for the reactions between the Lagrangian phases, the fluid continuum and the wall boundaries would therefore have to found as well. 7.6. FUTURE WORK 81 More models for the deposition of KCl from e.g thermophoresis and aerosols would be a natural step when simulating fouling. 82 CHAPTER 7. CONCLUSION AND FUTURE WORK List of Figures 1.1 1.2 2.1 2.2 2.3 3.1 3.2 Sketch over location of known corrosion damages on SH3 at Verdo. The SH3 is located at approximately 14 m directly above the grate. 3 Plot of the measured thickness’s at location A, B, C and D. The design and minimum thickness for the two tube thickness are indicated by the four horizontal lines. Location A,B and C have a design and minimum thickness of 6.3 mm and 5 mm respectively. Location D has a design and minimum thickness of 4.5 mm and 4 mm respectively. It is clear that the largest corrosions are found in the middle rows. . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Schematic figure of the running conditions at the Verdo plant. Adapted from reference [15]. . . . . . . . . . . . . . . . . . . . . . 8 Pictures from inside of the furnace in Verdo at shut down. The pictures was taken at the end of this project, as the SH3 was about to be replaced. a) Picture of a spreader stone used for spreading suspension fired fuel. b) Picture of the spreader for wood chips at the top. At the bottom the spreader for the old coal firing is seen. . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Geometry of the simulated second boiler at the Verdo heat and power plant. Fuel spreaders are located on the front wall in the lower left corner of the figure. SH3 is the fully resolved tube banks at the upper left part of the boiler. The orange blocks illustrates the SH2, SH1, ECO3, ECO2 and ECO1. . . . . . . . . . . . . . . 9 Principle sketch of how high the temperature corrosion processes occur on the metal surface and deposit layer, adopted from [34]. 13 Relative release of K during pyrolysis and combustion. A considerable increase in the release is noticed at T > 700 ◦ C. Adopted from [20]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 84 LIST OF FIGURES 3.3 Relative release of Cl during pyrolysis and combustion, where the experimental running time before sampling are indicated with e.g. 20 min in the figure. Adopted from [20]. . . . . . . . . . . . . . . 16 Relative release of S during pyrolysis and combustion. Adopted from [20]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 3.5 Upper and lower estimate of K release. Adopted from [20]. . . . . 17 3.6 Probability of corrosion due to metal surface temperature alone. 19 3.7 Thermodynamical stable species of potassium under oxidizing conditions (Λ = 1.3), straw fired case. Adopted from [5]. . . . . . 22 Approximated melting curves of potassium and silicate-rich particles. Adopted from reference [10]. . . . . . . . . . . . . . . . . . 23 Melt fraction and propensity of sticking due to T15 criteria. A ratio of 0.1 KCl and 0.9 Silcate particles was used. . . . . . . . . 24 3.10 Melt fraction and propensity of sticking due to T15 criteria. A ratio of 0.15 KCl and 0.85 Si-rich particles was used. . . . . . . . 24 3.11 Probability of sticking due to impact angle, αimpact [deg] . . . . . 26 3.12 Illustration of the grate discretisation. The blue area indicate the area used for the bed model. . . . . . . . . . . . . . . . . . . . . 27 3.4 3.8 3.9 4.1 5.1 5.2 5.3 5.4 Schematic figure of ref and parcels hitting the face but outside the ref . Red dots are cell nodes, green are boundary nodes and black is the impacting particle. . . . . . . . . . . . . . . . . . . . 37 Global mesh used for the main simulation. 11.557.129 volume cells with one prism layer on the furnace walls, and two on the SH3 tubes. The grey cells are wall boundary cells, tan is volume cells in the mid plane and yellow is in-place interface cells towards SH2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 Mesh around SH3 tubes. Two prism layer cells on the SH3 tubes with a stretching of 1.5 and total thickness of 1.5 mm. . . . . . . 43 Residuals for the full scale simulation with the steam in SH3 simulated as an integrated circuit. The large fluctuations in the turbulent kinetic energy, Tke, is caused by injection of Lagrangian particles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 Monitor plot of the average outlet temperature for the steam in SH3, in the the full scale simulation with the steam in SH3 simulated as an integrated circuit. Minor oscillations at are seen. 46 LIST OF FIGURES 5.5 85 Temperature profile of the mid plane. SH3 are located in the top left corner just above the narrow passage. The high temperatures indicates the grate and suspension firing. The temperature just before SH3 is around 1000 ◦ C, which is the melting range for silicate, see Figure 3.8. . . . . . . . . . . . . . . . . . . . . . . . 47 Horizontal temperature profiles at y= 0.1 m, 2.5 m, 4 m, 6 m, 8 m, 10 m, 12 m, 14.5 m, and 18 m above the grate. Notice the effect of the secondary nozzles, in how they push the main flow together. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 Vector plot of the velocities at the mid plane. Due to very high inlet velocities of 43 ms at the carrier jets, the velocity field is limited at 15 ms to more clearly see relevant velocities. A large low velocity eddy is seen at the upper back side of the furnace. . 48 Vector plot of the velocities at the mid plane near SH3. Just after the expansion an eddy is seen, as the flow separates from the wall. The maximum velocity near the SH3 is around 12 ms . . . 49 Vector plot of the velocities across the center of SH3. Two large eddies are seen in the top corners. The highest velocities are seen in the center of SH3. . . . . . . . . . . . . . . . . . . . . . . . . . 50 5.10 O2 distribution at the mid plane. Concentration plotted as mole fraction. Cutoff at 0.1 to highlight oxygen lean band. The suspension firing is clearly seen as the oxygen lean area. . . . . . . . 51 5.11 O2 distribution at the transverse plane used previously in the middle of narrow passage. Concentration plotted as mole fraction, cut off at 0.1 mol to highlight oxygen lean band. . . . . . . 51 5.12 CO distribution at the mid plane. Concentration plotted as mole fraction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 5.13 CO distribution horizontal plane right before outlet at the left. Concentration plotted as mole fraction. Surface average concentration at outlet calculated as 61.4 ppm. . . . . . . . . . . . . . . 53 5.14 KCl distribution at the mid plane. Concentration plotted as mole fraction, cut off at 0.0004 to highlight KCl rich band. . . . . . . . 54 5.15 KCl distribution at previusly used transverse plane in the middle of the narrow passage. Concentration plotted as mole fraction, cut off at 0.0004 mol to highlight KCl rich band. . . . . . . . . . 54 5.6 5.7 5.8 5.9 86 LIST OF FIGURES 5.16 Surface temperature of SH3. Inlet temperature of steam, Tin = 448◦ C. Average outlet temperature of steam, Tout = 492◦ C. Left colorbar represent the scale of the SH3 surface temperature. Colorbar at right represents the temperature of the flue gas in the plane just after the SH3. . . . . . . . . . . . . . . . . . . . . . . . 55 5.17 Surface temperature of OH3. Cutoff at T = 520◦ C in left colorbar representing the scale of the SH3 surface temperature. Colorbar at the right represents the temperature of the flue gas in the plane just after the SH3. It is clear how the uneven temperature field effects the load distribution of SH3 . . . . . . . . . . . . . . . . . 56 5.18 Temperature inside a middle row of tubes. It can be seen how the outer tube has a relative cold outlet temperature compared to the other tubes. An outlet temperature of nearly 515 ◦ C is seen for some tubes. . . . . . . . . . . . . . . . . . . . . . . . . . 57 5.19 Velocity inside a middle row of tubes. Higher velocities are seen near the bending of the tubes. . . . . . . . . . . . . . . . . . . . . 57 5.20 Risk of corrosion due to surface temperature alone. It is seen that the middle rows are the ones most prone to exhibit high temperature corrosion from this plot. In particular the back tubes have an elevated risk. . . . . . . . . . . . . . . . . . . . . . . . . 58 5.21 Detailed view of the risk of corrosion due to surface temperature alone in the lower middle part of the tube bank. The spacing between the rows in the middle, that is just a bit larger than between the other rows, is the middle of the SH3 and boiler. The small high risk areas near the bending indicates an elevated risk of corrosion on the center front side tubes near the tip. . . . . . . 59 5.22 The KCl mole fraction in the boundary cells of the SH3 plotted on the surface of SH3. A relative even distribution when looking at the scale. All though three main areas can be pointed out near the side walls and in the center. . . . . . . . . . . . . . . . . . . . 60 5.23 Comparison of the two models used to predict fouling a) The model developed in this work b) The model used by Kær in [9] . 62 kg 5.24 Deposition of ash particles at SH3. The units are h·m 2 and the kg cm value of 1 equals 0.96 day when using an ash density of 2500 m 3. Cutoff at 1 due to insufficient particle resolvement and to fine mesh, and thereby some unnatural high deposition rates locally. 100.000 parcels was used. . . . . . . . . . . . . . . . . . . . . . . 63 LIST OF FIGURES 87 5.25 Deposition of coarse ash particles at lower right part of SH3. The kg cm units are h·m 2 and the value of 1 equals 0.96 day when using an 6.1 6.2 kg ash density of 2500 m 3 . Cutoff at 1 due to insufficient particle resolvement, and thereby some unnatural high deposition rates locally. 100.000 parcels was used. Heavy depositions are seen at the front of the tubes facing the flow direction. . . . . . . . . . . 64 Pictures from inside of the furnace in Verdo at shut down. a) Little depositions are seen on the side walls. b) Heavy fouling is seen on especially the front side(bottom) of the SH3 tubes. . . . 70 Pictures from inside of the furnace at Verdo. a) Ash deposit at the top and front of the middle part of SH2. b) Ash deposits on the side of row 10 in SH3. . . . . . . . . . . . . . . . . . . . . . . 71 A.1 Illustration of the porous media blocks. . . . . . . . . . . . . . . 106 B.1 Geometry of furnace at the left, and detail drawing of the spreader at the right. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 B.2 Global domain and mesh used for investigations of particle parameters. Domain is cut of by 0.5 m from the side facing the view point in the z-direction(transverse), to shown internal mesh. The poor graphical resolution of the surface mesh is caused by a very fine surface mesh and poor quality file from STAR-CCM+. See Figure B.3 for better resolution of surface mesh . . . . . . . . . . 113 B.3 Detail of the mesh near spreader plate and carrier jet nozzle. A wake refinement was used to resolve the carrier jet. . . . . . . . . 114 B.4 The residence time of particles and velocity magnitude profile. . 115 B.5 Detail plot of continuum and wood chip velocities. Top colorbar: Velocity in continuum. Bottom colorbar: Velocity of particles. . . 115 B.6 Particle distribution with same restitution coefficient, βrest = 0.1. A normal Gaussian distribution of the Dwoodchips with a mean of 0.0284 m, SD= 0.5 and lower and upper bounds of 0.005 m and 0.06 m respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . 116 B.7 Particle distribution with different normal restitution coefficient, βn,rest = 0.2 for the back spreader and βn,rest = 0.05 for the spreader in the front. A constant diameter of Dwoodchips = 0.0284 m was used. The particles residence time is illustrated by color of trajectories. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 88 LIST OF FIGURES B.8 The domain and mesh used to simulate suspension firing and bed model convergence. A total of 983.530 polyhedral cells. The top section of cells before the domain outlet was extruded to avoid reversed flow in the simulation. . . . . . . . . . . . . . . . . . . . 118 B.9 Detailed mesh for a) The combustion area for suspension firing b) Surface mesh of the spreader stone used for spreading the suspension fired fuel. . . . . . . . . . . . . . . . . . . . . . . . . . 119 B.10 Analysis of size of particles for suspension firing. Index 1-3 have the mean diameters: D3,particle = 0.1 mm, D1,particle = 1 mm and D2,particle = 5 mm. . . . . . . . . . . . . . . . . . . . . . . . . . . 120 B.11 Ash fraction of the particles in the lower part of the boiler. a) Boiler seen from the back. b) Boiler seen from the side. It is seen that the small particles burn almost instantly . . . . . . . . . . . 121 B.12 Residence time of particles . . . . . . . . . . . . . . . . . . . . . . 121 B.13 Mole fraction of volatiles in the continuum at the mid-plane of the boiler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 B.14 Iso surface of the mole fraction of volatiles in the continuum seen from the front side. Iso value of 2 · 10−6 . . . . . . . . . . . . . . . 122 B.15 Iso surface of the mole fraction of volatiles in the continuum, seen from the side. Iso value of 2 · 10−6 . . . . . . . . . . . . . . . . . . 123 B.16 Temperature in the lower part of the boiler. a) Boiler seen from the side. b) Boiler seen from the back. . . . . . . . . . . . . . . . 124 B.17 H2 O mass fraction of the wood chips before landing on the bed. 125 B.18 Thermal picture of the grate during production. . . . . . . . . . . 126 B.19 Temperature at the grate calculated by the bed model, seen from the top of the boiler. . . . . . . . . . . . . . . . . . . . . . . . . . 126 B.20 Temperature at the lower part of the boiler. . . . . . . . . . . . . 127 B.21 Residuals for the simulation behind the results in the suspension firing and bed model analyses. The high spikes are the results of Lagrangian particles being injected. . . . . . . . . . . . . . . . . . 128 B.22 Detailed mesh showing the curvature resolvement for a) Refined mesh using the generalized cylinders tool b) Ordinary polyhedral mesh. No surface mesh was generated due to graphical memory problems caused by the many surface cells. . . . . . . . . . . . . 129 B.23 Temperatures in the SH3 tubes, when simulating the tubes separately, a constant heat flux as BC and altered Cp value. . . . . . 131 LIST OF FIGURES 89 B.24 The main monitor values for convergence in the separate steam simulation. a) The average outlet temperature. b) The residuals for the simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . 132 90 LIST OF FIGURES List of Tables 1.1 1.2 2.1 2.2 3.1 3.2 4.1 The design and measured thickness at all 14 tube rows in the four positions indicated at Figure 1.1, adopted from [17]. . . . . . . . 3 Corrosion rates for the two tube thickness’s. The measured thickness is the minimum measured thickness by Verdo and Force. The min. thickness is the minimum thickness estimated by Force Technology - Korrosion og Metallurgi before risk of rupture. The Corrosion rate are the estimated future rates estimated by Force Technology - Korrosion og Metallurgi based on corrosion history and temperature, adopted from [17]. . . . . . . . . . . . . . . . . 4 Proximate analysis of the fuel used at Verdo. 1) Wood chips. 2) Dark bio-pellets. 3) Light bio-pellets (oat peel). 4) Light biopellets 2. 4) Seed pellets. 5) Wood pellets. 6) Miscellaneous biomass dust. See Appendix C.6 on page 144 for analysis and pictures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Ultimate analysis of the fuel used at Verdo on dry basis with the Gross Calorific Value(GCV), and Net Calorific Value(NCV). 1) Wood chips. 2) Dark bio-pellets. 3) Light bio-pellets (oat peel). 4) Light bio-pellets 2. 4) Seed pellets. 5) Wood pellets. 6) Miscellaneous biomass dust. See Appendix C.6 on page 144 for pictures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Pressure for creation of metal chlorides, and the temperatures where the pressure is 10−4 atm (T4 ) and 10−6 (T6 ), adapted from [19]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 The used fractions for evaporation, devolatization, char burnout and release of KCl at the different zones. Zone A is the landing zone of the wood chips. . . . . . . . . . . . . . . . . . . . . . . . 28 Table of used Lagrangian models to simulate the Lagrangian phase for coal combustion . . . . . . . . . . . . . . . . . . . . . . 36 92 LIST OF TABLES A.1 Table over all heat exchangers and their main values. Enthalpies found from [38]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 A.2 Pressure drop values for the heat exchangers and the values used for the calculations. . . . . . . . . . . . . . . . . . . . . . . . . . 108 B.1 Table of the calculated velocities for the secondary air nozzels. The numbering is 1 as the bottom nozzles in the furnace and 4 as the top nozzles. . . . . . . . . . . . . . . . . . . . . . . . . . . 110 C.1 Monitor image of the air used for primary and secondary air inlets136 C.2 Monitor image of temperature and mass flow of cooling water/steam137 C.3 Monitor image of wood chip conveyor belt . . . . . . . . . . . . . 138 C.4 Monitor image of biomass for suspension firing conveyor belt . . 139 Bibliography [1] Wikipedia.org: 11-7-2012, http : //en.wikipedia.org/wiki/Greenhouse _gas#Greenhouse_gases [2] Energistryrelsen (DK) : 11-7-2012, http : //www.ens.dk/da − DK/U nder grundOgF orsyning/V edvarendeEnergi/bioenergi/Biomasse/Sider/F orside.aspx [3] Frandsen, J. Flemming; 2009, Ash Formation, Deposition and Corrosion When Utilizing Straw for Heat and Power Production - Doctoral Thesis, DTU-CHEC [4] Knudsen, N. Jacob; 2004, Volatilization of Inorganinc Matter during Combustion of Annual Biomass - Ph.D. Thesis, DTU-CHEC [5] Nielsen, P. Hanne; 1998, Deposition and High-Temperature Corrosion in Biomass-Fired Boilers - Ph.D. Thesis, DTU-CHEC [6] Mueller, C.; Selenius, M.; Theis, M.: 2005, Deposition behaviour of molten alkali-rich fly ashes - development of a submodel for CFD applications. Elsevier, pp. 2991-2998. [7] Yin, C.;Kær, S. K.; Rosendahl, L.; Hvid, S.L.: 2010, Co-firing straw with coal in a swirl-stabilized dual-feed burner: Modelling and experimental validation, Bioresource Technology, 101, p.4169-4178 [8] Yin, C.; Rosendahl, L.; Kær, S. K.; Clausen, S.: 2008, Mathematical Modelling and Experimental Study of Biomass Combustion in a Thermal 108 MW Grate-fired boiler, Energy & Fuels, 22, p.1380-1390 [9] Kær, S. K.: 2001, Numerical invetigation of deposit formation in straw-fired boilers, Ph.D thesis- Aalborg University [10] Kær, S. K.; Rosendahl, L. A.; Baxter, L. L.: 2006, Towards a CFD-based mechanistic deposit formation model for straw fired boilers. Fuel, pp. 833848 [11] Kær, S. K.: 2004, Numerical moddeling of a straw-fired grate boiler, Fuel, 83, p. 1183-1190 94 BIBLIOGRAPHY [12] Dudek S. A.; Wessel R. A.; Strempek J. R.: 119, Three-Dimensional Numerical Modelling of Stoker-Fired Power Boilers, Presented at the 1999 ASME International Mechanical Engineering Congress and Exposition, Nashville, Tennessee [13] Molcan, P.; Caillat, S.; Le Gleau, F.; Perdrix, E.: 2010, NUMERICAL SIMULATION OF WOOD CHIPS COMBUSTION IN 25 MW SPREADER STOKER BOILER, Third International Symposium on Energy from Biomass and Waste, Venice, Italy [14] Belosevic, S: 2010, Modelling approaches to predict biomass co-firing with pulverized coal, The Open Thermodynamics Journal, 4, 50-70 [15] Kühlert, K.; Nester, S.; Denison, M.: 2001, Development and validation of a Fluent-based CFD model for combined wood waste, sludge and natural gas combustion in a stoker boiler, TAPPI Engineering/Finishing & Converting Conference (Pulp & Paper Industry), San Antonio, Texas, Sept. 16-20, 2001 [16] Zevenhoven-Onderwater, M.; Blomquist, J. P.; Skrifvars, B. J.: 1999, The prediction of behaviour of ashes from five different solid fuels in fluidised bed combustion. Fuel, Volume 79, pp. 1353-1361. [17] Holst, J.N.: 2011, Vurdering af restlevetid på overhederslanger kedel 2, Force Technology, Department of Korrosion og Metallurgi, DK-Brøndby [18] Yin ,C.; Rosendahl, L. A.; Kær, S. K.: 2008, Grate-Firing of biomase for heat and Power production, Progress in Energy and Combustion Science, 34, p. 725-754 [19] Henriksen, N.; Busse, O.; Johnsen, J. B.: Forebyggelse af korrosion og belægningsdannelse, Elsam -PSO projekt nr. 3142, 2003. [20] Johansen, J. M.; Jakobsen, J. G.; Frandsen, F. J.; Glarborg, P.: 2011, Release of K, Cl and S during Pyrolysis and Combustion of High-Chlorine Biomass. Energy & Fuels, pp. 4961-4971 [21] Bjørkman, E.; Strømberg, B.: 1997, Release of Chlorine from Biomass at Pyrolysis and Gasification Conditions. Energy & Fuels, Volume 11, pp. 1026-1032 [22] Hansen, J. V.; Petersen, S. S.; Hove, E. A.: 2011, Katalog over Skadesmekanismer i forbrændingsanlæg, Force Technology, Department of Industrial Processes, DK-Lyngby [23] Westborg, S.; Sørensen, K.G.: 2011, Undersøgelse af belægning på overhederrør, Verdo, Force Technology, Department of Korrosion og Metallurgi, DK-Brøndby BIBLIOGRAPHY 95 [24] Pitman, J.: 2006, Probability, Springer, Third Edition [25] Hansen, J. V.; Petersen, S.S.; Hove, E.A.: 2011, Katalog over Skadesmekanismer i forbrændingsanlæg, Force Technology, Department of Industrial Processes, DK-Lyngby [26] White F. M.: 2006, Viscous fluid flow, McGraw-Hill, 3. edition. [27] Sumer, B. M.: 2007, Lecture notes on turbulence, DTU, MEK [28] Wilcox D. C. : 1994, Turbulence Modelling for CFD, DCW Industries, Inc.,La Canada, California. [29] Bingham, B. B.;Larsen, P. S.; Barker, V. A: 2011, Computational Fluid Dynamics - Lecture Note for Ccurse no. 41319, DTU, Lyngby, Denmark [30] STAR-CCM+ help file, vers. 7.02.011 [31] Larsen, S. Poul; Carlsen, Henrik; Teknisk Termodynamik: DTU-Institut for Mekanisk Energy, 2008 [32] Incropera, F; Dewitt, D; Bergman, T; Lavine, A; 2007, Introduction to Heat Transfer, 5.th Edition, University of California, John Wiley & Sons [33] White, F. M.: Fluid Mechanics, McGraw-Hill, 6. edition [34] Riedl, R; Dahl, J.:1999, Corrosion in fire tube boilers of biomass combustion plants, Proceeding of the China International Corrosion Control Conference ’99, Nr. 90129, China [35] http : //www.simetric.co.uk/si_wood.htm , 25-06-2012 [36] Schwager, T.; Becker, V.; Poschel, T: 2008, Coefficient of tangential restitution for viscoelastic spheres, The European Physical Journal, E. 27, 107-14 [37] Turns, R. Stephen; An introduction to Combustion - Concepts and Applications, 2. edition; McGraw-Hill, 2000 [38] http : //www.peacesof tware.de/einigewerte/wasser_dampf _e.html, may-2012 [39] Zbogar, A; Frandsen, F: 2003, Surface Emissivity of Coal Ashes, IFRF Combustion Journal, Article Nr. 200305 [40] Robinson, A; Buckley, S; Baxter, L: 2001, Experimental Measurements of the Thermal Conductivity of Ash Deposits: Part 1. Measurement Technique, Energy & Fuels, 15, p. 66-74 96 BIBLIOGRAPHY [41] Lu, H; Ip, E.; Scott, J.; Foster, P.; Vickers, M., Baxter, L.L.: 2008, Effects of particle shape and size on devolatilization of biomass particles, Fuel, 89, p. 1156-1168 [42] Rosendahl, L.A.; Yin, C.; Kær, S.K.; Friborg, K.; Overgaard, P: Physical characterization of biomass fuels prepared for suspension firing in utility boilers for CFD modelling, Biomass & Energy, 31, p. 318-325 [43] Syred, N.; Kurniawan, K.; Griffiths, T.; Gralton, T.; Ray .R.: 2007, Development of fragmentation models for solid fuel combustion and gasification as subroutines for inclusion in CFD codes, Fuel, 86, p. 2221-2231 [44] Saripalli, R.; Wang T.: 2005, SIMULATION OF COMBUSTION AND THERMAL FLOW IN AN INDUSTRIAL BOILER, Proceedings of the Twenty-Seventh Industrial Energy Technology Conference, New Orleans, LA Appendix A Models used in STAR-CCM+ The used models are explained in this appendix. For some models certain values was derived. These derivations can also be found in this appendix. A.1 Modelling Lagrangian particles For simulating suspension firing of biomass the STAR-CCM+ coal combustion model for Lagrangian particles was used. The set-up and main models used by STAR-CCM+ will be explained in the following. A.1.1 Momentum balance for particles The momentum balance for material particles are: mp dvp = Fs + Fb dt (A.1) Where mp is the mass and vp is the velocity of the particle, t is time, Fs is the surface force and Fb is the body force. The surface and body forces can be divided into: (A.2) Fs = Fd + Fp + Fvm Fb = Fg + Fu (A.3) where Fd is the drag force from the continuous phase, Fp is the pressure force from gradients in the static pressure of the continuous phase, Fvm is the virtual mass added to the particles as it accelerates the continuous phase, Fg is the gravity force and Fu is a user defined force. The Fvm is neglectable as the particle density is much higher than the continuous phase and there fore not 98 APPENDIX A. MODELS USED IN STAR-CCM+ activated, [30]. Nor is there defined any user defined forces. The drag force is calculated as: 1 (A.4) Fd = CD ρAp |vs |vs 2 Here vs is the particle slip velocity and CD is the drag coefficient calculated from the Schiller-Neumann correlation found in [30] and Ap is the surface area of the particle. A.1.2 Lagrangian Energy Model The energy model for Lagrangian particles is the heat transfer coefficient, hp , for the particle. This is found from the Nusselt number(Nu): hp = Nup · k ) Dp (A.5) Here Dp is the diameter of the particle and k is the thermal conductivity of the continuum phase. The Nusselt number is calculated by the Ranz-Marshall correlation: 1 1 (A.6) Nup = 2(1 + 0.3Rep2 Pr 3 ) Where Re is the Reynolds number and Pr is the Prandtl number of the continum phase, [30]. A.1.3 Turbulent Dispersion The turbulent dispersion introduces the effect of small unresolved turbulent eddies and thus randomness into the Lagrangian particle tracks. This is done through the RANS model and a random-walk technique explained in reference [30]. In the RANS model the fluctuating part u′ cause the random source. A.1.4 Two way coupling In the default one-way coupling only the continuous phase effect the Lagrangian phase through drag and heat transfer. However as the size of the Lagrangian particles are relative big in relation to the continuum volume cells, a two-way coupling between the continuous phase and the Lagrangian particles was chosen. This adds a source in the continuous phase equations. A.1.5 Coal combustion of Lagrangian particles The coal combustion model include the coal moister evaporation, first-order char oxidation and two-step devolatization submodels. A.1. MODELLING LAGRANGIAN PARTICLES 99 The evaporation model The evaporation model assumes that the water of the particle is located as film on the particle surface. Before the release of volatiles and char combustion can take place, all the water on the surface has to be evaporated. This is a reasonable initial model, as the water in the particles probably will keep the temperature of the particle at 100 ◦ C, until the majority of water is evaporated. The STAR-CCM+ formulation is that of the Quasi-Steady Single-Component Droplet Evaporation model as applied to a water droplet, and the associated Sherwood and Nusselt numbers are calculated using the Ranz-Marshall correlation, [30]. Devolatization models A one step or two step devolatization model can be used in order to simulate the release of volatiles from the particles. In eq. (A.7) to (A.9) the general devolatization model for n steps are described according to reference [30]: rawcoalp −→ V Mpn (g) + (1 − V Mpn )(char)p (A.7) Here V Mpn is the volatile matter content of the proximate analysis and (char)p is the char in the particle. The kinetic rate of volatile matter production for the n’th step, rvpn , is expressed as: rvpn = cpn V Mpn γcp (A.8) Here γcp is the mass fraction of coal and cpn is the reaction rate constant defined as: B A Epn cpn = Apn exp − (A.9) RTp Here Apn is a pre-exponential factor, Epn is the activation energy for the particle in the n’th reaction, R is the universal gas constant and Tp is the temperature of the particle. The default choice in STAR-CMM+ is a two step model with given release rates and activation energies for both steps. The two step model is more accurate when knowing the parameters for each step, as it is for the well tested combustion of coal particles. As little or no knowledge is know about these parameters for biomass, the simple one step model is used. The activation energy and pre-exponential release factor are fund in reference [14] J and Apn = 1 · 106 s−1 respectively. as Epn = 7.4 · 107 kg·mol Char oxidation The char is the residue after the devolatization, which is generalised to pure carbon. The used oxidation processes in this work is done by O2 , but could also 100 APPENDIX A. MODELS USED IN STAR-CCM+ have been by H2 O and CO2 . The oxidation reaction is: 2 C + O2 −→ 2 CO (A.10) The reaction rate is based on the diffusion of O2 to the particle and a mass balance of the V M produced from devolatization, and can be found in reference [30]. The reaction rate constant, kc , for the specific oxidation reaction is: A Ec kc = Ac Tp exp − RTp B (A.11) Here Ec is the activation energy for the char and Ac is a pre-exponential factor. Mass distribution of the volatile species The four main components in the coal model, rawcoal, char, ash and water are defined in the following way. The rawcoal consist of volatiles and char, where volatiles are the volatile species such as CO, CO2 , H2 and CH4 and the char is the fixed carbon in the particle. Thus the raw coal is the dry, ash free part of the biomass. Ash is an inert component and only contributes with mass in this study, while water evaporates from the particle. Derivation of mass distribution for volatiles To find the mass distribution of the volatile species, the fractions of VM, C, ash and H2 O in the proximate analysis in Table 2.1 must be used. These fractions must be combined with the ultimate analysis with the atom constitution. The volatile species only contain the atoms C, O, H, K and Cl in this study. To fulfil mass conservation the molar mass, M, of the atoms and the molar mass of the volatile species is used to set up a set of linear equations, [13]: 12 12 12 · XCH4 + · XCO + · XO2 = YC,vol 16 28 44 (A.12) 16 32 · XCO + · XCO2 = YO,vol 28 44 (A.13) 4 · XCH4 + XH2 = YH,vol 16 (A.14) XKCl = YKCl,vol (A.15) Here Yi,vol is the mass fraction of C, O, H and KCl of the volatiles, and Xi is the unknown mass fraction of the volatile species CH4 , CO, CO2 , H2 and KCl. A.1. MODELLING LAGRANGIAN PARTICLES 101 The coefficients are the ratio of the molar mass of the atoms and the molar mass of the volatile species. E.g. in eq. (A.12) the first coefficient is the ratio M of M C = 12 16 , here rounded of for simplicity in the text. The mass fraction CH4 Yi,vol can be found by dividing the mass fractions from the dry ultimate analysis, Yi,ult , with the mass fraction of the volatiles, volf rac , see eq. (A.16) to (A.20). volf rac = YC,tot − YC,f ixed + YO + YH + YKCl = 0.8796 (A.16) YC,ult − YC,f ixed = 0.4625 volf rac (A.17) YO,vol = YO,ult = 0.4681 volf rac (A.18) YH,vol = YH,ult = 0.0688 volf rac (A.19) YKCl,ult = 0.0007 volf rac (A.20) YC,vol = YKCl,vol = In order to have mass conservation, the sum of the mass distribution after the volatization should be the same as before volatization minus the mass fraction q Yi,vol = 1, see of the fixed carbon, YC,f ixed = 0.07 and ash, Yash = 0.0381, i.e. eq. (A.21). XCl2 + XCO + XCO2 + XH2 + XCH4 = YC,vol + YO,vol + YH,vol + YKCl,vol = (A.21) 0.4625 + 0.4681 + 0.0688 + 0.0007 = 1 Further more the ratio between CO2 and CO is temperature dependant, and is found using the same equation from the bed-model 3 −6400 K Ratio = 2510 · exp T 4 3 −6400 K = 2510 exp 1373 K 4 = 26.66 (A.22) Here it is assumed that the volatiles are released during suspension firing, where an average temperature off 1100◦ C is estimated. This gives an additional equation: xCO = 26.66 (A.23) xCO2 Equation A.12 to A.23 can be solved using a mathematical program e.g. Maple, where equation A.12, A.13, A.14, A.21 and A.23 are solved as a linear set of equations yielding the unknowns: XCH4 = 0.1594, XCO = 0.7817, XCO2 = 0.0293, XH2 = 0.0290 and XKCl = 0.0007 102 APPENDIX A. MODELS USED IN STAR-CCM+ Derivation of heat of formation for raw coal In order to calculate the energy release from the biomass correctly, STARCCM+needs the heat of formation for the rawcoal in [ kJ kg ], which in this case is the heat of formation of the dry, ash free biomass. Using eq. (2.35) from reference [37] the amount of heat removed from a system can be calculated as hR = hprod − hreac (A.24) Here hprod is the enthalpy of the volatile products CO, CO2 , H2 , H2 O and CH4 , hreac is the heat of formation for the reactants and thereby the raw coal heat of formation,hf ormation , and hR is the enthalpy of the reaction. The heat of combustion, hc , also known as the Gross Calorific Value for the fuel, is the same as the enthalpy of reaction, hR , but with opposite sign, see Table 2.1 for value. Eq. (A.24) can be rewritten to find hf ormation hR = −hc = hprod − hreac (A.25) hc = hreac − hprod (A.26) hf ormation = hc + hprod = GCV + hprod (A.27) As mentioned before the hprod is the enthalpy of the combustion products. The combustion products is better defined as the end products after all combustion reactions. Here it is assumed that water is evaporated so the raw coal devolatile to volatile gasses and char: Rawcoal −→ V olatiles + char (A.28) The volatiles was described earlier as volatiles −→ CO + CO2 + H2 + CH4 (A.29) CO, H2 and CH4 will react with O2 and burn as gasses: 1 CH4 + O2 −→ CO + 2H2 2 (A.30) 1 CO + O2 −→ CO2 2 (A.31) 1 H2 + O2 −→ H2 O 2 (A.32) From eq. (A.30) to (A.32) it can be seen that the end combustion products is H2 O and CO2 for a complete combustion, leaving H2 O and CO2 as hprod A.1. MODELLING LAGRANGIAN PARTICLES è 103 é kJ in eq. (A.24). The units of eq. (A.27) is in kmol , all though it can also be written è éon a mass basis. The units of the heat of formation in STAR-CCM+ was kJ kg , and the above equations are only general combustion equations with no considerations of the coefficients of the species in eq. (A.29). In order to find the heat of formation of the final combustion products, we therefore need to use the coefficients of the volatile gasses for combustion of one kg of bio-mass first. The distribution of the species is then converted to a /mol basis in order to use the enthalpies of the products to find the heat of formation. Finally the heat of formation can be converted back to a basis of /kg for the STAR-CCM+ input. Using the coefficients for the volatile species found in section Appendix A.1.5 on page 100 eq. (A.29) can be written 1 kgvolatile −→ 0.782 kgCO + 0.029 kgCO2 + 0.029 kgH2 + 0.159 kgCH4 (A.33) To convert eq. (A.33) to a mol basis the molar masses of the species are kg kg kg used, MCO2 = 44.008 kmol MCO = 29.009 kmol , MCH4 = 16.042 kmol , MH2 = kg 2.016 kmol : 1 kgvol = A 0.782 29.009 kg kg kmol B CO A + 0.029 44.008 kg kg kmol B A + CO2 0.029 2.016 0.027 kmolCO + 6.66 · 10−4 kmolCO2 + 0.015 kmolH2 kg kg kmol B A + 0.159 kg kg kmol 16.042 (A.34) + 0.010 kmolCH4 H2 Plugging the coefficients from eq. (A.34) into eq. (A.30) yields: 1 0.010 CH4 + 0.010 · O2 −→ 0.010 CO + 0.020 H2 2 (A.35) The CO produced in eq. (A.35) must be included in eq. (A.31) 1 (0.010 + 0.027) CO + (0.010 + 0.027) O2 ≫ 0.037 CO2 2 (A.36) Likewise must the H2 produced in eq. (A.35) be included in eq. (A.32) (0.020 + 0.015) H2 + (0.020 + 0.015) 1 O2 −→ 0.035 H2 O 2 (A.37) Thus the complete combustion of one kg volatiles produce: CO2 = 0.037 kmol + 6.66 · 10−4 kmol = 0.038 kmol (A.38) H2 O = 0.035 kmol + 0.015 kmol = 0.050 kmol (A.39) MJ kmol From reference [37], the heat of formation for CO2 = −395 and H2 O = MJ at 1400 K is found. This yields a hprod for one kg of volatiles: −250 kmol 3 hprod = 0.038 kmolCO2 · −395 MJ kmol 4 3 +0.049 kmolH2 O · −250 MJ kmol 4 = −27.2 MJ kgvol (A.40) B CH4 = 104 APPENDIX A. MODELS USED IN STAR-CCM+ As the preceding calculations was done for one kg of volatiles and the volatile fraction of one kg of rawcoal was found in eq. (A.16) as 0.8796, hprod for one kg of rawcoal is: hprod = −27.2 MJ kgvol · 0.8796 kgvol kgrawcoal = −23.9 MJ (A.41) kgrawcoal Using eq.(A.27) the heat of formation for STAR-CCM+ can be found: hf ormation = GCV + hprod = 19.5 A.1.6 MJ kg − 23.9 MJ kgrawcoal = −4.3 MJ kg (A.42) Particle Radiation The particle radiation model is a part of the radiation model for the continuous phase. There is a two way coupling of the radiation between the continuum and particles. The particle generates a source term in the continuum energy equation as explained in the next section. As the particle is able to store heat the term is also found in energy equation for the particle. A.2 Modelling radiation Different radiation models are available in STAR-CCM+ depending on the required physics. For a continuum with a media transparent to radiation but with boundaries having radiation properties the surface to surface, S2S, model can be used. However as this work models a real gas with combustion, the media participation of the gray gasses regarding absorbing, emitting and scattering can not be neglected. Thus the more computational heavy model Participating Media Radiation, DOM, is used. The model uses the Discrete Ordinate Method, hence the abbreviation DOM. The gray gasses in this work are H2 O and CO2 . The model also include particle radiation from Lagrangian particles when activated here, meaning that radiation has to be selected in the Lagrangian models. In the STAR-CCM+ help guide, reference [30], the basic model equation for describing a beam of radiation and how it looses energy due to absorption, gain energy by emission and redistributes energy by scattering is: ksλ dIλ = −βλ Iλ + kaλ Ibλ + ds 4π Ú 4π Iλ (Ω)d(Ω) + kpaλ Ipbλ + kpsλ 4π Ú 4π Iλ (Ω)d(Ω) (A.43) Here λ is the wavelength, Iλ is the radiant intensity at wavelength λ, Ibλ is the black body intensity, Ω is the solid angle, βλ is an extinction coefficient, kaλ is the absorption coefficient at wave length λ, ksλ is the scattering coefficient at wavelength λ, kpaλ is the particle absorption coefficient at wavelength λ, kpsλ is A.3. SIMULATING HEAT EXCHANGERS WITH POROUS MEDIA REGIONS 105 the scattering coefficient at wavelength λ and s is the distance in Ω direction. The extinction coefficient is defined as: βλ = kaλ + ksλ + kpaλ + kpsλ (A.44) The absorption, scattering and emissivity coefficients and the solid angle are parameters available for the user to define. The solid angle Ω is a three-dimensional measure of a sphere with Si-units steradians [sr]. A full sphere corresponds to 4π sr, [30]. In the DOM model in STAR-CCM+ the the number of ordinates must be defined, where 4 ordinates divide the sphere seen by a cell into 4 solid angles. Thus the higher the number of ordinates the more accurate a solution. The default value of 4 ordinates have been used in this work. The emissivity, ǫ, is used in the radiant intensity Iλ for a gray body as: I = ǫσT4 (A.45) where σ is the Stefan-Boltzmann constant. In [39] several experiments on coal ash emissivities are compared. It shows that the emissivity are temperature dependant, but can be regarded as constant for small temperature intervals as found in the SH3 region. No sure value could be found however, as the different experiments compared, showed a quite large spread in the found values, e.g. 0.38-0.65 at 1000K. A rough mean of the values, 0.55, is adopted for the emissivity used in this work at the SH3 surfaces. A.3 Simulating heat exchangers with porous media regions The SH3 was fully resolved. In order to save computational cells in the less important section of the boiler, superheater 1 and 2(SH1-2) and the economisers(ECO), are simulated with blocks of porous media. These induce a pressure and heat loss. See Figure A.1 for location of the heat exchangers. 106 APPENDIX A. MODELS USED IN STAR-CCM+ Figure A.1: Illustration of the porous media blocks. The pressure drop and heat flux must be calculated analytical and given as input to STAR-CCM+, this is done in the following. A.3.1 Energy extracted from the heat exchangers From Appendix C.2 on page 137 the temperature and mass flow values of the superheated steam can be found. The effect can be calculated as, [31]: Q̇ = ṁ · Cp · (Tsteam,out − Tsteam,in ) = ṁ · (hsteam,out − hsteam,in ) (A.46) Based on the running conditions of the plant the effect of each heat exchanger is calculated, and listed in Table A.1: A.3. SIMULATING HEAT EXCHANGERS WITH POROUS MEDIA REGIONS 107 Table A.1: Table over all heat exchangers and their main values. Enthalpies found from [38]. ECO 1 sec. 1 ECO 1 sec. 2 ECO 2 Walls SH1 SH2 SH3 ṁ [Kg/s] Pressure [bar] Tsteam,in [C] Tsteam,out [C] hin [kJ/kg] hout [kj/kg] Q̇ [MW] 29 29 29 29 29 29 29 155 145 135 120 120 115 109 159 200 242 316 325 379 448 200 242 316 325 382 448 490 680.13 857.89 1048.21 1434.95 2688.40 3000.55 3222.18 857.89 1048.21 1434.95 2688.4 3000.55 3222.18 3337.38 -5.16 -5.52 -11.22 -36.35 -9.05 -6.43 -3.34 T otoutput -40.72 from Heat ex. T otoutput from boiler The calculations can be checked with the output from the plant as a total. The total output for the plant is found in Appendix C.2 on page 137 to Q̇tot = Ẇelectrical + Q̇thermal = 32.5 MW + 106 MW = 138.6 MW. As the values in Table A.1 are for one boiler the total potential energy for Verdo would be 154.14 MW. An energy loss of 15.54 MW or 10% are there fore seen for the entire system which seems reasonable. A.3.2 Pressure drop over the heat exchangers The pressure drop over a heat exchanger can be estimated with analytical formulas, depending on the number of rows, spacing, size of tubes, flow velocity, viscosity and temperature. To estimate the pressure loss a calculating sheet developed by Force Technology was used. This sheet uses the formulas of (VDIWärmeatlas 7. auflage 1994), and is not explained further. The geometric values was found from the drawings of the heat exchangers. The results are presented in Table A.2. -77.07 108 APPENDIX A. MODELS USED IN STAR-CCM+ Table A.2: Pressure drop values for the heat exchangers and the values used for the calculations. ECO 1 sec. 1 ECO 1 sec. 2 ECO 2 SH1 SH2 U [m/s] 5 5 5 5 5 N 16 14 16 17 24 dlong [mm] 90 90 90 130 46 dtrans [mm] 200 200 200 200 215 ∆P [P a/m] 28 24.42 22 17.17 9.93 ∆P [kg/m4 ] 1.12 0.98 0.88 0.69 0.4 Appendix B Preliminary analyses The preliminary analyses conducted prior to the main simulation of this thesis can be found in this appendix. B.1 Primary and secondary combustion air The methods for estimating the inlet values for the primary and secondary air will be elaborated in the following. B.1.1 Primary air Very little exact information was available about the distribution of the primary air for the grate. The duct system is a main duct separating into two ducts, one for each side of the furnace. At each side, the ducts branches off into additional four ducts leading in and under 4 zones of the grate. The controls for adjusting the airflow was a simple mechanical leaver placed at each of the four branches at each side. The settings for these leavers were: 1/4, 1/2, 1 and 1, starting from the front of the furnace. Verdo provided a crude but simple method for 0.5 1 1 distributing the flow through each zone as: 0.25 2.75 , 2.75 , 2.75 and 2.75 for zone 1, 2, 3 and 4 respectively. The fractions were applied on the mass flow of the primary air and used in the bed model. B.1.2 Secondary air The inlet values for the secondary air nozzles were not available, and therefore had to be estimated. From reference [33] the pressure drop in a pipe can be 110 APPENDIX B. PRELIMINARY ANALYSES written as: 1 ∆p = Ktot ρU 2 2 (B.1) where Ktot is the total pressure loss coefficient due to minor and major losses, ρ is the density of the fluid and U is the velocity of the fluid. Due to lack of information about the pipe system, the pressure loss coefficient was difficult to estimate and a practical approach was used. All the air ducts have the same pressure drop as the separate ducts originate from the main feeding duct of secondary air and terminates in the boiler, see Appendix C.1 on page 136. The flow will automatically seek to satisfy this, when e.g. turning a valve i one pipe(increasing K). This decrease the velocity in the corrected pipe and increase the velocity in the other pipes to gain the same pressure drop. However the pressure loss coefficient is related to the friction coefficient, which is a function of the Reynolds number in the duct. Thus we seek the velocity which is a function of it self. An iteration program was written to solve this, see Appendix D.1 on page 154. It uses the pressure loss coefficients for an pipe exit, the valves and wall friction and solves eq. (B.1). As the pressure drop and mass flow is given the pipe flows are solve separately. It is assumed that the main factor for distributing the flow is the valves. Due to the lack of information of the rest of the pressure loss coefficients in each pipe the calculated velocity is normalised to produce the required mass flow in the end. The code produce the results in Table B.1: Table B.1: Table of the calculated velocities for the secondary air nozzels. The numbering is 1 as the bottom nozzles in the furnace and 4 as the top nozzles. Secondary nozzle string U [m s] B.2 Top-Front 20.00 1. Back 33.06 2. Back 35.48 3. Back 35.63 4.Back 35.85 Distibution of wood chips on the grate The wood chips are spread out on the bed by use of a spreader, see Figure B.1. B.2. DISTIBUTION OF WOOD CHIPS ON THE GRATE (a) Bottom of furnace 111 (b) Geometry of spreader plate and air nozzles Figure B.1: Geometry of furnace at the left, and detail drawing of the spreader at the right. The wood chips fall down the feeding duct gaining momentum. At the exit of the feeding ducts a carrier air jet and a small ramp is located in order to control the direction and spreading of the chips, see Figure B.1(b). Most of the main geometry was available but important information about the air jets such as the nozzle geometry and number of nozzles were not. As the nozzles contribute with a substantial amount of air to the free board, the penetration depth of the jet into the free board is very important. A simulation of the spreader part was therefore conducted in order to estimate the momentum and penetration of the carrier jet. Beside the momentum information of the jet, the distribution of the wood chips on the grate can be found from this analysis. The distribution is used to validate the momentum of the jet, as a correct jet will give the correct chip distribution on the grate. Most of the overall geometry of the spreader was available, including angle of ramp plate, αramp = 5◦ , and geometry of duct carrying the wood chips. The mass flux of wood chips for full load on the plant was given, ṁwood = 2.38175 kg s , see Appendix C.3 on page 138. Beside the information found in Appendix C.1 on page 136 and Appendix C.3 on page 138, the mass flux of the air jet ṁ(air,spreader) = 3.66 kg s , and fall height of wood chips in duct before reaching spreader, Hf all = 3 m was provided by Verdo. The unknown parameters were the exact particles size and distribution on the grate, the exact density of the wood, the restitution coefficient of wood chips bouncing of the duct and furnace walls, and geometry and numbers of air nozzles. The simulation was made before the size distribution analysis was available, which made the size an unknown. As the number and geometry of the air nozzles were not know, the nozzles was approximated with a rectangle covering the whole injection width. It is assumed that a series of round air nozzles placed in a row with close proximity will form one coherent jet after a small distance from the nozzle outlets, similar to that of the rectangular outlet. Having the width of the injection part, the 112 APPENDIX B. PRELIMINARY ANALYSES only free parameter is the height of the rectangle air nozzle when tuning the outlet velocity. A first guess of the height of the nozzle was 25 mm, from the illustration profile drawing of the spreader, Figure B.1(b). An average density of kg wood chips is found from reference [35] to approximately 500 m 3 . The size of the wood chips was according to Verdo approximately Lwoodchip = 3 − 7 cm, with a rectangular shape. As the only possible shape of particles in STAR-CCM+is spherical, the used diameter is calculated from the average rectangular chips conserving the volume, the effect of this is discussed in section 6.1 on page 65. E.g a wood chip with dimension Vchip = 1 · 2 · 6 cm = 12 cm3 yields a particle diameter of Dwoodchips = 2 · r = 2 · ó 3 3Vchip =2· 4π ó 3 3 · 12 · 10−6 m3 = 0.0284 m 4·π (B.2) An alternative method could have been using the Discrete Element Model (DEM) application in STAR-CCM+as this uses discrete solid particles interacting with each other. This application allows the construction of spherical particles joined together giving non-spherical shapes. Due to increasing complexity and little extra information, this was not chosen however. When using Lagrangian particles in STAR-CCM+ one needs to define and consider the particles interaction with boundaries. This is done by defining the normal and tangential restitution coefficient for the particles when colliding with walls. The normal restitution coefficient, β(n,rest) , is the ratio between the height of a particle released with zero velocity and the height it rebounds after a collision with a wall perpendicular to the particle trajectory. β(n,rest) = Hrelease Hrebound (B.3) Thus a particle with a coefficient of 1 is perfectly elastic. The tangential restitution coefficient, β(t,rest) , defines how much of the velocity is lost in the perpendicular direction of the wall. Thus it can be seen as the friction between particle and wall. In some work according to reference [36], the Columb friction coefficient is used for the tangential restitution coefficient. These parameters are essential when simulating particles interaction with the boundaries. No information of the restitution coefficient for wood chips was found in the literature. Thus, a small physical test was conducted with small irregular wood pieces released from 1 m, showing an approximate normal restitution coefficient of 0.1. The simulation was only conducted for one half of the lower part of the furnace with two spreaders. In this way, effects of e.g. particle size are easier illustrated, as particle with different properties can be injected from separate spreaders. The aim was that the momentum of the air jet should correspond to a reasonable distribution of the wood chips on the grate for minor, average and larger sized particles. In Figure B.2 the domain and mesh used for the analysis are shown. B.2. DISTIBUTION OF WOOD CHIPS ON THE GRATE 113 Figure B.2: Global domain and mesh used for investigations of particle parameters. Domain is cut of by 0.5 m from the side facing the view point in the z-direction(transverse), to shown internal mesh. The poor graphical resolution of the surface mesh is caused by a very fine surface mesh and poor quality file from STAR-CCM+. See Figure B.3 for better resolution of surface mesh The wall boundary condition was used on both sides of the section. For a better resolution of the surface mesh see Figure B.3, which shows the mesh used to resolve the spreader and carrier jet. 114 APPENDIX B. PRELIMINARY ANALYSES Figure B.3: Detail of the mesh near spreader plate and carrier jet nozzle. A wake refinement was used to resolve the carrier jet. A wake refinement tool was used to extrude the fine mesh from the nozzle. At Figure B.4 and Figure B.5 the velocities and residence time of the particles are shown. The inlet velocity of the carrier jet was found to vjet = 43.64 ms from previous mentioned geometry and mass flow. It is clear that the particles are accelerated by the carrier jet as the velocities are increased over the short distance of the spreader plate from vp ≈ 5 ms to vp ≈ 7 ms , where vp is the velocity of the particles. The deflection of the jet in the vertical direction is caused by the air from the bed. For this analysis the total mass flux of air from Appendix C.1 on page 136 where spread evenly over the entire bed, as no bed analysis had been conducted yet. The particles are suspended in the combustion zone for approximately 0.7-1 seconds. B.2. DISTIBUTION OF WOOD CHIPS ON THE GRATE 115 Figure B.4: The residence time of particles and velocity magnitude profile. Figure B.5: Detail plot of continuum and wood chip velocities. Top colorbar: Velocity in continuum. Bottom colorbar: Velocity of particles. 116 APPENDIX B. PRELIMINARY ANALYSES At Figure B.6 the distribution of particles are shown for a normal distribution of the particles with a mean diameter of Dwoodchip = 0.0284 m, standard deviation SD= 0.5 and lower and upper bound of 0.005 m and 0.06 m respectively. The high SD generates a near uniform distribution over this interval. It shows that the smaller wood chips travel longer out on the grate than the bigger chips. The carrier jet cannot accelerate the larger particles as fast as the small particles because of the higher inertia of the large particles. In Figure B.6 the particles have the same normal restitution coefficient, β(n,rest) = 0.1. Figure B.6: Particle distribution with same restitution coefficient, βrest = 0.1. A normal Gaussian distribution of the Dwoodchips with a mean of 0.0284 m, SD= 0.5 and lower and upper bounds of 0.005 m and 0.06 m respectively. In Figure B.7 the normal restitution coefficient is analysed. Here a value of βn,rest = 0.2 was used for the particles injected in the spreader in the back, and βn,rest = 0.05 was used for the particle in the front spreader. A value of βt,rest = 1 was used. It shows as expected that the particles with highest coefficient travels the longest, as these do not loose as much energy when colliding with the duct walls. B.2. DISTIBUTION OF WOOD CHIPS ON THE GRATE 117 Figure B.7: Particle distribution with different normal restitution coefficient, βn,rest = 0.2 for the back spreader and βn,rest = 0.05 for the spreader in the front. A constant diameter of Dwoodchips = 0.0284 m was used. The particles residence time is illustrated by color of trajectories. An analysis was conducted with other values of βt,rest besides 1. These showed that for values other than 1, the particles basically stopped/froze on the walls in the duct leading to the spreader and carrier air nozzles. Even for high values as 0.99. As the above analysis with βt,rest = 1 showed good results, it was concluded to use the value of 1. The value of 0.15 was adopted for the normal restitution coefficient for the rest of the work. Further more it was concluded that the momentum of the carrier air was accurate enough, as it produced good results for the distribution of particles on the grate. According to Verdo roughly 90% of the chips land on the back 2/3 of the grate, but no certain measurements had never been recorded. The analysis results match this result, all though maybe overshooting with the main wood chip mass flux on the back half of the bed. The final distribution analysis of the wood chips is conducted in Appendix B.4 on page 125, where the bed model and suspension firing are also investigated. 118 B.3 APPENDIX B. PRELIMINARY ANALYSES Simulating suspension firing The suspension firing was simulated together with the bed model. The results in Appendix B.4 on page 125 were therefore produced along with the results presented in this section. In Figure B.8 the domain and mesh used for this analysis are presented. A total of 983.530 polyhedral cells was used. The mesh was refined by volumetric control volumes in the area where the combustion process of the suspension firing was expected to take place. This area is visible as the ordered cells with uniform cell sizes in Figure B.9(a). Figure B.8: The domain and mesh used to simulate suspension firing and bed model convergence. A total of 983.530 polyhedral cells. The top section of cells before the domain outlet was extruded to avoid reversed flow in the simulation. B.3. SIMULATING SUSPENSION FIRING (a) Refined mesh in the combustion zone of the suspension firing. 119 (b) Surface mesh of the spreading stone. Figure B.9: Detailed mesh for a) The combustion area for suspension firing b) Surface mesh of the spreader stone used for spreading the suspension fired fuel. In Figure B.9(b) the spreader stone used to spread the biomass fuel are illustrated by the surface mesh. In Figure B.10 a plot of the trajectories for three particle sizes are shown. The red trajectories are particles with a diameter of 0.1 mm, the blue trajectories are particles with a diameter of 1 mm and the green trajectories are particles with a diameter of 5 mm. The different diameters are controlled by three different injectors, hence the title of the colorbar. The trajectories show, that the small particles are lifted up by an up going jet right after injection into the furnace. The medium sized particles are partially lifted up in the middle of the furnace and partially injected all the way to the back wall. Here they fall down onto the bed. All the large particles end up on the bed. 120 APPENDIX B. PRELIMINARY ANALYSES Figure B.10: Analysis of size of particles for suspension firing. Index 1-3 have the mean diameters: D3,particle = 0.1 mm, D1,particle = 1 mm and D2,particle = 5 mm. In Figure B.11 the particle trajectories are coloured with the fraction of ash. Here red indicates a complete burnout of the fuel leaving only ash left in the particles. Comparing Figure B.11 with Figure B.10 it is clear that the small particles burn all most instantly while the large particles will burn on the bed. It it also clear that the maximum residence time in the Lagrangian particle solver is set to low for the large particles to burn out. In Figure B.12 the residence time are plotted for the same trajectories as in Figure B.10 and Figure B.11. The small particles leave the domain in approximately 2 seconds. In relation to Figure B.11 the small particles are completely burned out in less than one second. The results in Figure B.10 to Figure B.12 clearly show the significance of the particle size. Thus for a correct simulation of the suspension firing the particle distribution has to be correct. B.3. SIMULATING SUSPENSION FIRING (a) Ash fraction in particles seen from the front of the boiler 121 (b) Ash fraction of particles seen from the side of the boiler Figure B.11: Ash fraction of the particles in the lower part of the boiler. a) Boiler seen from the back. b) Boiler seen from the side. It is seen that the small particles burn almost instantly Figure B.12: Residence time of particles In Figure B.13 the mole fraction of the released coal volatiles from the suspension firing are shown for the mid plane of the furnace. It shows two areas with relatively high concentrations. The area just after the entrance into the furnace corresponds to the volatilization of the small particles. The area in the middle of the furnace corresponds to where the medium sized particles devolatize. The reaction coefficients in the process from the STAR-CCM+ coal volatile specie to the actual volatile species found in Appendix A.1.5 on page 102 eq. (A.29) are set to 1 · 106 s−1 and an activation energy of zero. This was done in order to get 122 APPENDIX B. PRELIMINARY ANALYSES a fast, none heat demanding reaction as the reaction is not an actual chemical reaction, see eq. (A.29) in Appendix A.1.5 on page 102. Figure B.13: Mole fraction of volatiles in the continuum at the mid-plane of the boiler In Figure B.14 and Figure B.15 an iso-surface is shown for the coal volatiles. Plotting the volatiles in this way shows the volume where the main combustion of the suspension fired fuel will take place. The main combustion is seen to take place approximately 2/3 out from the front wall of injection. This correspond to the approximately flame length that Verdo sees with a live camera in the furnace. One can also see the volatilization of the medium sized particles on the back side of the bed in Figure B.15. Figure B.14: Iso surface of the mole fraction of volatiles in the continuum seen from the front side. Iso value of 2 · 10−6 . B.3. SIMULATING SUSPENSION FIRING 123 Figure B.15: Iso surface of the mole fraction of volatiles in the continuum, seen from the side. Iso value of 2 · 10−6 . In Figure B.16 temperature profiles for the mid plane and a cross sectional plane are shown. As expected from a steady state simulation on a symmetric geometry, symmetry is found. Some small none symmetric properties of the geometry is present in the lowest string of secondary air jets on the back wall. The effect of these in this simulation is apparently insignificant. In Figure B.16(a) the top(and only active) jet on the front wall and the second jet from the top on the back wall are visible. The other jets are not visible as the section does not cut through their center. It is clear that the secondary jets push the main column of hot air together to a relative narrow band. In Figure B.16(b) the carrier jet for the wood chips are visible as the four light blue areas. Also one may notice the high temperature areas surrounding these jets. This is due to the combustion reaction between the rich CO flow from the bed and the rich oxygen flow from the jets. 124 APPENDIX B. PRELIMINARY ANALYSES (a) Temperature at the mid-plane seen from the side (b) Temperature at a plane 0.5 m from the front wall. Plot seen from the front side. Figure B.16: Temperature in the lower part of the boiler. a) Boiler seen from the side. b) Boiler seen from the back. B.3.1 Summary on suspension firing An analysis of the Lagrangian particles simulating the biomass fuel was conducted showing the effect of particles sizes. Small particles with size of 1 mm or less burns almost instantly while the majority of particles of 1 cm falls to the bed and burn there. Further more the used models predicts the flame generated by the combustion accordingly to the description by Verdo. The secondary jets push the main column of flue into a narrow band. The coal combustion model was found sufficient for this work. B.4. TUNING IN THE BED MODEL B.4 125 Tuning in the bed model The bed model was calibrated along with the analysis of the suspension firing. First the final mass flux of wood chips was analysed. In Figure B.17 the distribution and water content of the wood chips are shown along with temperature profiles described in the previous section. It shows that almost no evaporation of the water in the wood chips occurs. Thus one can assume that the fuel input for the bed model is the same as found in the proximate fuel analysis. The mass flux of the wood chips onto the different bed zones was integrated and used as input for zone A-D described in section 3.6 on page 27. Figure B.17: H2 O mass fraction of the wood chips before landing on the bed. The bed model used, depends on the incoming thermal radiation to calculate the temperature and combustion process occurring on the bed. This is found using a surface integral of the boundary irradiation in STAR-CCM+. As the radiation in the boiler depends on the temperature, the simulation conducted in STARCCM+ and the bed model are coupled. Thus, one has to make a convergence study on the bed model. Here one guesses at a temperature and irradiation at first. The generated output values from the bed model are then used as input values for the STAR-CCM+ simulation. The STAR-CCM+ simulation then have to converge before finding the irradiation on the bed again. These values are then used to generate a new output from the bed model and so on. 11 iterations were made before finding equilibrium between the bed model and the STAR-CCM+ simulation. In Figure B.18 and Figure B.19 a comparison between a thermal picture of the bed temperature and the calculated temperature from the bed model are shown. 126 APPENDIX B. PRELIMINARY ANALYSES Figure B.18: Thermal picture of the grate during production. Figure B.19: Temperature at the grate calculated by the bed model, seen from the top of the boiler. In Figure B.18 the picture only covers one side of the grate, which is why there is no information on the lower part of the picture. The image reflects the temperaure just above the bed. The large yellow area corresponds to a temperature of 1350 − 1400◦ C. The bed model was calibrated after this picture. This means that an additional iteration process, beside the one mentioned above, was needed in order to gain the correct temperature profile of the bed. This process involved tuning the combustion percentages in Table 3.2 in section 3.6 on page 27. No academic foundation was used to do this other than the experience in the department of Industrial Processes at Force Technology and the thermal B.4. TUNING IN THE BED MODEL 127 image. In Figure B.20 an attempt to show the temperatures near the bed are made. The white areas of the grate right at the front and back, are the part of the grate not included in the bed model, as the primary air ducts did not cover these regions. An adiabatic boundary condition was used in these areas. Figure B.20: Temperature at the lower part of the boiler. In Figure B.21 the residuals for the simulation behind the generated results above are presented. The fluctuation spikes are seen when the Lagrangian particles are injected. This magnitude of the spikes indicate the high coupling between the particles and the fluid continuum. Convergence is assumed due to decreasing fluctuations towards reasonable level of residuals for a particle combustion simulation. Also a high level of under relaxation, ωLagrange = 0.3, was used on the Lagrangian particles to achieve the decreasing fluctuations. The bed model was converged and tuned in the first 6900 iterations not shown. These were left out to focus on the end result. 128 APPENDIX B. PRELIMINARY ANALYSES Figure B.21: Residuals for the simulation behind the results in the suspension firing and bed model analyses. The high spikes are the results of Lagrangian particles being injected. From the results and residuals the bed model was concluded to be a good representation of the grate firing. The values produced by this model was used for the full boiler simulation in the next section. Thereby avoiding the time demanding task of converging the bed model on the very large simulation. B.5. SIMULATING THE STEAM IN THE SH3 TUBES B.5 129 Simulating the steam in the SH3 tubes There were some difficulties with numerical instabilities in the Bi Conjuncuated Gradient stabilizer(BiCGstab) when simulating the steam inside of the SH3 tubes, coupled to the flue gas side, in one big simulation. After a very extensive analysis of the problem in cooperation with CD-adapco the problem was suspected to be a bad meshing of the SH3 tubes. As the simulation already had reached and exceeded the cell count limits, 14 million polyhedral cells, for doing any calculations in time, a second approach to the problem was needed. First the SH3 tubes were simulated in a separated simulation, as described in the following sections. The solution and mesh from the separate SH3 simulation was imported into the full boiler simulation, and thereby initiating the simulating the steam region with a stable solution. The proces and results for the separate simulation is presented in the following. B.5.1 Mesh used to simulate the steam in the SH3 tubes The SH3 tubes was simulated separately in order to gain better numerical stability in the steam region. By doing this it was possible to use a meshing tool called Generalized cylinders, which is an optional meshing tool available once polyhedral meshing has been selected. This tool gives the possibility to stretch the cells in the longitudinal direction of the tubes. Using this tool it is possible to resolve the curvature of the tube walls much better, but still keeping the same or less, overall number of cells in the steam region, see Figure B.22. (a) Refined mesh using the generalized cylinders tool. 3.115.040 generalized cylinder volume cells in the steam region. (b) Mesh using ordinary polyhedral cells. 5.941.189 polyhedral volume cells in the steam region. Figure B.22: Detailed mesh showing the curvature resolvement for a) Refined mesh using the generalized cylinders tool b) Ordinary polyhedral mesh. No surface mesh was generated due to graphical memory problems caused by the many surface cells. 130 APPENDIX B. PRELIMINARY ANALYSES Not only was the curvature resolved better, but better cell quality was gained by using this tool with only half the number of volume cells. A target size of 1 cm and minimum size of 0.5 cm was used. B.5.2 The physics inside the SH tubes To simulate the physics a steady state turbulent case with real gas simulation was set up. The K-ǫ turbulence model was used to model turbulence and the DOM radiation model to include radiation. The default Cp value for steam in STAR-CCM+ was found to be incorrect for high pressures as in the SH3 tubes. The default value is a sixth order polynomial over a temperature range from 100 to 5000 K. The polynomial fit the values for steam at 1 atm but under estimate the values for steam at 109 bar, which is the actual pressure in the tube, by almost a factor 2. To fix this an additional H2 O specie was added to the simulation. As the temperature range for the steam was suspected to be in the range of approximately 440 − 500◦ C with a Cp value kJ kJ and 2.63 kg·K respectively, a constant value in between seamed to of 2.83 kg·K be a reasonable first choice. The Cp value for this specie was set as a constant kJ value of 2.73 kg·K at 470◦ C and 109 bar, using a zero order polynomial, [38]. This values deviates 5% from the minimum and maximum values expected in the region. For BC, a velocity inlet was used instead of the mass flow inlet, as the velocity inlet is more stable. The velocity was set to get the right mass flow according to the running conditions. For outlet conditions the pressure outlet was used, with a value of 109 bar. A BC for the tube walls of constant heat flux was used to fulfil the heat extracted from the system, found in Table A.1. This does not take the local variances in heat flux into account, and will produce some areas with higher and lower temperatures than actual are, but used as a initial estimation. B.5.3 Results for flow in SH3 tubes separately The simulation gave good results regarding the averaged outlet temperature, giving a final average temperature of 488.5◦ C missing the mark of 490◦ C with only 1.5◦ C, see Figure B.24(a) for average outlet monitor. See Figure B.23 for at temperature profile through the pipes. B.5. SIMULATING THE STEAM IN THE SH3 TUBES 131 Figure B.23: Temperatures in the SH3 tubes, when simulating the tubes separately, a constant heat flux as BC and altered Cp value. The simulation was used to export a better mesh into the full main simulation with the coupled steam region. Furthermore initialization values for for the converged SH3 region consisting of temperature, pressure, turbulent dissipation rate, turbulent kinetic energy and velocity was imported in to the main simulation. No prism layers was used in the tubes as the volume cells were already small. As the outlet temperature only differed 1.5◦ C from the target, it seemed to be an OK trade off between accuracy and wanted export values, see Figure B.24(a). The set up of the simulation described above can be done in a matter of hours with meshing and physics setup. The convergence on 24 processor cores was approximately 2 hours, see Figure B.24(b) for residuals. Last drop in residuals is caused by changing the under relaxation factor for the energy from 0.9 to 0.7. 132 APPENDIX B. PRELIMINARY ANALYSES (a) Monitor plot for the average outlet temperature for separate steam simulation. (b) Monitor plot of residuals for separate steam simulation. Figure B.24: The main monitor values for convergence in the separate steam simulation. a) The average outlet temperature. b) The residuals for the simulation. B.6. AVERAGE OUTLET TEMPERATURE, RESIDUALS, PRESSURE AND HEAT FLUX FOR SH3 IN FULL SIMULATION 133 B.6 Average outlet temperature, residuals, pressure and heat flux for SH3 in full simulation 134 APPENDIX B. PRELIMINARY ANALYSES Appendix C Production values at Verdo during full load Se following pages APPENDIX C. PRODUCTION VALUES AT VERDO DURING FULL LOAD 136 C.1 Air monitor at Verdo Table C.1: Monitor image of the air used for primary and secondary air inlets C.2. STEAM MONITOR AT VERDO C.2 137 Steam monitor at Verdo Table C.2: Monitor image of temperature and mass flow of cooling water/steam 138 C.3 APPENDIX C. PRODUCTION VALUES AT VERDO DURING FULL LOAD Wood chip monitor at Verdo Table C.3: Monitor image of wood chip conveyor belt C.4. BIOMASS FOR SUSPENSION FIRING MONITOR AT VERDO 139 C.4 Biomass for suspension firing monitor at Verdo Table C.4: Monitor image of biomass for suspension firing conveyor belt 140 C.5 APPENDIX C. PRODUCTION VALUES AT VERDO DURING FULL LOAD Fuel analysis See following pages 144 C.6 APPENDIX C. PRODUCTION VALUES AT VERDO DURING FULL LOAD Size distributions for biomass fuel See following pages Prøver 1. Flis 2. Mørk biopiller Prøver 3. BioPiller Lys Havre skaller 4. Biopiller Lys 5. Frøpiller Prøver 7. Smuld 6. Træpiller Vand, Aske og brændværdi (modtaget DTI) Aske %, tør Vand %, våd Brændværdi (NCV) GJ/ton , våd 1 Flis. 0,77 42,60 10,49 2.Mørk biopille. 7,72 17,02 15,67 3.Lys biopille havreskaller 3,68 8,42 17,27 4.Lys Biopille 2 2,79 13,18 16,34 5.Frøpiller. 7,28 14,23 15,61 6.Træpiller. 1,06 7,90 18,45 7.Smuld 0,58 6,17 18,88 Partikelstørrelsefraktioner Partikelstørrelsefraktioner Akkumuleret partikelstørrelsefordeling Andel < 3 mm Appendix D Developed calculation codes 154 D.1 1 2 3 4 5 APPENDIX D. DEVELOPED CALCULATION CODES Matlab code for pipe flow %Denne kode bygger på bogen Fluid Mechanics af F. White og ligning (6.49) for %beregning af friktionsfaktoren og (6.78) for beregning af %mundingshastigheden. % %Ud arbejdet af Svend Skovgaard Petersen 28.06.2012 6 7 8 9 10 clc clear all %masseflow af hele systemet MasseflowTarget=9.62 %kg/s 11 12 13 14 15 16 17 %Spjaeld position: Spjaeld_pos(1) = 0.59; Spjaeld_pos(2) = 0.76; Spjaeld_pos(3) = 0.80; Spjaeld_pos(4) = 0.78; Spjaeld_pos(5) = 0.80; 18 19 20 21 22 23 24 %Spjaeld diameter: Dh(1) = 0.508; Dh(2) = 0.457; Dh(3) = 0.457; Dh(4) = 0.508; Dh(5) = 0.3239; 25 26 27 % air fan pressure % PB = 26−(−1.79)*100; 28 29 30 % temp luft Temp_luft = 182; 31 32 33 34 %Spjæld tryktabskoefficienter KS = [7.3 1.5 1.5 1.5 1.5]; 35 36 37 %Dyse tryktabskoefficienter KD = [1 1 1 1 1]; 38 39 40 %densitet af luft (182 C) rho=0.7787; 41 42 43 %kinematisk viskositet ved ca. 200C og 120 bar vis=2.8e−5; 44 45 46 %Længde af rør L=6; 47 48 49 %Dysemunding diameter Dm = [0.1397 0.1397 0.1143 0.1016 0.038]; 50 51 52 53 %Antal dyser på hver streng n = [7 6 7 12 24]; 54 55 56 57 %Pressure in Pa b(1:5,1)=2600; D.1. MATLAB CODE FOR PIPE FLOW 58 59 60 155 %pre−allokere U vektorer U=ones(5,1)*1; Uold=zeros(5,1); 61 62 63 64 65 66 %iterationsløkke while all(abs(U−Uold)>1e−5) Re=U.*Dm'./vis; %Reynoldstallet f=(1./(−1.8*log(6.4./Re+(0.0001/3.7)^(1.11)))).^2; Km=f.*L./Dm'; %Tryktabkoefficient pga. friktion 67 %sum(K)*rho*0.5 eq.(6.78) A=[(KS(1)+KD(1)+Km(1))*0.5*rho; (KS(2)+KD(2)+Km(2))*0.5*rho; (KS(3)+KD(3)+Km(3))*0.5*rho; (KS(4)+KD(4)+Km(4))*0.5*rho; (KS(5)+KD(5)+Km(5))*0.5*rho]; 68 69 70 71 72 73 74 %gemmer den gamle hastighed Uold=U; 75 76 77 %løser ligningssystemet U=sqrt(b./A); 78 79 80 end 81 82 83 84 %Beregner masseflow ud fra de fundne hastigheder masseflow=sum(rho*Dm'.^2./4*pi.*U.*n'); 85 86 87 %tjekker forholdet mellem givet og beregnet massflow ratio=MasseflowTarget/masseflow; 88 89 90 %korrigere det beregnede masseflow U_final=U*ratio %Friktionsfaktor 156 D.2 1 2 3 4 5 APPENDIX D. DEVELOPED CALCULATION CODES Matlab code for depositing particles %Script to find which boundary faces lagrangian particles hits, and and %sum up on flux on that face. %Svend Skovgaard Petersen 15−06−2012 clear all clc 6 7 8 filename='furnacein_kaer_100000.csv'; %name of output file 9 10 11 12 area=load('furnacearea.csv','csv'); %file with cell face areas data1=load('furnaceout_kaer_100000.csv','csv'); %file with fluxdata %data2=load('tableformatlab2.csv','csv'); %file with boundary cells 13 14 15 diff=length(data1); l=length(area); %number of flux hits %number of faces 16 17 18 19 20 %Initiating zero flux on boundaries flux=zeros(l,4); flux(:,2:4)=area(:,2:4); %setting coordinates 21 22 23 %matrix with cell coordinates cellpos=[area(:,2)'; area(:,3)'; area(:,4)']; 24 25 26 %Effecktive radius of each face radius=sqrt(area(:,1)/pi); 27 28 29 for j=1:diff 30 31 32 temp=0; %resetting vector 33 34 35 36 % If the deposition is not zero if data1(j,1)Ó=0 diff−j %How far again is the calculation 37 38 39 40 %Calc. ditance between particle and cells dis=sqrt((cellpos(1,:)−data1(j,2)).^2+(cellpos(2,:)−data1(j,3)).^2 ... +(cellpos(3,:)−data1(j,4)).^2); 41 42 ratio=min(dis)/radius(find(dis==min(dis))); 43 44 45 46 47 48 49 indice=find(ratio*1.1*radius(:)>dis(:)); %which face does particle hit temp=dis(indice); disttemp=min(temp); %finding the closets/rigth particles k=find(disttemp==dis(indice)); %the the right indice ind=indice(k); if indÓ=isempty(ind) %The particel is within effective radius 50 if flux(ind,1)==0 %is this the first particle hitting flux(ind,1)=data1(j,1)/area(ind); %writing the flux to the face else %if multiple particles on face flux(ind,1)=data1(j,1)/area(ind)+flux(ind,1); %adding flux end 51 52 53 54 55 56 57 else D.2. MATLAB CODE FOR DEPOSITING PARTICLES 58 disp('error') 59 60 end 61 62 end 63 64 65 end flux(:,1)=flux(:,1)*3600; %rewrite to hor basis 66 67 68 69 %writing flux to commaseperated file dlmwrite(filename,flux) 157 158 D.3 APPENDIX D. DEVELOPED CALCULATION CODES Maple code for calculating volatile mass fraction See following pages. Program for calculating the mass fraction of the different vaolatiles and heat of formation for Raw Coal used by STAR-CCM+ > Loading LinearAlgebra Finding the mass fraction Molar mass of species [kg/kmol] > (1) > (2) Procentage of C, H og O. (Fixed carbon=0.07) > (3) > (4) > (5) > (6) > (7) > (8) Ratio between Co og CO2, CO/CO2. From the temperature: > (9) > (10) > (11) > (12) > (13) > (14) > (15) > (16) > (17) Checking if the stoichiometric coefficients is 1: > (18) Rounding of to get 1: > (19) > 1.000 (20) For STAR-CCM+ Number of atoms for starccm: > (21) > (22) > (23) > (24) Finding how many mol rawcoal per kg rawcoal there is > (25) > (26) > (27) > (28) > (29) > (30) > (31) >The number of Atoms, assuming that C should be one gives the: > (32) > (33) > (34) > (35) Heat of formation i kJ/kg enthalpy for end products [kJ/kmol] > (36) Heat of formation for the products in [kJ/kg], /kg because one kg of rawcaol was used. > (37) Calorific value of fuel [kJ/kg] > (38) Heat of formation for fuel in [kJ/kg] > (39) > (40)