Ash deposition propensity of coals/blends combustion in boilers: a modelling analysis based on multi-slagging routes

1 A method that is based on the initial slagging routes and the sintered/slagging route has been developed and 2 used for predicting the ash deposition propensities of coal combustion in utility boilers supported by the data 3 collected from power stations. Two types of initial slagging routes are considered, namely (i) pyrite-induced initial 4 slagging on the furnace wall, and (ii) fouling caused by the alkaline/alkali components condensation in the 5 convection section. In addition, the sintered/slagging route is considered by the liquids temperature, which 6 represents the melting potential of the main ash composition and is calculated using the chemical equilibrium 7 methods. The partial least square regression (PLSR) technique, coupled with a cross validation method, is 8 employed to obtain the correlation for the ash deposition indice. The method has been successfully applied to 9 coals/blends combustion in boilers, ranging from low rank coals to bituminous coal. The results obtained show that 10 the developed indice yields a higher success rate in classifying the overall slagging/fouling potential in boilers than 11 some of the typical slagging indices. In addition, only using the SiO2/Al2O3 ratio to predict the melting behaviours 12 and slagging potential is inaccurate since the effect of the SiO2/Al2O3 ratio is dictated by both the original ash 13 composition and the way in which the SiO2/Al2O3 ratio is changed. Finally, the influence of the acid components 14 (SiO2 and Al2O3) on the ash deposition prediction is investigated for guiding the mineral additives. It is noticed that 15 the predicted ash deposition potentials of the three easy slagging coals investigated decrease more rapidly by 16 adding Al2O3 than by adding SiO2. 17


21
Considerable progress has been made in the last decades in understanding the ash deposition mechanisms of 22 various coals. For example, Eastern US coals (such as Illinois and Appalachian coals) have higher concentrations 23 of Fe components than Western US coals [1], and the initial slagging caused by the pyrite is one of the main issues 24 related to slagging problems [1][2][3]. For low rank Western US coals (such as Wyoming and Montana coals) which 25 have higher concentrations of alkaline/alkali components than Eastern US coals, fouling in the convection section 26 is a serious problem [4][5][6]. Figure 1 shows the main ash deposition mechanisms for US coals in utility boilers [1, 7, 27 8]. Generally, it is regarded that ash deposition can be mainly dictated by three different routes: (i) Pyrite-induced 28 initial slagging route generates from the pyrite particles due to its large density and low melting temperature under 29 reducing atmosphere on the furnace wall [3,8,9]; (ii) Fouling-induced initial slagging route generates from the 30 condensation of alkali vapours and thermophertic deposition of aerosol/fume particles on the superheaters or 31 economizers; (iii) The sintered/slagging route is triggered by the molten matrix generated from the major basic 32 components reacting with clay and quartz, etc., and the reducing atmosphere can promote this process when a high 33 Fe concentration is present in the coal [1,7]. Furthermore, severe slagging in the furnace chamber could increase 34 the furnace exit gas temperature (FEGT) and hence this may further aggravate the ash deposition in the convection 35 section. Therefore, the severe ash deposition in boilers could be triggered by the three different routes and a 36 successful ash deposition indice should be capable of predicting the deposition formation from these three 37 formation routes. 38 conventional slagging indices to predict the slagging potential of coals/blends in an entrained flow reactor. It was 50 found that by incorporating the aerodynamic diameter of fly ash particles into the conventional slagging indices one 51 can improve the prediction performance because the aerodynamic diameter is proportional to the particle Stokes 52 number which determines the particle impaction efficiency [1,15]. It should be noted that the fluid velocity in the 53 EFR is as low as approximately 0.5 m/s [14,16], which means that particles may not have enough kinetic energy to 54 rebound from the deposition surface after impaction and hence deposition accumulation could increase with an 55 increase in the aerodynamic diameter under this low velocity condition in EFR [17]. However, the fluid velocity 56 could be as high as 10-25 m/s in pulverised coal boilers and, for the particles with similar aerodynamic diameter, it 57 is possible to have high enough kinetic energy (proportional to the square of the velocity, possible 20 2 -50 2 times 58 higher than in the EFR) to rebound from the deposition surface after impaction [17,18]. Therefore, the conclusions 59 from the low velocity conditions of the EFR may not be suitable for the real conditions in boilers. Moreover, for 60 some of the existing typical slagging indices (B/A, B/A*Sulfur, Si value, etc.), the slagging prediction for the 61 sintered/slagging route directly employs the mass fractions of ash components and assumes the same contribution 62 of each basic or acid component to the slagging prediction. However, the sintered/slagging layer is not linearly 63 This paper takes a new approach to build an ash deposition indice for fuel slagging propensity analysis. The 70 ash deposition indice takes into considerations the multi-ash deposition routes which exist in industrial boilers and 71 the indice is developed with support from the actual observations made in a range of industrial boilers. The 72 sintered/slagging route is predicted by using the overall melting potential of the major ash components through the 73 chemical equilibrium calculations; the initial slagging route caused by either the pyrite or the alkaline/alkali 74 components is predicted by using the amount of the related basic ash components; the known slagging observations 75 for coal combustion in boilers are used as the training data to acquire the correlation of the slagging indice. The 76 partial least square regression (PLSR) method, coupled with cross-validation, is employed to develop the slagging 77 indice. The study reported in this paper is mainly based on the available data for typical US coal, and the results are 78 to represent the sintered/slagging route [29,30]. This assumption is employed for all types of coals. In order to 97 evaluate the melting capability, liquidus temperature ( ) is employed and predicted by using the chemical 98 equilibrium calculations. 99 (v) For those coals with Fe 2 O 3 as the major basic oxide, the deposition is formed mainly in the radiant section 100 with the slag rich in the iron content [8]. However, for the coals with alkaline/alkali constituents as the major 101 fluxing mineral, serious ash deposition (fouling) is observed in the convection section [31]. Hence coal ash can be 102 classified into two types, the lignitic and bituminous types of ash [14,32]. For lignitic type ash defined as the 103 amount of either alkaline or alkali components being greater than the amount of Fe 2 O 3 , only the initial slagging 104 route caused by the alkaline/alkali condensation is considered as the major initial slagging route. For bituminous 105 type ash defined as the amount of Fe 2 O 3 being greater than the amount of alkaline and alkali components, only the 106 initial slagging route caused by the pyrite is considered as the major initial slagging route. 107 Therefore, based on the above assumptions, the proposed method to build the ash deposition indice is 108 developed as follows: for bituminous type coal, the liquidus temperatures under the oxidizing atmosphere and the 109 reducing atmosphere ( and ), the SiO 2 +Al 2 O 3 content, the Fe 2 O 3 content and the total sulphur content can be 110 employed as the independent variables; for lignitic type coal, the liquidus temperatures under oxidizing atmosphere 111 and reducing atmosphere, the SiO 2 +Al 2 O 3 content, and the alkaline/alkali content can be employed as the 112 independent variables. The overall slagging/fouling observations can be employed as the dependent variable. The 113 partial least square regression (PLSR) technique, coupled with a cross validation method, is employed to obtain the 114 correlation for the indice. This is because (a) in this work, the data of slagging observations is limited and the 115 independent variables in the method to build the ash deposition indice are highly correlated, and (b) the PLSR 116 method is specifically designed to deal with multiple regression problems where the number of observations is 117 limited and the correlations between the independent variables are high [33]. 118

Prediction of the liquidus temperature 119
The liquidus temperature is the temperature at which the first solid phase just starts to precipitate on the 120 cooling of a slag-liquid oxide melt [21]. The temperature is predicted based on the major ash composition (Al 2 O 3 , 121 SiO 2 , Fe 2 O 3 , CaO, and MgO) by using the chemical thermodynamics software FactSage 6.4 [21]. The software is 122 based on the minimization of the Gibbs free energy from the system subject to the mass balance constraints [34,35]. 123 The calculations were performed by using the equilibrium module together with the databases ELEM, FToxid, 6 immiscibility [21]. Five major ash components (Al 2 O 3 , SiO 2 , Fe 2 O 3 , CaO, and MgO) were included in the 126 calculations; for lignite, Na 2 O is also included due to its high amount; K 2 O is excluded due to its low amount in the 127 ash; however, the other components (SO 3 and P 2 O 5 ) were also neglected due to the fact that S and P are volatile 128   Based on the ash compositions listed in Table 1, the ash deposition indice for the bituminous type is 161 calculated in Case 1 and the ash deposition indice for the lignitic type is calculated in Case 2. The training data, 162 which cover fuels of low, medium and high slagging propensities, contain less than half of the total data set and 163 therefore the testing data contain more than half of the total data set; see the Supporting Information for details. 164

PLSR and Cross-Validation
After performing the PLSR and Cross-Validation calculations, the correlations for calculating the ash deposition 165 indice, for these cases are as follows: 166 Case 1: (1) Case 2: (2) For both cases, the liquidus temperature and have negative coefficients, which implies that the 167 predicted slagging observation will decrease with an increase in the liquidus temperature and . 168 However, the parameters related to the initial slagging routes ( , , CaO+MgO and Na 2 O+K 2 O) have a 169 positive coefficient which means that the predicted slagging/fouling observation increases with a higher content of 170 these four parameters. 171 observations. In addition, the uncertainty of the predictions may be attributed to the number of the training data set. 175 In our calculations, we tried the number from 5 to 9. The predicted average relative errors range from 16.8% to 176 19.3% for Case 1 and from 9.0% to 9.4% for Case 2, which indicates that the prediction performance may not be 177 greatly affected by the number of the training data. Figure 4 illustrates a comparison between the predicted 178 performance of the ash deposition indice, in this study and some of the conventional slagging indices based on 179 the ranked slagging observations. It can be observed that the ranking for the accuracy of the prediction performance 180 . In contrast, conventional slagging indices had limited success rates, ranging from 1 to 7 for case 1 (out of 13) 186 and 0 to 12 for case 2 (out of 17). Therefore, the indice built by considering multi-slagging routes yields a higher 187 success rate in classifying the overall slagging/fouling potential in boilers than that of the typical slagging indices. 188 In addition, Fig. 5

Sensitivity of the method 205
Adding mineral additives is common practice in order to control the slagging and fouling problems in boilers. 206 Therefore, the influence of adding acid components to coals that show higher deposition potential was investigated 207 to test the sensitivity of the developed method. 208 Either SiO 2 or Al 2 O 3 was added as an additive to three easy slagging/fouling US coals and the predicted values 209 of the indice are plotted against the added SiO 2 or Al 2 O 3 content of the fuel and the SiO 2 /Al 2 O 3 ratio as shown in 210 Fig. 6. The sensitivity study indicates that by adding either SiO 2 or Al 2 O 3 can reduce the predicted slagging 211 11 potential. This is because the added acid components could reduce the melting potential due to the increase in the 212 liquidus temperature. In addition, the acid components could capture the alkali/alkaline vapour phase to decrease 213 the condensation potential. Also the analysis shows that the value of the predicted slagging potential decreases 214 more rapidly by using Al 2 O 3 than when adding SiO 2 . Van Dyk et al. [44] and Li et al. [45] also found that Al 2 O 3 is 215 more effective than SiO 2 due to its higher ability to increase the ash fusion temperature than that of SiO 2 . However, 216 the analysis shown in the right section of Fig. 7, indicates an opposite trend corresponding to the SiO 2 /Al 2 O 3 ratio 217 when adding SiO 2 compared to adding Al 2 O 3 . It is noticed that Song et al. [20] found that ash fusion temperatures 218 (AFTs) are increased with increasing the SiO 2 /Al 2 O 3 ratio from the fusion experiments and chemical equilibrium 219 calculations. However, Liu et al. [46] found that AFTs are decreased with increasing the SiO 2 /Al 2 O 3 ratio from the 220 fusion experiments. This is because, see Ref.
[20], the SiO 2 was added into the ash with a relatively low CaO 221 content (approximately 15%) and adding the SiO 2 can lead AFTs to move from the low temperature region into the 222 high temperature region [20]; However, see Ref.
[46], when the SiO 2 is added into the ash with relatively high CaO 223 content (approximately 40%) the added SiO 2 can react with CaO to generate the low-melting anorthite and 224 gehlenite and this leads the AFTs to move from a high temperature region to a low temperature region [46]. In this 225 study, all the three coals with higher slagging propensity have relatively low/medium CaO content (ranging from 226 2.9% to 21.8%) and adding either SiO 2 or Al 2 O 3 could increase the liquidus temperature from the chemical 227 equilibrium calculations. In addition, further calculations, using chemical equilibrium methods, were undertaken to 228 investigate the influence of the SiO 2 /Al 2 O 3 ratio on the melting potential. In the calculations: (i) In addition to the 229 three coals tested in this study, coal ashes from Ref. with the slagging observations for the 30 US coals/blends with a history of ash deposition issues. The indice built 249 by using the proposed method yields a higher success rate in classifying the overall slagging/fouling potential in 250 boilers than those existing slagging indices. It is postulated that this method has a potential to be used as an 251 alternative tool to build an ash deposition indice for industrial use with a better prediction performance compared to 252 existing slagging indices. In addition, an advantage of this method is that the newly developed indice based on the 253 known slagging/fouling history from multiple boiler units makes it more suitable for different boiler configurations 254 and coal types, although some of the aspects regarding the ash chemistry need to be further investigated in order to 255 improve the accuracy and extend the application range of the proposed method. Without addressing the specific 256 conditions in a boiler, the performance of a predictive method could be less accurate [47]. The index reported in 257 13 this study does not consider the combustion conditions explicitly in its formulation and this may be limited to the 258 conditions observed in the units used to validate the index. Incorporating changes in combustion conditions could 259 be an ideal path moving ahead to further improve the accuracy of this index. 260 It should be noted that the initial slagging route caused by the pyrite is represented by the contents of Fe 2 O 3 and 261 sulphur for US coals since they are known to contain iron, predominantly in the form of pyrite [ increase the accuracy of predicting deposit formation from alkali condensation. Also, it should be noted that the ash 273 loading, which can affect the deposit accumulation in boilers [10], is not considered in the proposed slagging indice 274 because there is no significant difference in the ash loading for the tested US coals. However, if there exists a great 275 difference in the ash loading, the parameter should be considered in the prediction model and this can be done by 276 using the value of the ash compositions/ ash loading to replace the existing value of ash compositions [1,10]. 277

278
A novel method to build an indice is developed and used for predicting the overall slagging/fouling potential 279 of coal/blends combustion in boilers. The method couples the initial slagging route caused either by pyrite or by 280 alkaline/alkali components and the sintered/slagging route. The initial slagging route is predicted based on the 281 corresponding ash components and the sintered/slagging route is predicted based on the overall melting potential 282 using the liquidus temperature calculated from chemical equilibrium methods. Utilizing the available slagging 283 observation data from US coal fired boilers, PLSR coupled with the cross validation method was employed to 284 develop the new ash deposition indice. 285 14 It should be noted that both SiO 2 and Al 2 O 3 can reduce the slagging potential, but the drop in slagging 286 propensity is more significant by adding Al 2 O 3 compared to SiO 2 as confirmed by chemical equilibrium 287 calculations. Finally, using the SiO 2 /Al 2 O 3 ratio alone to predict the melting behaviours and slagging potential of 288 coals is inaccurate owing that the SiO 2 /Al 2 O 3 ratio alone cannot dictate the overall melting behaviour. The 289 proposed method has been validated against the field performance of slagging observations on 30 sets of US 290 coals/blends combusted in utility boilers. The results obtained indicate that the developed indice shows a much 291 higher success rate for ranking the overall slagging potential in boilers than the other five conventional slagging 292 indices. 293