Ash fusion temperatures under oxidizing conditions - Energy & Fuels

Effect of Coal Ash Composition on Ash Fusion Temperatures. Energy & Fuels. Song, Tang, Zhu, Wu, Zhu and Koyama. 2010 24 (1), pp 182–189. Abstract: T...
0 downloads 0 Views 531KB Size
Energy & Fuels 1993, 7, 490-494

490

Ash Fusion Temperatures under Oxidizing Conditions William G. Lloyd,* John T. Riley, Shiyong Zhou,l Mark A. Risen, and Richard L. Tibbitts Department of Chemistry and Center for Coal Science, Western Kentucky University, Bowling Green, Kentucky 42101 Received March 2, 1993. Revised Manuscript Received April 14, 1993

The ash fusion temperatures (AFTs) of 70 ashes (obtained from blends of 7 North American source coals of varying rank) have been determined under oxidizing atmospheres in accordance with ASTM method D 1857. These oxidizing AFTs have been compared with the corresponding AFTs obtained under reducing conditions. The temperature spreads, [AFTOXIDIZING - A F T ~ u c I N Ghave I , been found to vary with ash composition, from -16.5 O F (-9 K) to +425 O F (+236 K). A procedure is described for the development of regression equations to estimate each of the oxidizing AFTs, based upon ash compositional data. The best regressions provide AFT estimates with root mean square errors of 33-36 O F (18-20 K) ( n = 67).

Introduction The laboratory determination of ash fusion temperatures (AFTs) originated as a guide to the clinkering behavior of coal ash in stoker furnaces.2 It is used worldwide nowadays to predict the slagging potential of coals fired in pulverized coal utility boilers. Most AFTs are measured under reducing conditions. Gas-phase burning is associated with reducing conditions a t the particle surfaces and is the characteristic mode of combustion for large coal particles. Heterogeneous burning involves oxygen-solid contact and hence oxidizing conditions at the particle surfaces. For coal particles between 65 and 15 pm combustion appears to follow both gas- phase and heterogeneous processes,while for coal particles below about 15 pm initial burning appears to be completely heterogeneous.3 In large pulverized coal boilers, moreover, most of the coal mass occurs in particles smaller than 65 pm and perhaps 20-30% is in particles smaller than 15 pm.3 Most coal is in fact burned under oxidizing condit i o n ~ We . ~ have reported previouslyupon the relationship of ash composition to AFTs for 70 ashes measured under reducing atmosphere^.^' In the present work we have measured AFTs of the same 70 ashes under an oxidizing (air)atmosphere and have examined the relationships with reducing atmosphere AFTs and with compositional data. Particular attention has been given to the iron oxide component, since of the major ash components only Fe is expected to exist in different oxidation states in reducing and oxidizing (1) Visiting scientist from Anshan Institute of Iron and Steel Technology, Anshan, China. (2) Hough, D. C. ASME Ash Fusion Research Project Report FTI 88/01. American Society of Mechanical Engineers Report CRTD-18; ASME: Fairfield, NJ, 1988. (3) Ceely, F. C.; Daman, E. L. Combustion Process Technology. In Elliott, M. A., Ed. Chemistry of Coal Utilization; Wiley: New York, 1981; 2nd suppl. vol., Chapter 20. (4) Reid, W. T. Coal Ash - Ita Effect on Combustion Systems. In Elliott, M. A,, Ed., Chemistry of Coal Utilization; Wiley: New York, 1981; 2nd suppl. vol., Chapter 21. ( 5 ) Lioyd, W. G.; Riley, J. T.; Risen, M. A.; Gilleland, S.R.; Tibbitta, R. L. Energy Fuels 1990, 4, 360. (6) Lloyd, W. G.; Riley, J. T.; Risen, M. A.; Gilleland, S.R.; Tibbitta, R. L. Prepr. Pap.-Am. Chem. SOC.,Diu. Fuel Chem. 1991,36 (I),235. (7) Lloyd, W. G.; Riley, J. T.; Riaen, M. A.; Gilleland, S.R.; Tibbitta, in nrens R. L. J . Coal Chin1 ~--..,.__ (8)Gray, 17. R. Fuel 1987,66, 1230. _--I.

0887-0624/93/2507-0490$04.00l0

Experimental Section Laboratory Procedures. Seven source coals, of rank from lignite A to medium volatile bituminous, were selected. Sources and characteristics have been reported previously.6 After reduction to -60 mesh (-250 pm) these coals were blended in 3:1,1:1, and 1:3 ratios, providing 63 blends. Ash samples from the source coals and blended coals were prepared in accordance with ASTM Method D 1857.9 AFTs under oxidizingconditions were determined on ash splits using a LECO Model AF-600 aeh fusibility system operating under a continuous air stream in accordance with the above method.g Ash splits were analyzed by XRF and ICP spectrometry,using overlapping ash standardsq6 The compositional range of the ashes studied is shown in Table I. Statistical analyses were conducted using PC release 6.03 of the Statistical Analysis System.10 Selection of Regression Terms. The simplest set of compositional predictor terms is the set of weight fractions of the 10 principal oxides of Table I. To this we added the 45 second-order terms (cross terms). (Cross terms are abbreviated in the following discussion,for example, Na-S for [Na201[SO3].) Of these 55 candidate predictors, many pairs of terms were highly correlated with one another. In multiple linear regression (MLR) analysis a high correlation between two predictor terms can lead to the condition of excessive collinearity, in which good-looking values for such measures of fit as R2 or root mean square error (rmse) can mask regressions of no real predictive power. Mason, Gunst, and Hessll suggest 0.95 as a reasonable upper limit for the correlation coefficient Rij between predictor terms in MLR analyses. We selected 0.92 as the maximum permissible value for Rij, at which value 19 of the 55 terms were free of excessive correlation with any other terms. The remaining terms occurred in correlation clusters, for example, a cluster of 10 terms containing [CaO] or [K201.6 Selection of the most useful (9) Test Method for Fusibility of Coal and Coke Ash, Method D 1867. Annual Book of ASTM Standards, American Society for Testing and Materials: Philadelphia, PA, 1992; Vol. 6.06 (published annually). (10) SAS Institute, Inc., Cary, NC, 27513. (11)Mason, R. L.; Gunst, R. F.; Hess, J. L. Statistical Design and Analysis of Ezperiments; Wiley: New York, 1989; pp 696 ff.

0 1993 American Chemical Society

Energy & Fuels, Vol. 7, No. 4,1993 491

Ash Fusion Temperatures under Oxidizing Conditions

component &Os

Table I. Comporitiond Ranger of Coal Arher 10 ashes with lowest ST spread' min,% ma,% mean,%

min,%

28.7 26.9 25.1 2.63 12.5 1.66 2.36 15.4 50.5 1.66

13.9 0.6 3.5 0.16 1.05 0.51 0.01 0.72 28.6 0.82

CaO FezOs K2O Mgo Na2O PZOS

sos

10 ashes with highest ST spread0

all 70 ashes ma,% mean,%

13.9 2.7 3.9 0.16 1.06 0.63 0.63 1.08 28.6 0.82

22.2 10.6 8.8 1.37 2.28 0.91 0.77 7.79 40.1 1.11

25.7 26.9 5.6 1.91 12.5 1.34 2.36 13.4 50.4 1.66

min,%

max, %

mean, %

19.2 0.6 3.7 0.79 1.06 0.51 0.01 0.72 40.7 0.94

22.3 14.0 25.1 2.63 1.36 0.74 0.91 5.10 47.0 1.14

21.0 3.8 19.2 2.16 1.14 0.68 0.22 3.23 44.1 1.05

21.2 14.3 4.8 1.24 2.91 0.96 1.10 7.84 39.9 1.05

Si02 Ti02 O ST spread is [T"G(oxidizing) - Tsop~~"~(reducing)]. The average spread inG "T for the full 70-ash set is 122 O F (68 K); for the lowest-spread subset, it is 3.2 OF (1.8 K);and for the highest-spread subset, it is 331 OF (184 K). 450450 400-

350350

i

I

I /

I I

/

/I

D / D

.

/ I

II D D

I / .

1

-4 m I

I

I

I

I

I

I

I

I

I

350

300 Y

f

I

LL

250

F

F"

Q

150

EL

100

100-

.. . . D m

1 . .

1

2100

2630

A00

2400 2 h 2$00 ST(O), degree$ F

2;00

2800

2&

-50-1

0

I

I

0.05

,I

0.1

I 0.15 FeZO3 fraction

,I

0.2

I

I

0.25

0.3

Figure 1. Distribution of spreads between ~somdoxidizing) and Twmdreducing). Ashes in the upper leg (A) have a median Fe208:Ca0ratio of 7.0; those in the lower right leg (B) have a median ratio of 1.6.

Figure2. Distribution of spreadsbetween TmmG(0Xidizing) as a function of Fez03 fraction. Ra = and T"G(reducing)

terms in these clusters was carried out using the following sequence for each AFT: (1)The group of 19 "free" terms was selected as the base pool. (2) To this pool was added separately each term appearing in the largest cluster. The term making the largest improvement in the regression fit (R2)was added to the base pool. (3)All terms in the cluster which were correlated with the selected term at Rij > 0.92 were now excluded from the pool. (4) Steps 2 and 3 were repeated until all terms in the cluster had been either added to or excluded from the pool. (5) Each of the other clusters was examined similarly by repeating steps 2-4 until all of their terms had been added to or excluded from the pool. (6) When all clusters were examined, the entire selection process was repeated through at least two more full iterations and until the selected pool terms were unchanged. Thus for six clusters, A, B, ...,F, the selected terms from B...F were retained while cluster A was reexamined; then terms from A, C, D, ...,F were retained while cluster B was reexamined. This pattern was continued until every cluster of correlated terms had been reexamined. This procedure reduced the number of candidate predictors to a pool of 28-30 terms for each

rapidly, then progressively more slowly until there is little or no incremental improvement. At the same time total collinearity among predictor terms increases progressively. For this set of ashes under reducing conditions the optimum number of terms was found to be 10.' To select the "best" 5 1 0 4 " regressions for estimating an AFT, two criteria were applied: (1) A group of 20 10-term regressions was obtained, selected by maximizing R2 measurements of fit from a large field of possibilities (over 3 X 107 for a 30-term pool). Each selected regression was expanded to provide coefficients and standard errors for each coefficient. (2) The one term common to every regression is the intercept. As a second screening for excessive collinearity, the standard error of the intercept (SEI) was extracted from each regression. Arbitrarily, only those regressions with SEIS less than 90 O F (50 K) were retained. Approximately two-thirds of the regressions generated in step 1 met this criterion. Outlier Cases. When the five best MLR equations had been selected,estimates for the individualcases (ashes) were examined. When the error of estimate E was more than 2-fold greater than the standard error of the residual a,the case was flagged. Data for all five regressions were compared. If for any case the average value of E / - for the five best regressions exceeded 2.0, that case was identified as an outlier and was dropped. Outlier testing

AFT. Selection of Specific Regressions. As the number of terms in a MLR equation is increased, its predictive power as gauged byR2measurementsof fit at first improves

0.72.

492 Energy & Fuels, Vol. 7, No. 4, 1993 Table 11. Estimating- Oxidizing- AFTs from Beducing AFTs and Selected Composition Terms. R2AD.J rmse 0.928 41.5 ID0 = 1686 + 0.392ID~ 22 6OO[Al-S] 1.04 X l@[Na-Ti] - 25 3OO[S-Sil 0.926 42.2 ID0 = 832 0.551ID~ - 14 soO[Ca-P] + 41 400[Fe-K]+ 46 7OO[Si-Til 0.924 42.5 ID0 = 816 + 0.558ID~ 42 lOO[Fe-K] = 21 200[P-S] + 46 7OO[Si-Til 43.1 0.922 ID0 = 808 0.554ID~+ 46 100LFe-KI + 46 3OO[Si-Til 47.6 0.905 ID: = 589 0.626IDR + 55 1001Fe-KI + 23 5OO[Til 0.901 48.7 ID0 = 1005 + 0.573ID~- 92 lOO[Ca-K] + 24 GOO[K-Sil 40.3 0.933 STo = 399 + 0.678s"~ 814[Si] + 1.18 X l@[Fe-Til 0.923 43.3 STo = 1366 + 0.552ST~- 348O[s-si] - 361O[Ca-Sil 44.0 0.920 STo = 237 + 0.742ST~+ 874[Sil+ 1230lFel 44.3 0,919 STo = 1336 0.569ST~- 65 000[K-S3- 4890[Ca-Sil 0.919 44.5 STo = 328 + 0.7485"~+ 2850[Fe-Sil + 625iSiI 44.9 0.917 STo = 1304 + 0.573ST~ - 3800[Al-S] - 417O[Ca-Sil HTo: no valid regressions with R2 > 0.90 FTo: no valid regressions with R2 > 0.90

+ + +

+ +

+

+

+

4

Lloyd et al. SEI 89.9 88.8 88.8 90.0 86.1 82.7 74.7 89.3 85.5 90.3 84.1 90.4

Temperatures and temperature ranges are in OF.

was done separately for each AFT. Sometimes more than three consistent outlier cases were identified, but no more than three outliers (those with the highest average values of EIBE)were dropped.

Results Most AFTs determined under oxidizing conditions are higher than the corresponding fusion temperatures determined under reducing conditions. Figure l shows the scatter of temperature differences between T S O ~ N I N G (oxidizing) (STo) and Tmi+"+mNG(reducing)(STR).The average STRfor this set is 2229 O F (1221 K)while the average STo is 2351 O F (1288 K). For individual ashes the spreads are highly variable, from -16.5 OF (-9 K) to +425 O F (+236 K). At STo values greater than 2450 OF the data become bifurcated; the ashes in the upper leg (region A of Figure 1) have high Fe:Ca ratios while those in the lower leg (region B) have much lower Fe:Ca ratios. Figure 2 shows the distribution of [STo - STRI as a function of Fez03 fraction. The data at higher temperatures are no longer bifurcated. For the 67 cases R2 = Figure 3. Estimate of Tsommo(oxidizing) from T"G0.72;that is, 72 % of the total dispersion of the temperature (reducing), [Si], and [Fe-Til (Table 11). R2 = 0.9362; rmse = spread [STo - STR]is accounted for by this single term. h40.3 OF; average error = 29.4 O F . However, further exploration has shown that Si- and S-containing terms have comparable predictive values. Winegartner and Rhodes have reported [Ca-Si] to be the Table I1 lists regressions estimating T m D ~ R M A ~ O N - best single predictive term for oxidizing AFTs.12 Among (oxidizing) (IDo) and STo from the corresponding the reducing atmosphere regressions the frequency is reducing atmosphere AFTs and additional terms. Many somewhat different, but again only two of the eight most of the better regressions do not contain Fe terms. The fit common terms ([Ca-Fel and [Fe-Mgl) contain an iron of the best of the STo regressions in Table I1 is shown in component. The most commonly occurringterms are [All Figure 3. No comparably satisfactory regressions on (positive), [Al-Si] (negative), and [P-SI (negative). WinTHEMIspHERICAL(0XidiZing)(HTo) 01 On TpLun)(OXidiZing) egartner and Rhodes found [Fe-Si] to be the strongest (FTo) were attained by this approach. single term.12 The best predictive expressions for all of the AFTs are In this data set 5 of the 10 Fe-containing terms (Al-Fe, those using compositional terms. Table I11presents eight Fe, Fe-K, Fe-Si, and Fe-Ti) are strongly correlated with such regressions. For oxidizing atmosphere AFTs the one another, with Rij > 0.94 in every case. Following an goodness of fit, as indicated byR2mJ, is in the range 0.942earlier strategy: we subjected these five terms to factor 0.955, with root mean square errors of estimate in the analysis (principal component analysis). Eigenvalues for range 33-36 O F (18-20 K). Reducing atmosphere AFT the first two Varimax factors are 4.88 and 0.07, indicating ~ estimates show similar precision of estimates: R 2 m in a single significant common factor. A number of regresthe range 0.940-0.963 and rmse 29-37 O F (16-21 K). Figure sions on STo were carried out using this factor, the 4 shows the fits of the best regressions found for STo and corresponding STR,and the other five lightly correlated STR. iron terms (Ca-Fe, Fe-Mg, Fe-Na, Fe-S, and Fe-P). No Among the 20 best oxidizing atmosphere regressions regressions were obtained with SEIS below 90 O F . Even the most commonly occurring terms are [S-Si] (negative with high SEIS, implying excessive collinearity among cofficient), [All (positive), [Al-Fel (positive), [Ca-Fel predictor terms, the goodness of fit judged by R2 was (positive for IDo, negative for FTo), [P-SI (negative), inferior to those of the regressions of Table 111. [P-Si1 (positive), [Mg-PI (positive),and [Nal (positive). (12) Winegartner,E. C.; Rhodes, B. T. J. Eng. Power 1975,97,395. Only two of these eight terms contain an iron component.

Energy & Fuels, Vol. 7, No. 4, 1993 493

Ash Fusion Temperatures under Oxidizing Conditions

Table 111. Selected AFT Regressions* AFT intcpt Coefficients x lo-' K-S Fe-Na K-Na K-P Ca-Ti AI-Si Ca-Fe IDo Al -24.9 +416 -43.7 -49.7 +4.06 -22.1 -1.15 1868 +0.738 P Mg-Na Na Na-Ti Fe-Ti Al-Si Fe-Mg STo AI +19.9 +2.84 +3.82 -168 -4.83 +19.4 1629 +0.533 -0.697 Ca-Na Mg-P P-s NaSi Al-S Ca-Mg Al-Fe AI-Na HTo -22.1 +5.98 +35.2 -19.1 +44.0 -2.64 -2.13 2463 +0.227 KSi Na-P K-S Na AI-Fe Ca-Fe Fe-Na FTo AI -6.22 -200 +16.8 +5.78 -1.49 -23.8 +1.62 2288 +0.160 P-s Fe-Na Na-Si AI-Si Ca-Fe Ca-Ti IDR AI MiT-P -8.00 -42.0 +13.8 +20.4 -1.62 +4.56 -21.5 1631 +0.964 Na-P Fe-Mg K-Na Mg-Si AI-Si Ca-Fe AI AI-Fe STR +429 -7.15 +2.16 +2.19 +195 -1.10 1611 +0.762 -0.402 K-Mg K-P Mg-P Na Fe-Mg A1-Si Ca-Fe AI HTR +4.52 +27.3 +111 +30.8 -7.64 +2.29 1592 +0.531 -0.670 K-Mg Mg-P P-s Na AI-Si Ca-Fe Fe-Mg FTR AI -17.0 +29.9 +21.3 +4.80 -6.15 1676 +0.705 -0.953 +0.759 a Temperatures and temperature ranges are in OF.

P-Ti +136 P-s -22.6 P-Si +2.98 P-Si +6.15 P-Ti +111

P-s -42.6

P-s -15.5 P-Si +3.94

R2m

rmse,OF

S-Si -0.745 s-SI

0.9472

35.5

-0.860

0.9550

33.1

0.9468

34.4

0.9416

35.8

0.9401

36.8

0.9470

35.2

0.9627

29.1

0.9491

34.9

S-Ti +13.3 S-Si -2.02 S-Si -2.50 S-Si -1.71 S-Si -1.61 S-Si -1.90

j

Ij 2350

I

/

I

,/ /

I

IL

2300

!

I

,250

I

I

I 2000

0

0.25

0.5

0.75

1

Fraction of hvBb coal

Figure 6. Ash fusion temperatures (reducing)for ashes obtained from blends of two coals. The left-hand axis represents ash from 100% subbituminous A coal (Jacob's Ranch, WY), CaO content 18.5%. The right-hand axis represents ash from 100% high volatile B bituminous coal (Western Kentucky #12 seam, Muhlenberg Co., KY), CaO content 0.6%. Figure 4. Estimates of Tso"G(oxidizing) (solid squares) and Tso-G(reducing) (open squares) from the regressions in Table 111. For the oxidizing atmosphere, rmse = f 33.0 O F and average error = 24.8 OF. For the reducing atmosphere, rmse = f 35.2 O F and average error = 28.9 OF.

in SO3 and because it is suspected that laboratory test cones may lose much of their sulfur content before deformation temperatures are recorded.12 The relevancy to boiler slag composition is an appropriate question to raise. The suggestion that SO, is substantially lost from Discussion laboratory test cones warrants experimental investigation. The best predictive regressions for the estimation of Our data, particularly with ashes from blends of highAFTs under oxidizing conditions are based upon comcalcium and low-calcium high-sulfur coals, show that the positional terms. Notwithstanding our expectations and AFTs of intermediate blends are far from linearly interthose of 0thers,~*8J2 iron-containing terms were not of polated values (Figure 51, suggesting that at least a dominant importance in accounting for the differences substantial fraction of captured sulfur is retained during between oxidizing and reducing AFTs. From our data, A F T measurements. Lacking data to the contrary we have silicon- and sulfur-containing terms are the most importreated the analyzed SO3 component as real and relevant tant, with high levels of Si and S associatedwith substantial the estimation of laboratory AFTs. Also, we get better spreads in the difference (AFTOX~IZING - AFTREDUCING)to regressions when sulfur terms are included. for all four AFT pairs. Regression based upon simple chemical analysis, using Some workersaJ2have modified their raw compositional no more than binary interactions and ignoring mineraldata by normalizing it prior to analysis. Others have ogical data, might not be expected to be sufficient for converted weight fraction data to a molal basis. These prediction of AFTs. On the basis of careful phase studies, operations change the numerical values of the regression however, Huffman and Huggins13J4 have concluded that coefficients but have no impact upon the precision of prediction of an ash fusion temperature by regression (13)Huggins, F. E.; Kosmack, D.A.; Huffman, G.P.Fuel 1981,60, analysis. 517. Sulfur-containing terms have been excluded from other (14)Huffman,G.P.;Huggins, F. E.; Dunmyre, G.R.Fuel 1981,60, 585. studies,8J2 because boiler slag has been found to be low

494 Energy &Fuels, Vol. 7, No. 4, 1993

even the initial deformation temperature is preceded by extensive formation of liquid phase. The strength of the present regression analysessupports the view that slagging ash is a mixture of relatively simple compounds. Multiple linear regression analysis, when applied with the systematic exclusion of excessive collinearity, is a powerful tool for the estimation of fusion temperatures of coal ashes, for both oxidizing and reducing atmospheres. For a given well-analyzed family of coal and coal blend ashes,if reliable ash compositionaldata are available,MLR estimates offer precision comparable with that of laboratory measurements of fusion temperatures.

Lloyd et ai.

Based upon several testa for excessive co1linearity:J we have set two entirelyarbitrary screeningcriteria in selecting the "best" regressions: Rij between predictor terms