Document z58jn3oRZ3VaE769R0QqEDq0

Chcmosphcrc. Vol.26, No. 11, pp 2039-2054.1993 Printed in Great Britain 0045-6535/93 S6.00 + 0.00 Pergamon Press Ltd. CORRELATION OF CHLORINATED ORGANIC COMPOUND EMISSIONS FROM INCINERATION WITH CHLORINATED ORGANIC INPUT John C. Wagner and Alex E.S. Green Clean Combustion Technology Laboratory, SSRB, University of Florida, Gainesville, FL 32611 (Received in USA 21 September 1992; accepted 22 February 1993) ABSTRACT This paper summarizes the results of stack emission measurements made since 1988 at the University of Florida-Tacachale-Clean Combustion Technology Laboratory. We find several statistically significant relationships between HC1 emissions (a surrogate for PVC In the waste) and the emissions of a number of chlorinated organic compounds. These relationships have important implications for clean combustion technology and public policy. Key words: emissions, chlorinated hydrocarbons, HCI, PVC, incineration. INTRODUCTION Is there a relationship between polyvinyl chloride (PVC) input and the chlorinated hydrocarbon (C1HC) emissions from incineration? The prevailing (almost official) viewpoint Is "No" [1-2]. A minority says "Yes" [3-8]. In this paper we report the data and recent statistical analyses which support Yes and examine the possibility of reconciling'these answers. The Clean Combustion Technology Laboratory (CCTL) has been engaged in a program of research and development on co-combustion since 1980 [9-13]. In 1987 a donated 500-lb/hr incinerator was reconstructed and retrofitted in front of the steam plant at the Tacachale center in Gainesville,FL The facility was made operational on December 24, 1987, and has been used for a variety of co-combustion experiments and emission measurements [14-15]. Here we report a detailed analysis of the results of measurements of emission of hydrogen chloride (HCI) and volatile organic compounds (VOC's) [3,6] when firing non-hazardous institutional waste with various levels of PVC. The incinerator system consisting of a primary combustion chamber (PCC), a secondary combustion chamber (SCC) and its auxiliary equipment (see Figure 1) has been described previously |(>]. PVC, HCL AND VOST DATA FROM CCTL While the non-hazardous waste used in the various trial burns reported here were quite uniform it is useful to have a surrogate to indicate the level of chlorine in the waste that can be compared with the chlorinated organic emissions. Since the best surrogate is HCl.lhe major product of chlorine combustion, an EPA Method 26 [16] sampling train has been used to sample hydrogen 2039 2040 chloride (HC1) emissions from the horizontal stack of the CCTL incinerator The CCTL has made over 30 measurements of HC1 emissions with different levels of chlorine in the input 13) (see Table 1) The PVC pipe added during the earlier experiments was assumed to contain 40% PVC and 60% filler material. PVC itself contains 56 8% by weight chlorine Figure 1. Incinerator system layout The stoker has rake fingers that push unburned garbage back toward the PCC b u rner The biomass feeding system has an educing blower, a main hopper, and 12inch auger and two side hoppers with 6-inch augers Each combustion chamber has a natural gas burner. A main blower provides air to the two burners A auxiliary blower provides air to the secondary chamber An external blower provides underfire air A butterfly valve at a "T" in the horizontal portion of the stack allows ambient air to cool and dilute the su ck gases The 100-fool Ull chimney provides sufficient draft for the combustion gases and the dilution air to maintain a slight negative pressure 1 (VOC) em between sampling sampling compour often en (mono)c I hazardoi NHW wa. runs is i any addt mg/kg f STATIST model fj is best d mean of indicate model fi include correiat for a mt 3 for th Table 1. Paie/ru P-PVC f  e over 1). filler 2041 In this work data from CCTL sampling runs where both HC1 and volatile organic compound (VOC) emission measurements were made will be analyzed to determine if any relationships exist between chlorinated VOC emissions and chlorine input. An EPA Method 0030 (17) volatile organic sampling train (VOST) has been used to measure VOC emissions from the same location as the HC1 sampling The VOST measurement procedure examines 43 compounds. Twenty-seven of these compounds are chlorinated. However, only a few compounds show up in sufficient amounts and often enough to be considered indicators of overall VOC emissions. Benzene, toluene. (mono)chlorobenzene, the dichlorobenzenes, and chloroform fit this criteria. Eighteen successful VOST runs were made while sampling for HC1 and burning just nonhazardous waste (NHW) or NHW and PVC. Six runs had PVC resin added into the incinerator while NHW was burning. Data for feed rales, temperatures. CO, HC1, and many VOC emissions for these 18 runs is shown in Table 2. The CO, HC1. and VOC emissions were normalized to the waste (including any added PVC) feed rate. Carbon monoxide emission is listed in (m g /h r C 0)/(kg/hr input waste), or mg/kg for short; HC1 is listed in g/kg; and the VOCs are listed in pg/kg. STATISTICAL PROCEDURES AND EQUATION SEARCHING A frequently used statistical quantity for determining how well a multivariate regression model fits a set of data is the coefficient of determination. R2 (181. The coefficient of determination is best described as a ratio of how well the data is fit by a model to how well the data is fit by the mean of the data An R2 value of 10 indicates a perfect fit to the data while an R2 value of 0.0 indicates that the model fits the data no better than its mean did. The R2 can be less than 0.0 if the model fits the data worse than the data's mean does. This can happen only if the model does not include a constant or intercept term For linear regression R2 is equivalent to the square of the correlation coefficient, r An R2 value of 0.500 is equivalent to an r value of 0 707. The correlation for a model is statistically significant at the 5% level if it exceeds the critical R2 value listed in Tablo 3 for the model's degrees of freedom (DOFs) (number of data points minus Table 1 CCTL PVC and HC1 Test Data Date/run NHW input lb/hr PVC input lb/hr HC1 output lb/hr Date/run NHW PVC inpuit in p u t Il b/h r l b / h r MCI output lb/hr go back and 12al gas he i the foot tall a slight p-PVC pipe (22.7% Cl). r-PVC resin (56 8% Cl) 2042 i Table 2. Hydrogen Chloride, Temperature,<Carbon Monoxide, and Volatile Organic Compound Data CO m g/kg 2.239 2.201 Z CO/100 dg/kg 2.211 1.862 2.704 2.814 2.224 0.615 .205 Toluene C6H5CH3 pg/kg 471.10 Benzene C6H6 pg/kg mi 9.8 9.96 10.630 14.480 20.660 2.165 2.301 2.266 2.2_ 21.:121275 2.237 CB DCB MC CF C6H5C1 C6H4C12 CH2C12 CHCJ3 mi mpg/kg 28.48 pg/kg Pg'kg pg/kg m N/A 8.01 m 4.9} 0.00 m N0./0A0 ' MB 1.16 48.01 S i 88.51 M .87 m 102.20 N/A 48.60 48.14 N/A 38.23 Note tp - (Tpcc*460)/1000 and ts - CTscc* 460V1000 W 32:?8 45.84 15.9S 12:^2 N/A 22,31 3100..4734 24.84 2t t f o 19.04 TCA CC13CH3 pg/kg N/A 42.16 6.02 1.73 .22.39 299.90 9.18 $ TCE C2HC13 Hg/kg N/A 1.162 0.000 N/A 2.350 Wa 8:888 um 0.0 00 34.770 0.000 3.043 9.936 N/A 7.948 >9.8 19.6 ' 4 m 746.4 Perc C2C14 pg/kg 0.01 0.000 4.954 m 0.000 % 0.000 num ber of param eters in the model) and its num ber of parameters. A P% significance level indicates th at th ere is only a P% likelihood that the data ended up along the lines of the model by random chance. A coefficient is statistically significant [18] at the 5% level if the range defined by the coefficient * (its l-value times its standard error) does not include zero, i.e., there is a 95% or greater probability th at the coefficient is not zero. The t-value is from the Student's t test for a 2.5% level (h alf of the 5% level, indicating a two tailed (*) test) and the model's DOFs. The t-values for 5% significance level are listed in Table 3. Table 3. Critical Values of t and R? at the 57 Significance Level (m * num ber of param eters excluding th e in tercept constant) [Adapted from 18-19], DOF t(5%/2) m -l m-2 m-3 m-4 m-5 9 2.262 11 2.20! 13 2.160 15 2,131 17 2.110 0.363 0.306 0.264 0.232 0.207 0.486 0.420 0.370 0.329 0.297 0.563 0.495 0.440 0.397 0.361 0.617 0.550 0.495 0.449 0.411 0.659 0 593 . 0.538 0.492 0.452 1UJ\vV> with a Iff Tfti | those yh strap) pjet written u naine.. Ji dopondep temper?*1 of ffpedo (Rr2). ffce interacts EFFECT OF F: output BP normalizj H with a p Ib/hr cop. coofficipi. HCl pu- 12.00 1 0 . p() 1B. 00 ,r> K 2r .0r0' 0.00 - ,0 Figure 2. 400-Jb/hr Data 0 000 4 954 3 497 N/A Q 000 0 000 2 262 5 509 0 000 2.424 0 000 0.000 9 048 N/A 0 000 el odel by efined by 55% or for a 2.5% ics for 57 Two strategies are useful in multivariate regression analysis. One strategy is to start with a linear model and add independent variables, one at a time, hoping to significantly increase R2. The other strategy is to start with many independent variables and throw out. one at a lime, those whose coefficients are least defined, until only well-defined coefficients remain. Both strategies are used here in the analysis of the CCTL CIHC data A computer program in BASIC was written to facilitate linear regression analysis of the CCTL data The program reads a file containing tho data shown in Table 2, with each column headed by the compound s or temperature s name The user specifies a dependent variable and the independent variables to which to fit the dependent variable. These variables can be made up of any function of the compounds and/or temperatures The program determined the number of valid data points (N). the number of degrees of freedom (DOF), the coefficients by least-square calculation, the coefficient of determination (R'2), the standard error of the coefficients, and the t-ratios of the coefficients The program is interactive, can plot the data and the fit on the computer screen EFFECT OF CHLORINE INPUT ON HCL EMISSIONS Figure 2 shows a graph of HC1 output plotted against added Cl input where both the- HC1 output and added Cl input are in Ib/h r and are normalized to a 4D0-lb/hr waste feed rate The normalization is necessary to present the data on a 2-dimension.il plot The linear fit is HC1 - 1.114 l b /h r 0 7478 Cl - 0 505 kg /h r 0 7478 Cl. (1) with a coefficient of determination. R2. of 0 735 (or a correlation coefficient, r. of 0 857) The 1 114 lb/hr constant indicates a 0 2785% chlorine content in the waste One would expect that the coefficient of the Cl term be near unity The 0.7478 coefficient implies that not all of the chlorine HC1 o u t p u t ( l b / h r ) Cl added (lb /h r) Figure 2 HC1 emissions compared to added chlorine input with both HC1 and added Cl normalized lo a 400-lb/hr waste feed rate 2044 ends up os HCl, or at least not immediately. When the unnormalized data is subject to a regression analysis with HCl output as the dependent variable and with NHW food rate and added Cl input as independent variables, th e resulting fit is HCl - 0.002647 NHW 0.6878 Cl (2) with R^-0791 (or r-0.889). This fit is reasonably close to the fit using the normalized data. These are strong correlations for sampling data, which is usually noisy. These correlations indicated that HCl is a good surrogate for chlorine level in the waste when comparing chlorine in p u t to C1HC output. REVIEW OF CCTL HCL AND VOST DATA Over 460 single- and m ulti-variable linear models of CCTL VOST data were subject to linear regression analysis. Dichlorobenzene data was also subjected to 5 n on-linear regression models. Selected models w ith coefficients of determination exceeding 0.22. with few and well defined param eters and w hich appear physically interpretable are listed in Tables 4 and 3- L in ja j Regression oJ CCTL VAST Data Using lin ear regression analysis, certain VOC emissions (see Table 2) were compared to PCC and SCC tem peratures (Tpcc andTscc), COand HCl emissions, and other VOC emissions. The emissions of the three dichlorobenzenes isomers were added together to reduce noise and rep resen t total dichlorobenzene (DCB) (CgH^C^) emissions. The most abundant chlorinated VOCs are usually the chlorinated benzenes. Table 4 shows the more significant results of the lin ear regression analysis of the CCTL VOST data. When com paring DCB emissions with only one other variable (Tpcc, Tscc, CO. HCl, benzene (benz) (CgHg), or a constant), DCB correlated best with HCl (see Table 4). Benzene and HCl were the only variables with w hich DCB correlated better than with the constant. Adding a constant as a second independent variable to HCl did not improved DCB's correlation with it, raising R^ to statistical significance at the 5% level (above 0.219 from Table 3 for 16 DOF and 1 coefficient excluding the constant). Adding benzene as a second independent variable to HCl brought the to over on e-h alf (or r to over 0.7), Adding a third, fourth, fifth, or sixth compounds as independent variables did little to improve the correlation. In all cases with HCl as an independent variable, the coefficient of the HCl term was well-defined (statistically significant at the 3% level, i.e:. its t-ratio was above 2.12 from Table 3). Also, the coefficient for HCl was positive in all of these cases, which indicates that increasing HCl emissions leads to larger chlorinated organic emissions. VKt 1 1-> 11 { contami) DichlOfp benzene HC aqW F thaj the ! were t^lso with W \ `. DCltMil wj ;;i! outlying! agaipsj a constant. Ta N JB DC DCB * 12.71 l-rallos; 6. N 18 DO DCB .1 1 .9 ' l-ratios: A l N -.18 DD DCB - 47.02 l-retios; l.t t-r-al io's;",I3*I8 n - 1, j>or DCB - 1731 * 7.157446E t-ratios: 6.7 t-ralios: |3DNCB - 2.3U7Q4/ t-raliof; .U N - 16 J>QF. CB - ) 17374 l*ranos: .p; CNDi .1&^.85W00 t-rallos; 3:4f C ir.^4 .2 o f` t*retios: 1.1) N - 16 DOF Ct-Ureti.o4s5;.6245, p1. . 'r . i. osson ul as (2) These ned that HC tnear dels. to PCC lissions ai the i&lysis zene e the a a R2 to ient 0, the ratio lieh Chlorobenzene (CB) (CgH^CI) emission data existed for 16 of the 18 runs. The other two had contamination problems w hich resulted in indeterminable values for the em issions of CB. Bichlorobenzene correlated very well with just CB (see Table 4). Adding a constant. HC1, an d /o r beozene improved this correlation only slightly. CB itself did not correlate as well as DCB did with, i HC1 and/or benzene, w ith o r without a constant term, though the correlations w ere still well-fit. Figure 3 shows the relation between DCB and HC1 emissions. It is ap parent from this graph that the highest DCB emission case is an outlier. DCB emissions without this outlying point (DCBNH) were also compared to Tpcc. Tscc, CO, HC1, Cbenzene, CB, or a constant. DCBNH correlated much bettor with HCL than DCB did (see Table 4), but not as well with CB and not at all with benzene. When DCBNH was compared to HC1 and CB togeLher, the R2 rose to 0.793 with w ell-defined.coefficients The chlorobenzene data also shows an outlying data point, w hich corresponds to DCB's outlying data point. Chlorobenzene without its outlying data point (CBNH) correlates nearly as well against a constant and HCl (R20.520) as DCBNH does (R2-0.586). When CBNH is correlated against a constant, HCl, and benzene, R2 raises to 0.625 (see Table 4). : Table 4. Selected Results of Linear Regression of CCTL VOST Data, N - 18 DOF-1 7 R`2 -0.43094 DCB - 12.78747 HCL t-ratios: 6.040355 N -18 DOF-16 R`2 -0.43842 DCB - 11.93067 11.64256 tfCL t-ralios: .4615735 3-534248 N - IS DOF-1 6 R`2 - 0 . DCB - 47.02865 .1187602 __ t-ratios: 1.865959 2.182596 f e 1-8 Ioo^ iV hCL11*V24?128fe4-02 BENZ t-ratios: 3.843409 1.631948 N -1 8 ,,DOF-15 R`2 -0.51229 DCB - 1.731203 9.932466 HCL 7.157446E-02 BENZ t-ratios: 6.71488IE-02 2.94955 1 50734 N - 16 DOF *15 R`2 - 0.86688 DCB - 1.172582 CB l-ralios: 13-57282 N - 16 DOF - 14. R`2 0.87694 DCB 2.374752 I 1-raUos: 1.06991 1CL *6 1. CB N -1 6 DOF-15 R`2 -0.37296 CB - 11.73746 HCL t-ralios: 6.03128 3 N -16 DOF-1 4 R*2 - 0.47377 CB - 8.850081 HCL 6.353128E-02 BENZ t-ratios: 3-467547 1.63/655 N -1 6 D O F-14 R`2 -0.42450 CB - 24.22029,* 9.315372 HCL t-ratios: 1.119753 3:213522 N- 1 6 DOF-1 4 R`2 -0.26619 CB - 45.64516 * 9.704254E-02 BENZ t-ratios: 2.159265 2.253371 DOF - 16 R`2 0.56711 S_c_B_N__H -T1O0..39; HCL l-ratios: 7.776< &BNH -Dl2F`j62?3 ^94 854?45H?L t-ratios: .8232941 4.60617 M - D& i o 42CRB 2 ' 0716% t-ratios: 9.196751 ,, R`2 -0.79317 DCBWk ->2.244332255 HCL * .6699222 CB t-ratios: 2.1*8--8558 3.038759 N -1166 a DOF -1144 R. .`.2 - 0.23870 CF - 7-000669922_*,2.058866, HCL L-ratios: .9555842 2.095115 N - 17 DOF ,,16 .R. `2 -0.22603 BENZ - 3.553316 DCBB1NH t-ratios: 3.,992200'989 18 DOF i/pC{P`2 0-59837 ?SCC - 1.6815 -- t- ratios: 175.0104, N - 15 DOF 13 R`2 *0.52028 CBNH - 2J.86j|2_* 6.934021 HCL t-ralios 47 3.75487 cV n !? 1d9o.8f3"26162 * R`2 -0.62495 5.62701 HCL * 4.365779E>02 B. ENZ t-ratios: 1.56476 3.052403 1.830047 N - 11 DOF 9 R*2 -0.36506 MCNH = 1.619011 ,,3._2_6_9_773 HCL t-ralios: .1362864 2.274755 N - 11 DOF- 9 R`2 -0.53558 CF -6.07818 * .4133894 MCNH t-ratios: 1.293594 322163 2046 DCB (p g /k g ) Figure 3. Comparison of DCB emissions lo HCi emissions with and w ithout outlying DCB data point. Most o ther VOCs were not detected enough times during the 18 VOST ru n s for statistical analysis. Even the most commonly detected VOCs have several aero-valued (not detected) data points. Chlorobenzene and carbon tetrachloride (CT) have 1 zero point each; m ethylene chloride CMC) has 2; 1,1,1-trichloroethane (TCA) has 3: chloroform has 4: trichloroethylene (TCE) has 5; and perchloroethylene (PERO has 10. Reasonable correlations may still be achieved with these data, especially if the zero points fo r one VOC correspond to another VOC, o r with low HCI data points. The 15 data points of CT (CCl^) do no1 c r relaie with HCI and a constant (R2-0.000). The 12 m ethylene chloride (MC) (C^Ctg) data points do not correlate w ith HCI and a constant, but one value is an order of magnitude higher than the re sta n d is probably the result of contam ination, since MC is frequently used in chemical analyses where the VOST traps were analyzed. MC without that outlying point (MCNH) correlates somewhat with HCI and a constant (R20.3&5). Chloroform correlates well with a constant and MCNH (R2 0.536)- The 16 data points of perchloroethylene (PERO (CgCi4) do not correlate with HCi and a constant (R20.007). Quadratic and Cross. P ro d u c ts g re s sions of CCTl VO-SUPai^ l When DCB emission is compared to the square of any other one variable (see Table 5), the best correlations are w ith CB2 and HCI2, All fits with other independent variables w ere worse than with the unit power of the same independent variable. The correlations improve wiih additional squared terms, reaching R2-0.936, with HCI2. CB2. and constant terms. All th ree coefficients here are well-defined, and the HCI2 coefficient is positive. Replacing the HCI2 term w ith an HC1*CB term slightly raises Lhe R2 to 0.940. Correlation of DCB with benzene, benzene2, and HCI*benzene produces R20.699, with all th re e cot HClibenz coefficjoi incfcasin v*I HC11pop z HCL?benz dofiped. 1 rospltod ii (r-0.88I). I ' M (sec Tablf constant parameter agaipst 0 defined. R20.1192' .T both CTVf against al R2-0.573. wcll-defir leaving ft Table 5. IS T DCB - .76: irratios: < t-ratios; 9 N * 18-^Pi DCB - 45:4 i-ratios; 2 N * 16 Vpi DCB i 36. t-ratios, a f e J6 .34; t-ralios: 3 N - 16 BCD * 28. * 3 691222 1*ratios; * N 16 DCB * 29 CIT2. t-ralios: 1)CBNH three coefficients well-defined and; the HCl*benzene coefficient positive. Replacing the HC1*benzene term with a benzen/HCl term slightly raises the R2 to 0.729. Again all three coefficients are well-defined. The benzene/HCl coefficient is negative, w hich still indicates that increasing HC1 emissions leads to larger chlorinated organic emissions. When CBNH is correlated against all quadratic terms of HC1 and benzene (HC1, HCl2, HCl*benzene,benzene2*benzene, and a constant), the resulting R2 is 0.836. The coefficient for the HCL*benzene term Is most defined (had thehighest t-ratio), and that for the HCL term is least defined. Successively dropping the least defined term until only well-defined term s rem ain resulted in a form containing only the constant and the product HCl*benzene, w ith R2-0.776 (r-0.881). When DCBNH is correlated against a constant and HCl*benzene, the resulting R2 is 0.6B5 (see Table 3). Since DCB has two chlorine atoms p er molecule, DCBNH is correlated against a constant and HCl2*benzene. w hich resulted in R2-0.816 (r-0.903), the h ig h est R2 for a two parameter model so far. The coefficients are very well-defined. DCBNH also correlated v ery well against a constant and HC1*CBNH, with R2-0,870 (r-0.932). The coefficients are again v ery well- defined. When DCBNH is correlated against a constant. HCI2*benzene, and HCI*CBNH, the result Is R2-0.892, though the coefficient of the HC1*CBNH term is not well-defined. When CT was correlated with a constant and HC1*CF, the result for the 13 data points with both CT and CF data was R2-0.176. Adding an (HCL*CF)2 term raised R2 to 0.477. A fit of CT data against all quadratic term s of HC1 and CF (a constant. HCL HC12. HCl' CF, CF2, and CF) produced R2-0.573. Again the strategy of throwing out the least well-defined term s o n e'at a time until only well-defined term s rem ain is used here. In this case the constant and HC1*CF term s can be dropped, leaving four well-defined coefficients and R2-0.567. Table 3. Selected Results of Quadratic and Cross Product Regressions of CCTL VOST data. N * 18 DOF-1 7 R`2 -0.24015 DCB - .76319 HCL'2 t-ratios: 4.80169 N - 16 DOF-1 5 R*2 -0.76221 DCB - 5.109778E-03 CB`2 t-ratios; 9.824875 N - 18 DOF - 16 R;2 -0.42562 DCB - 45.49169 5721405 HCL'Z t-ratios: 2.272987 3-443274 l-ratios: 3 63494 ."6.290484 N -15 DOF-1 3 R*2 -0.77618 CBNlf- 40,96126 9.305392E-03 HCL*BENZ t-ratios: 3 856131 6-714425 N - 17 DOF 15 R`2 - 0.68548 DCBNH 39.27749 * 1.205203E-02 HCL*BENZ t-ratios: 3-92413 5-717672 N - 17 DOF - 15 R`2 - 0.81574 DCBNH - 43.63743 * 7.3228551-04 HCL`2*BENZ t-ratios: 3-94337Z 8.148909 BcBl-6 .3 4 9 it*?/4.& 0254E -03 CB'2 t-ratios: 3-415531 7,983147 3691222E-Q3 CB`2 t-ratios: 3 647856 3-519627 9.678116 N -16 DOF - 13 R*2 - 0 93962 DCB 29.98737 .041425 HdL*CB 2.454528E-03 t-ralios: 3.969738 3-735022 4.058922 N - 17 DOF 13 R*2 - 0.72512 DCBNH 34.97016 - .5006529 HCL'2 Hcbi -^f.704l3 .R623688i e?2 HCL*CB t-ratios: 368109 9.309327 DCB& -D2?32535 ^ . ^ l l ^ L ^ C B N I I -589553 N - 13 DOF-10 R*2 - 0.47732 . CT - 10.65122. * .112461 HCL*CT t-ratiost 3^538813 2.698114 2.401374 CF*-1^.312712* 0338^138^$ ^MCNH l-ratios: 1.872917 3-917113 2048 Chloroform correlated v eil with a constant and HCl'MCNH (R2-0.630), but only 11 data points are shared between CF and MCNH. A fit of PERC data against all quadratic term s or HC1 and TCE produced R2-0.760. The constant and TCE terms can be dropped, leaving four well-defined coefficients and R2-0.709. Nonlinear Least-Squares Fits In a phenom enological study of data from the Pitlsfield-Vicon in cin erato r 121 Green el al. 13-5) fit total chlorinated dioxin, furan, benzene, and phenol emissions with the equation Y aXexpCn X) b Xexp(mX) F(t^.p) Zq (3) where F(t;p,p) ft/ l ))* e x p (p /1 - p/l). (4) Here Y is total emission of the group of compounds of interest at the tertiary duct in (pg compound ouLpul)/(kg waste input) o r pg/kg for short, X is HC1 emission in g/kg, t is absolute tem perature in the duct in R (Rankine) divided by 1000. and Z is COemission in m g/kg divided by 100. The first term In Equation 3 represents baseline emissions with ideal combustion conditions (no CO). The second term rep resen ts the effect of the temperature and CO on emissions. The tem perature, HC1, CO, and chlorinated emissions were measured at the duct after the Vicon's secondary combustion chambers. A n o n lin e a r least-squares regression program. NLF1T. was used to determ ine parameters when modeling the CCTL PCB data with nonlinear equations. With this program , the user can specify th a t certain param eters remain fixed while the program searches on the other parameters. In the previous analysis of the Vicon data 13-81, the value t - 2 was chosen as a divisor in Equation 4 since it was n e a r the average value of Lfor the Vicon's tertiary duct tem perature. In the CCTL data, the average for t in th SCC is 2.360. The F(t;p,p) function in Equations 4 acts somewhat orlhonorm al to Z** and the function of X and has values near 1. w hich makes the b param eter less sensitive to changes in p and p. The first term in Equation 3 was dropped in this work (i.e. a - 0) sin ce, with thB small range of tem peratures in the CCTL data, the distinction between the first and second terms was not discernable. With m>0, p*l, and q0, all fixed, the NLFIT program iterated on param eters b and p and produced R2*0.432, w hich is not significantly belter than R2 <0.431) for a lin ear regression of DCB upon just HC1. Letting q v a ry Lraised R2 slightly to 0.443- Letting m vary as well did not increase R2. When m, n, p, and q were varied an excellent R - 0.701 was achieved. However in this case the parameters p and p had unreasonably large absolute values. To determ ine if another, function could also represent F(t:p,p), but without the unroesonablo param eter values F(t;p.p) was plotted against l for these large param eter values. The resulting curves appeared Gaussian (shaped like anorm al bell curve). This suggested the function J(t;u,s)-exp(-((t-u)/s)2/2 (5) . i TV ", ,T program R2r0.7p|. 0.651. Fh mottei. ^ :1 Z). :Th?,f Tahlp 6 / ; pcb ; DCBNII DISCUSS!j K iak-I Tj c; ` Thi?? i f i \ j. w1 hose raii.t pi ' '* JThis Slight Ofi.i y* Ml ' % \ : The rpte'o. con nd TCE t al. )Und *e in st le IC1. CO. meters eters. in in the hat less 0) and I DCB se R2. 2049 - \ f fi t-*1 *:V ,' t.!`( -. t could replace F(t;p,p). Equation 3 was then rew ritten as (Gaussian Model) Y -b X exp(mX) J(t;u,s) Zq, (6) The param eters u and s were estimated from the plots for initial values for the NLFIT program. Allowing the NLFIT program to vary b, m, u, s. and q, the program again produced R2-0.701. The param eter m was not well-defined here as well. When m was fixed to zero, R2 became 0,651- Fixing q a t zero but letting m search gave R2 - 0.694. Table 6 gives the results w ith this model. When the outlier is excluded the widLh parameter s increases to a more satisfying level, To test a different model of the dependency of DCB on COemissions, 2^ was replaced by exp(w Z), The resulting R.2 and param eters were nearly identical wilh those for the model withZ^. . f *;i Table 6. Param eters for Nonlinear Equation (Gaussian) Fits to CCTL DCB data. bm us q DCB 15.1 0.137 2.32 0.0157 0 49.0 0.0778 2.32 0.0214 -1.280 64.0 0 2.32 0.0256 -0.771 36.7 0 2.32 0.0222 0 DCBNH 12.2 12.0 11.5 0 0 0.0379 2.41 0.152 2.42 0.159 2.76 0.350 -0.0208 0 0 R2 0.694 0.701 0.651 0.587 0.594 0.593 0.640 DISCUSSION Kinetic-Type Fits The best lin ear regression formula for CBNH (Table 5) has the form CBNH - A B HCl*benzene (7) This is suggestive of a kinetic model of chlorobenzene formation from benzene: C6H6 HC1 --> C6H5CU H2, (8) whose rate of reaction is dependent on the product of the HCI and benzene concentrations. Dichlorobenzene can be formed from benzene by reacting it with 2 HCls: CsH6 2 HCI C6H4C12 2 H2. (9) This suggests a formula to fit the DCBNH data to HCI and benzene which also is v ery good: DCBNH A B HCl^benzene. ( 10) Dichlorobenzene can also be formed from chlorobenzene by reacting it w ith HCI: c6h 5c i * h c i~ > c6h 4ci2 * h2. n i) The rate of this reaction is dependent on the product of chlorobenzene concentration and HCI concentration, This suggests a model correlating DCBNH data with CBNH and HCI w hich is even " ]* /, jS hi f i: 2050 DCBNH A B HC1*CBNH. ( 12) The fact that the data are Tit well by these models makes a chemical kinetic description pJausable. An Attemn la t Reconciliation of CCTL and P-V results It is im portant to compare our CIHC emissions with other results reported in the literature, particularly the data set reported in the widely quoted Piusfield-Vicon (P-V) study. Since the CCTL m easurements are made after the secondary combustion cham ber (SCC) we concentrate on emission factors after the SCC as listed in the P-V reports. The P-V data sets include hydrogen chloride (X HC1), chlorobenzene (Yj CIBz), chlorophenols (Yg ClPh), chlorodioxins (Y3 PCDD), and chlorofurans (Y^ - PCDF) emissions. The P-V study found good linear correlations between Y3 and Y4 (r*0.91). Y j and Yi (r-0.50), and Y^ and Yj (r0.53) Thus it is reasonable to use chlorobenzene as a surrogate for PCDD/PCDF 1 The P-V lin ear regression (Y - a*bX) analysis for the emissions at the exit of th eir secondary combustion cham ber gave correlation coefficients with X for PCDD (r-0.30), for PCDF (r-0.25) for CIBz (r*0.11) and for ClPh ( r --0.03). On the other hand the CCTL lin ear regression results with respect to X give for CB (r - 0,652) and CBNH (r *0721) and for DCB ( r 0.662) and DCBNH (r 0 765). When X is replaced by XB (B - benzene), we obtain even belter correlations for CB (r 0.726), for CBNH (r 0.881). When X is replaced by X^B we obtain for DCB (r - 0.686) and for DCBNH (r 0.903). As an alternative to the P-V's use of linear regression we have fit the P-V data sets with many analytic functions Yj(XXZ) where X HC1 emission factor, T - absolute tem perature at the exit oi the secondary cham ber, and 2 - CO concentration. By adjusting the param eters in these functions using the n o n -lin ea r least square fitting routine we found that the functions which did best (such as Equation 3) showed an increasing dependence upon HC1 (4,8), Thus o u r analyses of the P-V data is in general agreem ent with our analyses of the CCTL data and contrary to the P-V conclusions. The d iffering conclusions as to the interpretation of the P-V data may be accounted for by. (1) The general noisiness of toxic emission measurements at the parts per billion level. Here It should also be noted th a t w hereas the CCTL emission factors are a factor of about 10 lower than those of typical California medical waste incinerators, Lhe P-V emission factors (apart from HCL) are . about a factor of 10 below the CCTL's emission factors. Thus the P-V signal to noise ratios for CIBz measurements are probably lower than those of Lhe CCTL. (2) There were marked variations of Lhe P-V input waste between trial burns whereas the CCTL's waste input has been quite homogeneous. (3) There were marked variations of the P-V combustion conditions, as indicated by th e ir much . wider range of tem peratures; and carbon monoxide levels. (4) The P-V study gave a limited number of data sets collected at th eir SCC exit (7 pairs plus one unpaired set). (5) The P-V study had a very limited num ber of SCC data sets in which PVC levels were varied (one pair). (6) For the P-V PVCfree set the in p u t waste was heavily contaminated with chlorinated dioxins and furans. Thus the PV experiments, while very useful for many purposes, were not suited for finding correlations between PVC inpuLs and chlorinaLed aromatic oulpuls. In contrast the CCTL experiments used relatively homogeneous sets of non-hazardous waste ps the p non-spi burns w slfpng t pq pyc % CONCJ-p: - I* hTM * 1 dpende ftp %- *. conclus in ch|q. pqlysly. lhe CCT HP? fcpfl noi sufi and leaare in 1 one cop ch|ofpb S r 2;5 ( iM, .Vi Hp3i 4t1^?;^` y herei Yw y f r n \1%P^t1;.:toxic, ep cfjlorjn BtyS?j0. ' P I able ilure. i CCTL mission (X- ^ and fcene as condary for h 0,765); for 1.903). ith h e exit :h did es of P-V or by. >it in those re CIBz of the leous ich lumber very 3VC- . i the P* is s waste 2051 as the prim ary inputs apart from the small spiking percentages of PVC. The 6 spiked runs and 12 non-spiked ru n s gave 18 ru q s with chlorine ranging from 0.2 to 1,5% by weight. The CCTL trial burns were almost always under good combustion conditions (low CO, high Lemperatures). Thus the strong monotonic correlations found between the CCTL Yj$ and X reflects a greater concentration on PVC related effects in the CCTL study. , CONCLUSIONS The main conclusions from this study of CCTL's VOST, HC1. CO and tem perature study are; (1) there is a d irect dependence of HC1 emission levels on the level of PVC in the waste; (2) there is a strong dependence of C1HC (notably DCB and CB) emissions on HC1 emissions; (3) th ere is a strong dependence of DCB and CB on each on other and on products of HC1 and aonchlorinated benzene; (4) and there is almost certain evidence of nonlinear relations existing between C1HC emissions and CO., and temperature. In nearly all of the different regression analyses performed here, the relationship between a chlorinated organic emission and HC1 emission was positive and welldefined. These results, contrary to the prevailing opinion [11, lead to the physically reasonable conclusion that decreases in the levels of organically bound chlorine in the input leads to decreases in chlorinated organic emissions. Our best regression equations suggest that compounds with benzene rings, such as polystyrene, should be avoided particularly when chlorine is present. Here it m ight be noted '-that the CCTL plastic input (39%) was mostly polystyrene. Many models, including our kinetic models, use benzene as a precursor to chlorinated benzenes and PCDDs/PCDFs. If Lhe benzene molecules are not sufficiently destroyed in the incineration-process, they can be chlorinated by the HC1 present and lead to more complex aromatic hydrocarbons. In this regard the CCTL study and the P-V study are In full agreem ent thaL good combustion is of major importance. It should be noted that the CCTL used VOST to measure volatile organic compounds while Modified Method 5 (MM5) (201 was used at P-V to measure semi-volatile organic compounds. Only one compound of interest, dichlorobenzene, overlaps both data sets. In the.jP-V data the total chlorobenzene (sum of di- through hexa-chlorobenzene) emissions were.-on average, 2.5 limes as great as the dichlorobenzene emissions. The CCTL dichlorobenzene emission data can be multiplied by 2.5 to approximate total chlorobenzene (TCB) emissions. One can relate TCB to HC1 by TCB-AHCl2, . (13) where A - aB and B is average benzene emissions (in pg/kg). For the CCTL data A-l .9 and a-O.OOl 1 .while for the P-V data A0.049. Since the CCTL waste was heavily loaded in polystyrene, while the PV waste was probably not, the benzene emissions from CCTL may be many times g reater than ihusc from P-V, and the param eter a for P-V may be close to that from CCTL. In final summary the CCTL's experimental, phenomenological, and theoretical studies of toxic emissions from incineration all support the physically intuitive hypothesis that reduction of chlorinated plastics in the input waste stream results in reduction of aromatic chlorinated organic emissions 1211. While the CCTL's measurements have been limited Lo volatiles and lig h t semi volatiles these result are expected to ap'ply to other chlorinated aromatic hydrocarbons emissions % ' . ^ f jL '^ 'V I\ f t-' r, I,- - I*' j Vi* }r> ' t- V 2052 t such as phenols, dioxins, and furans, which we have not measured. A num ber or other publications support a PVC-PCDD association 122-26). Bulley has also found experim entally that reduction of PVC input leads to reduced chlorinated dioxins and furans (27). Thus we are convinced that, when all oLher factors are held constant, there is a direct correlation between input PVC and output PCDD/PCDF and that it is purposeful to reduce chlorinated plastics inputs lo incinerators. On th e o th er hand it must be acknowledged that signal to noise problems and insufficient data sets limit the sharpness of the conclusions that can be reached from the CCTL data set or any other data set now in the literature. While the CCTL has acquired much VOST data in the 1 to 10 g HCl/kg waste range w hich should be applicable to municipal waste problems, more data is needed in the 10 lo 20 g /k g range to better determine relations between chlorinated organic emissions and HC1 emission from medical waste incineration and possibly even a h ig h er range for toxic waste incineration. More and better data is also needed at lower HCI values to understand the impact of pollution prevention or precombustjon measures on C1HC emissions since here the noise in the data itself makes determ ining various correlations particularly difficult. Many more data runs are needed with an expanded tem perature range (1300 to 2100 *F PCC tem peratures ), various CO levels and possibly various benzene levels to properly determ ine the effects of tem perature and CO on ClHC emissions. For good ranges of HCI, tem perature, COand possibly benzene levels some 30-100 runs are probably needed to chart the dependence of the ClHCs on these three or four variables. With the hope of carrying out this many ru n s at a reasonable cost we are attempting to.simplify and extend the measurement of volatiles into the domain of semi volatiles. Our successful correlations with only IS or so VOST data sets suggest that with 30-100 extended data sets we can gain an insighL into Lhe basic relationships which govern toxic emissions from incineration. ACKNOWLEDGMENTS The au th o rs wish to thank the. Mick A. Naulin Foundation, the University of Florida Gatorade Fund, an anonym ous donor, Supelco. Tacachale, the University of Florida, Tennessee Valley Authority, Florida Governor's Energy Office, New River Waste Association, and the Flrida Department of Environm ental Regulation,for various forms of project support. They also wish to thank Bruce Green for refurbishing and maintaining the incinerator and helping install the control system.Charles Schmidt for VOST analysis; Kaifa Awuma for HCI analysis and.Don aid Clauson, Joseph Blake, Xie-Qi Ma, Mahadcvan Sadanandan. Henk van Ravenswaay, Jonathan Carter. Chris Jeselson.Britta Schmidt, Jason Weaver, Scott Quarmby. Robert Driskell. David Roskein, Donald Adams Todd Y urchisin, Ronald Slorer, William Calhoun, Thomas Cherry, Song Mu. Wilfred Schnell, Rodnie Barbosa, and others who helped In various other ways over the past 3 years. calions of PVC i all tient any 10 g rCdcd in and le Cl of he data F PCC Iho ut o nues liti COSI 11* J jaiorade sii to ine Carier, Donald e h nell. 2053 REFERENCES ) R.S. Magee, Plastics in31ur.iciEai.SaUd ffasieJnrineraUon: A Literature Study., prepared by the New jersey Institute of Technology Hazardous Substance Management Center for the Society of Plastics Industry, Inc., January. 1989 2 R. Meulicht. Results of ine Combustion and Emissions,Research Project at the Vican Incinerator Facility-in Pittsfield, Mass. Final Reran, Volume I. prepared by Midwest Research Institute for New York Suite Energy Research and Development Authority Report 87-16 June. 1987 3 A. Green. J Wagner. J. 3ial:e. and X-Q. Ma, ' Strategies For Tonic Minimization when Cofiring with Biomass*', 1991 Southern Biomass Conf. Paper IV B2, Baton Rouge. LA. Jan 7-10. 1991 4. A. Green. J, Wagner. C Saltici, and M Jackson. Pollution Prevention and Institutional Incineration". Proc the ASME Sciiti Wasu? Processing Conference. Detroit. MI, May 17-20. 1992. 5. A Green, j. Wagner. C Saltici, j Blake, and X-f). Ma Medical Waste Incineration with a Toxic Prevention Pnuotol", Pi oe n d in a;, of.t he.$4th_An ftua i ,Mc_oUn k oLt h e.Ain & . aste.. Management Association. Paper 91 535 Var.wiivcr PC. June 16-21. 1991. 6 A. Green. C Batich J Wagner, el al `Advances in Uses of Modular Waste to Energy Systems". Published in ASME booklet ,Advances in Solid Fuels Te chnologies. FACT-Voi 9, Voi. G00522, AES Green and WE. Lear (eds ). 7. A.E S. Green, J Wagner, and S Mahadcvan "Chlorinated Toxics from Incineration ' presented at the ASME Joint Power Generation. Conference Atunla. GA. October 18-22. 1992. 8 a.E.S. Green, J.C. Wagner, and K j. Lin. ' Phenomenological Models of Chlorinated Hydro carbons". Chemosnhe r e . Vol 22. Nos 1-2. pp 121-135 1991 9 A. Green and B Green. Development of Gas-Coal Combustor", ( 1981-), U S. PalenL No 4,561.36-1. `Method of Iietrofiting An Oil-fired Boiler to use Coal and Gas Combustion". December 1985- 10. A. Green, J. Wagner, B. Green, et a!.. `Coai-Water-Gas. An AJi American Fuel For Oil Boilers". Proc. Eleventh Int. Conf on Slurry Technology. Hilton Head, SC. pp 251-262, March 16-18. 1986 II A. Green. J. Wagner, B Green and J. Schwartz "Synergisms in Coburning Gas and Coal". ASlME Paper 87-jPGC-FACT-iO Join ASMF. 'IEEE Power Gen. Conf , Miami Beach. FL. Oct. d-8 1987. 12. A. Green, j. Wagner, B Green, et al . "Radiation Enhancement by Coal Slurries". Proc of the Twelfth lntnl Conference on Slurry Technology. New Orleans. LA. March 31- April 3. 19S7 3- A.E.S. Green. B A.S Green, and jC Wagner. "Radiation Enhancement in Oil/Coal Boilers Converted to Natural uas". US Patent No 4.978 367, Dec 1990, and No 5 8 . -d33, Sept, 1991 H. A Green. G. Prine. D Roekwood. ct a!.. `Co-Combustion in Community Waste to Energy Systems", Co-Combustion. ASMF FACT Vol 4, pp l?.-2S 1988 2054 15 A Green j Wagner b Green el al u.-Cuinl-ustior. of Wi-Mc Biomass ind Natural Gas linin'!ass 20. pp 249-262 1989 16 Method 26- Determination cl liydmgen Chloride Emissions 1i-m Stan nary Sources i ederai p r .i.-.v i Yul 54 No 24.' pp 5226) i22**/ Deceinbei 2i> 198 17 Test Metnods for Evaluating Syjid Wa-dt SV-M6 3rd Ed Meil. t.: s U 'A Wash DC IS Kl. Scheafier and James T McChve S ^ u ^ c s fo^lm.^aLs.rs Puxbury Press Boston 1982 19 j ' Wag lie; Coin lu s t ion a nd P r t i >niLuvnon < in tre j 6ieib.pdLii2-yiL`im.`l L j j 3 . lS^i0P f fffP Mcd>)1a1 i p i.111 e ra t>'r s Phi; Jusserial ion lopt of Mtch Fngr University of I iorida Aug 1992 26 j..ft Mutii>mI> f,.r f v i 11a11j\g Wasps: SW-S46 3rd CJ Meth 001 0. V) S EPA Wash DC 1%' 21 A r S Giecn and JC. Wegner. To*.c Pn duiii of Medical Waste inrineration Chapter 1 in Medical Viasto Incineration end Poliuyipji PuAentoji. A Green <cd > Van Nostrand Reinhold New York. NY. 1992 22 11 Vogg. M Metzger, and 1. StiogJitz P-.trni Findings on the Formation and Decomposition 01 PCMj.PCDI m Solid Municipal Wat t. 11.. n.- ra.>< .. SptMalizid Sem.i.ai on F.mssion of Trace Grgar.it* from Municipal Waste In cmer.11.01 ( ->jo nr.zgen Dtnmail .hi. 2f-22 198" 2?. W Christman n et al Combusti n ! Polyvinylchlor Me An Imp;-r tan t Source for the Formal!.>n of I'CDD'PCDF CfiejHiiFPiliI.L c>! l '1 Nos !-b pp 3b.- 392 19S8 24 UK Uulled K ft Bruce I. 0 and i'cac J. loi f;,.n,.*n >)" Cniorinutco i rg.u.ics During Solid \i uste ( i inbust ion Waste Manaatin.r rl 11 <! Pc search. Vol 8 26 ? ` -t 199(5 25 WM Shaun and W Tsang Diokiii .Mr.r..lion in Jr.* .*eialt rs hr v won Sci 1echini! Vol 17. pj> .<'2 1-730. 1983 26 V' M Shaub and W Tsang `Civersit-v oi Dm/in Formation in Gas and Solid Phases Under Municipal In c in e ra to r Conditions in ch lo rin ated Dmxins U pibe n j o f y r a n ?_i n. the. Total I. 1J Keith r Pappe an 1 (. Cnc..idltary ?tds ) huitervorih Stoneham MA. pp 469- L ' 6l M IiuJiej Medical islv Manage in` fit in /.utiiol ihifi P.Ct 2 I Ghflp .*> Chcmosplicrc.' Printed in Grca