Multivariate Analysis of Exhaust Emissions from Heavy-Duty Diesel

Michael SjÖgren, Hang Li, Ulf Rannug, and Roger Westerholm*. Department of Genetic and Cellular Toxicology, Wallenberg Laboratory, and Department of ...
0 downloads 0 Views 318KB Size
Environ. Sci. Technol. 1996, 30, 38-49

Multivariate Analysis of Exhaust Emissions from Heavy-Duty Diesel Fuels MICHAEL SJO ¨ GREN,† HANG LI,‡ ULF RANNUG,† AND R O G E R W E S T E R H O L M * ,‡ Department of Genetic and Cellular Toxicology, Wallenberg Laboratory, and Department of Analytical Chemistry, Arrhenius Laboratory, Stockholm University, S-106 91 Stockholm, Sweden

Particulate and gaseous exhaust emission phases from running 10 diesel fuels on two makes of heavyduty diesel engines were analyzed with respect to 63 chemical descriptors. Measurements for one of the fuels were also made in the presence of an exhaust aftertreatment device. The variables included 28 polycyclic aromatic compounds (PAC), regulated pollutants (CO, HC, NOx, particles), and 19 other organic and inorganic exhaust emission components. Principal components analysis (PCA) was applied for the statistical exploration of the obtained data. In addition, relationships between chemical (12 variables) and physical (12 variables) parameters of the fuels to the exhaust emissions were derived using partial least squares (PLS) regression. Both PCA and PLS models were derived for the engine makes separately. The PCA showed that the most descriptive exhaust emission factors from these diesel fuels included fluoranthene as a representative of PAC, the regulated pollutants, sulfates, methylated pyrenes, and monoaromatics. Exhaust emissions were significantly decreased in the presence of an exhaust aftertreatment device. Both engine makes exhibited similar patterns of exhaust emissions. Discrepancies were observed for the exhaust emissions of CO2 and oil-derived soluble organic fractions, owing to differences in engine design. The PLS analysis showed a good correlation of exhaust emission of the regulated pollutants and PAC with the contents of PAC in the fuels and the fuel aromaticity. Also, the emission of soluble sulfate was directly related to the contents of sulfur in the fuels. In conclusion, the PCA clearly indicates that emissions of PAC, particulates, and sulfates are the most descriptive exhaust emission factors. These exhaust emissions can, as shown from the PLS analysis, be much reduced by a decrease of aromatics, PAC, and sulfur in the fuels.

Introduction When diesel fuels are combusted, a vast number of components are formed and emitted with the exhausts (1, 2). Some of these products are considered by the International Agency for Research on Cancer (IARC) as being probably carcinogenic to humans (3). IARC also classifies diesel exhaust emissions to be probably carcinogenic to humans (4). The compounds present in these emissions comprise regulated and unregulated pollutants. The pollutants that in many countries (5, 6) are regulated by law include carbon monoxide (CO), nitrogen oxides (NOx), hydrocarbons (HC), and exhaust particulates. The amount and chemical composition of exhaust emission is related to the chemical composition of the fuel (7-14), including the presence of fuel additives (15), but also to the efficiency by which the fuel is combusted. Thus, the selection of fuel, the choice, tuning, and running of the engine as well as the type of exhaust aftertreatment device (16-18) are important factors for reducing the adverse effects of diesel exhaust emissions (19). To minimize these effects, it is important to characterize the underlying factors. In an earlier study (20), eight diesel fuels and associated diesel exhaust particulates (DEP) were analyzed using principal components analysis (PCA) and partial least squares regression (PLS). The most influential variables with respect to emissions included fuel density, final boiling point, total aromatics, and PAH. These variables were also shown to have an impact on the biological responses studied, the mutagenicity in Ames Salmonella test, and the ability of the extracts to inhibit the specific binding of radiolabeled TCDD to the Ah receptor. With more extensive chemical analysis data for the same set of fuels, it was later shown that mutagenicity, in the absence of metabolic activation (i.e. , -S9), correlated well with the contents of 1-nitropyrene (21). With S9, a higher correlation was observed with PAH, specifically with fluoranthene and anthracene. The correlation between PAH in diesel fuels and PAH in the exhaust emissions have previously been studied by Westerholm and Li (22). Using linear regression, PCA and PLS, a significant correlation between PAH of the fuels and PAH in the exhaust emission was observed. The results indicated that a reduction of PAH in the fuel, from 1 g/L to 4 mg/L, would decrease PAH in the emission by 80%. These results corroborated results previously observed by others (7-9). As new data became available through the inclusion of three new fuels, one fuel (denoted D6) was shown to be significantly higher regarding PAH than the other fuels. This outlier was therefore excluded from further studies. Using PCA, the 10 remaining fuels were described in terms of physical and chemical fuel parameters (23). Results showed that these fuels could physically (26 variables) be described in a few general terms, e.g., viscosity, density, and cetane number. The most descriptive chemical factors * Corresponding author fax: +46-8-156391; e-mail address: [email protected]. † Department of Genetic and Cellular Toxicoloty. ‡ Department of Analytical Chemistry.

38

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 30, NO. 1, 1996

0013-936X/96/0930-0038$12.00/0

 1995 American Chemical Society

TABLE 1

List of Fuel Variables, Notations, and Measured and Calculated Values per Fuel variable

fuel unit

cetane no. density at 15 °C viscosity at 40 °C initial BP (IBP)a 50% BP (BP50)a final BP (FBP)a dist. PC1 (≈BP40)b dist. PC2 (≈BP95)b dist. PC3 (≈FBP)b dist. PC4 (≈IBP)b dist. residue CF point cloud point flash point specific energy sum of 14 PAH sum of 27 PAH nitrogen sulfur ethylhexyl nitrate water aromatics olefins naphthenes paraffins benzene toluene

g L-1 mm2 s-1 °C °C °C

% °C °C °C MJ L-1 mg L-1 mg L-1 mg L-1 ppmw wt % ppmw vol % vol % wt % wt % mg L-1 mg L-1

abbr

D1

D2

D4

D5

D7

D8

D9

D10

D14

D15

fp1 fp4 fp5 fp6 fp12 fp18 fdp1 fdp2 fdp3 fdp4 fp21 fp22 fp23 fp24 fp26 fc28 fc29 fc34 fc35 fc36 fc37 fc38 fc39 fc41 fc42 fc44 fc45

52.8 811.7 2.11 220 236 261 1.59 -2.65 -0.19 0.02 1.3 -40 -34 87 35.1 0.73 1.57 0.21 100 0 15 2.7 1.4 53.6 45.6 85.5 215.6

50 821.3 2.11 223 239 260 2.00 -2.70 -0.11 -0.03 1.8 -39 -35 87 35.4 0.94 4.12 0.34 100 0 20 14.7 2 38.8 45.5 4.9 7.4

47.2 832 2.09 221 241 261 2.29 -2.63 -0.12 -0.07 1.4 -39 -34 92 35.7 0.73 1.39 3.9 2900 0 70 22.5 2.2 29.6 43.8 10.7 31.9

47 831.3 2.26 190 248 323 2.34 1.29 0.74 -0.15 1.2 -32 -24 75 35.7 101.8 340.7 29.2 200 0 50 26 1.6 27.2 47.6 30.7 1225

44.7 808.3 1.41 180 206 300 -3.78 -1.08 0.36 0.17 1.7 -40 -40 64 35 47.64 231.6 14.3 200 0 60 19.8 0.9 32.9 48.5 7.8 59.5

55.7 808.7 1.44 176 204 299 -4.18 -1.15 0.23 0.18 1.6 -40 -40 64 35 41.05 183.5 207 100 0.2 73 19.4 0.2 32.9 47.7 7.9 72

52.8 813.2 1.96 175 231 301 -0.02 0.71 -0.13 -0.24 1.5 -40 -40 75 35.1 106.5 313 11 100 0 58 16 0.7 25.5 58.1 5.9 299.9

nac 817.5 1.7 133 219 323 -2.79 2.38 -0.30 -0.46 1.5 -40 -25 48 35.2 0.8 0.8 0.13 7 0 34 20.7 1.8 32.5 47.1 na na

50 814.3 2.02 164 236 307 -0.21 1.91 -0.63 0.33 1 -39 -37 59 35.2 1.2 1.2 na 1 0 12 7.7 na 50.7 44.9 na na

56 817.2 2.57 172 257 337 2.77 3.91 0.15 0.26 1.1 -20 -17 60 35.2 0.6 0.6 na 1 0 17 3.7 na 48.8 47.1 na na

a The distillation curve data have been detailed elsewhere (13, 15). In the present analysis, these data were replaced by principal components of the distillation data as discussed in the experimental section. b Principal components. c na, not analyzed.

(45 variables) of the fuels included PAC, represented by 1-methyl phenanthrene. The first goal of the present study was to develop PCA models for describing exhaust emissions from two heavyduty diesel engines of different makes. Secondly, using PLS, relationships between fuel parameters and exhaust emission characteristics were derived and will be briefly discussed. The results provide indications for future development of heavy-duty diesel fuels with desirably low adverse effects on the environment and human health. In separate papers, relationships between fuel parameters and exhaust emissions have been covered in more detail (20, 22). What is more important, the relationships between factors of particulate exhaust emissions and genotoxic potency have been presented (24).

The engines were operated on a chassis dynamometer according to a well-defined transient driving cycle (Stochasticher Fahrzyklus fu ¨ r Stadtlinien Omnibusse “bus cycle”) simulating public transportation within a city (25). It should be noted that the engine models from both engine makes were changed after the D1-D9 fuels were run. An initial PCA was performed on the characteristics of a wide set of heavy-duty diesel engines (data not shown). The results indicated that the differences between the two engine models, from each make, would have no major impact on the exhaust emission. For each engine make, the two engine models were therefore treated as one. A synthetic highquality lubricating oil was used in all engine models so as to minimize the possible contribution of engine oil-derived PAC to the exhaust emissions (26, 27). Physical parameters of the engines are outlined in Table 2.

Experimental Section

The exhaust emissions of HC, CO, NOx, CO2, particulates, aldehydes, oxygenates, light aromatics, and PAC (Table 3) were all assessed in parallel with taking filter samples. Hence, it should be noted that the term ‘sample’ in this context denotes all measurements made during the same run of a fuel and engine. The fuels and the engines as well as the sampling and analysis methods of exhaust emissions have been described in detail elsewhere (20, 22, 23, 25, 28). The notations of fuels and engines are consistent with these previous papers.

Fuels and Engines. Ten different diesel fuels, denoted D1, D2, D4, D5, D7-D10, D14, and D15, were combusted in two makes of heavy-duty diesel engines. Fuels D1, D2, and D4 were derived from the same base fuel and together with D10, D14, and D15 were from the same supplier. Fuel D5 was a blend of kerosenes. Fuel D8 was derived from D7 through the addition of 2000 ppm ethylhexyl nitrate (EHN) as an ignition improver. Fuel D9 was a blend of highly hydrated fuel composed from cracked gas oils with low sulfur contents. For fuel D10, additional sample sets (henceforth denoted D10f) were taken using exhaust aftertreatment devices, i.e. particulate traps. Physical and chemical characteristics of the fuels are shown in brief format in Table 1. A more extensive description and the chemical and physical relationships of these fuels have been published elsewhere (20, 22, 23).

Statistical Pretreatment. A total of 72 exhaust emission samples from the 10 diesel fuels and two engine types was analyzed with respect to 63 chemical variables. Some variables, i.e., sum of 14 PAH (ec28), sum of 27 PAC (ec29), soluble organic fraction (SOF, ec33), SO42- (ec34), and sulfate-bound water (ec39), were calculated from other, already-included variables and were therefore left out from

VOL. 30, NO. 1, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

39

TABLE 2

Engine Data engine make 1a

engine model deplacement (L) max power (kW/rpm) max torque (Nm/rpm) compression ratio cylinder diameter (mm) stroke length (mm) no. of cylinders a

engine make 2a

fuels D1-D9

fuels D10-D15

fuels D1-D9

fuels D10-D15

DSC1104 11 191/1800 1210/1000 17 127 145 6

DSC1127 11 180/1800 1235/1350 18 127 145 6

TD101 F 9.6 229/2050 1230/1300 14.3 121 140 6

THD101 C 9.6 227/2200 1220/1400 15 121 140 6

PCA of diesel engine physical data (not shown) indicate that the two engine models within each engine make are to be considered equal.

TABLE 3

List of Measured and Calculated Chemical Diesel Fuel Exhaust Emission Variablesa variable no.

variable name

sampled by

variable no.

variable name

sampled by

ec1 ec2 ec3 ec4 ec5 ec6 ec7 ec8 ec9 ec10 ec11 ec12 ec13 ec14 ec15 ec16 ec17 ec18 ec19 ec20 ec21 ec22 ec23 ec24 ec25 ec26 ec27 ec28 ec29 ec30

Polycyclic Aromatic Compounds (PAC) phenanthrene* anthracene* 4-methyldibenzothiophene 3-methyldibenzothiophene 3-methylphenanthrene 2-methylanthracene 4 & 9-methylphenanthrene 1-methylphenanthrene fluoranthene* pyrene* 1-methyl-7-isopropylphenanthrene benzo[a]fluorene 2-methylpyrene 1-methylpyrene benzo[ghi]fluoranthene* cyclopenta[cd]pyrene benz[a]anthracene* chrysene/triphenylene* benzo[b & k]fluoranthene* benzo[e]pyrene* benzo[a]pyrene* perylene indeno[1,2,3-cd]fluoranthene indeno[1,2,3-cd]pyrene* picene benzo[ghi]perylene* coronene* sum of PAC (14)*,b sum of PAC (27)b 1-nitropyrene

a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a

ec31 ec32 ec33 ec34 ec35 ec36 ec37 ec38 ec39 ec40 ec41 ec42 ec43 ec44 ec45 ec46 ec47 ec48 ec49 ec50 ec51 ec52 ec53 ec54 ec55 ec56 ec57 ec58 ec59 ec60 ec61 ec62 ec63

Non-PAC particles organic soluble fraction soluble organic fraction (SOF)b sulfate, SO42- b SOF, fuel derived SOF, oil derived particle carbon content soluble sulfate sulfate bound waterb nitrate in particle HC, hydrocarbons CO, carbon monoxide NOx, nitrogen oxides CO2, carbon dioxide formaldehyde acetaldehyde methanol ethanol acrolein acetone 2-propanol methacrolein 3-buten-2-one N-butanal methyl ethyl ketone ethyl acetateb isovaleraldehyde benzene toluene xylene ethylbenzene ethylene propylene

a a a a a a a a a a b b b b c c c c d d d d d d d d d d d d d d d

a Variables were determined after sampling by (a) filter, (b) bag, (c) cartridge, or (d) absorbent. b Variable analyzed but excluded from calculations: ec28 is composed of the sum of 14 PAC marked with an asterisk (*); variable ec29 is the sum of PAC variables ec1-ec27; ec33 is derived from ec 32; ec34 and ec39 correlate by definition to ec38; ec56 contains >50% missing data.

further statistical procedures. In the PCA analysis, ethyl acetate (ec56) was excluded due to a high proportion of missing data (>50%). Similarly, samples with a high proportion of missing data were excluded. Thus, 64 samples and 57 variables were retained in the final PCA data set. In the PLS analysis, an additional 12 physical and 12 chemical parameters of the fuels were included. Since there was a very good correlation between adjacent distillation curve points (detailed in ref 23), these variables were pretreated using PCA (described below). The 13 originally measured distillation variables were reduced to four orthogonal, i.e., mutually noncorrelated, principal component score vectors, henceforth denoted fdp1-fdp4. Approximately, fdp1 corresponds to the 40% boiling point

40

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 30, NO. 1, 1996

(BP40), describing 55% of the distillation curve data. Similarly, fdp2 (43%) corresponds to BP95, fdp3 (1.1%) corresponds to final boiling point (FBP), and fdp4 (0.1%) corresponds to initial boiling point (IBP). Since these vectors comprise PC scores and not original data, some distillation points are presented in Table 1. They were, however, not included in the analysis. The use of different engine makes introduces additional causes of variation in the emission of diesel exhausts. Two separate PLS models were therefore derived. Also, the D10f samples were omitted in both PLS analyzes, since the fuel characteristics cannot explain the differences between exhaust emission samples derived from the same fuel (D10) with and without exhaust aftertreatment devices.

FIGURE 1. Loadings (a and b) and scores (c and d) plots for the first three PCs of exhaust emissions from engine make 1. Notations of the variable points are presented in Table 3, whereas sample notations are consistent with the denotation of the fuel from which the samples emanate.

Multivariate Analysis. The data matrix was statistically analyzed using principal components analysis (PCA) (2932) and partial least squares analysis (PLS) (32-38). The fundamentals for these methods are the same. Both methods use the stepwise reduction or projection of matrix data onto orthogonal vectors. However, while PCA is used merely for exploration of correlation patterns in complex data matrices, PLS can be used for finding plausible causal relationships between variables. Principal Components Analysis. Every sample (row in a data matrix) may be described by a number of variables (columns in the matrix). In geometrical terms, a sample can be represented as a point in a multidimensional space, where the axes are defined by the variables. With many samples, correlations between variables may be observed. A principal component, PC can be expressed as a straight line through the sample points in the multivariable space. The direction of the line is calculated to minimize the distance of the points to the line in a least squares manner. The contributions of individual variables to the PC are expressed as loadings (cosine of the angle between the variable axis and the PC). Sample points are projected to the PC, and the positions of the projected points make up a vector of sample scores. For each PC, a loadings vector (p) and a scores vector (t) is obtained. By multiplication of the two vectors, the original data matrix (X) may be approximated. A matrix of residuals (E) is calculated as X - p × t. Next PC is then extracted from E. The iterative extraction of PCs proceeds until a limit of relevance is reached, i.e., no more systematic variation remains in the residual matrix. At that point, the raw data have been filtered to a loadings matrix (P) and a scores matrix (T), which contain the systematic variable and object information of the data set, respectively. Thus, the raw data may be reproduced from the PCs such that X ) 1X h + TP′ + E. Since every PC explains a maximum of the remaining matrix

variance, all PCs are orthogonal. The results of a PCA may easily be interpreted by plotting scores and loadings, respectively, in two- or three-dimensional graphs. Relationships between samples and between variables can be identified. Also, objects with high (positive or negative) scores are explained by variables that have high loadings in the same PC. In the present analyzes, loadings and scores are presented in separate, but superimposable, graphs in each figure. Hence, for every presented PC, the most descriptive variables (panels a and b) and the samples with high values (panels c and d) of these variables are easily identified, since they are located in the same directions in the graphs. PC models can be compared, providing identical sets of variables have been used, by plotting the loadings for one model versus the loadings of the other model. If all variables were to have the same impact in both PC models, a plot would then produce a diagonal of loading points (slope ) (1, r ) (1). Hence, the more diverse the models, the less correlated the loading points. If samples in both models have the same origin (fuels in this context), the score means and spreads can be calculated. The agreement between the models can then be visualized by plotting the range of scores, one model versus the other. Partial Least Squares Projection to Latent Structures, PLS. In PLS, the components (denoted latent variables or β) are extracted from one X matrix (descriptor matrix) and from one Y matrix (response matrix). Components are extracted so as to obtain a maximal predictivity of the response data but also to approximate the X and Y matrices. As with PCA, the components are orthogonal, and the loadings may easily be graphically interpreted. Since components are extracted for both matrices, scores are for each PLS dimension projected to two separate vectors, t (X scores) and u (Y scores), forming two matrices of scores,

VOL. 30, NO. 1, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

41

42

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 30, NO. 1, 1996

unit

µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km µg/km

g/km g/km mg/km mg/km mg/km mg/km mg/km

g/km g/km g/km g/km mg/km mg/km mg/km mg/km

mg/km mg/km mg/km

variable

ec1 ec2 ec3 ec4 ec5 ec6 ec7 ec8 ec9 ec10 ec11 ec12 ec13 ec14 ec15 ec16 ec17 ec18 ec19 ec20 ec21 ec22 ec23 ec24 ec25 ec26 ec27 ec30

ec31 ec32 ec35 ec36 ec37 ec38 ec40

ec41 ec42 ec43 ec44 ec45 ec46 ec47 ec48

ec49 ec50 ec51

18.7 12.1 0.2

1.24 2.46 11.5 1082 111.7 47.3 3.8 24.9

0.39 0.18 11.90 108.0 171.0 na 1.10

3.01 0.36 bd 0.16 1.68 2.11 2.33 1.84 10.22 13.63 0.24 0.24 0.55 0.49 1.24 0.70 0.32 0.80 0.23 0.13 0.03 0.01 0.01 bd 0.01 0.01 bd 0.64

D1

15.1 9.1 0.2

1.29 3.48 12.5 1083 140.0 40.5 1.1 10.8

0.41 0.21 8.80 169.0 181.0 na 2.10

0.16 0.01 bd bd 0.83 1.19 1.23 0.95 9.09 14.40 0.05 0.49 0.92 0.67 2.45 0.78 0.73 1.50 0.40 0.35 0.02 bd bd 0.01 bd 0.02 bd 0.09

D2

8.4 22.9 0.4

1.40 3.28 12.7 1113 81.5 24.0 na 11.0

0.46 0.18 21.90 128.0 184.0 28.20 3.20

0.59 0.12 bd bd 0.58 0.68 0.75 0.72 7.95 9.13 0.14 0.55 0.51 0.39 1.40 0.32 0.73 1.13 0.58 0.51 0.24 0.17 bd 0.04 0.04 0.10 bd 0.85

D4

15.8 12.1 0.3

1.23 2.88 12.9 1186 91.5 57.5 na 39.6

0.45 0.18 45.30 79.0 244.0 2.00 0.60

8.16 1.20 bd bd 7.75 9.46 9.73 8.63 22.92 23.25 2.09 2.27 4.46 2.60 2.49 2.50 1.21 2.85 0.61 1.09 0.85 1.04 bd bd bd 0.05 bd 0.25

D5

12.1 12.9 0.1

1.43 3.33 12.5 1156 258.0 49.5 3.8 26.1

0.45 0.18 32.00 96.0 236.0 5.30 2.80

2.95 0.20 bd bd 6.89 7.32 10.33 8.29 18.61 29.88 0.41 1.19 1.43 1.09 2.30 1.48 0.67 2.16 0.16 0.78 0.13 0.03 bd bd bd 0.06 bd 0.07

D7a

13.4 13.1 0.4

1.57 2.81 11.9 1167 127.0 75.0 3.0 63.7

0.45 0.16 29.90 68.0 266.0 na bd

3.46 0.34 bd bd 4.25 5.09 7.70 6.99 14.11 18.96 0.66 1.03 0.93 0.48 1.37 0.80 0.41 0.91 0.09 0.30 0.18 0.35 bd bd bd bd bd 0.18

D8

engine make 1

15.0 8.9 0.1

1.32 2.44 12.1 1183 63.5 59.0 na 19.7

0.42 0.16 35.40 73.0 221.0 na 0.60

9.37 1.21 bd bd 19.92 20.81 9.52 10.20 18.75 27.49 0.35 1.31 1.70 1.29 2.48 1.22 0.97 1.80 0.34 0.71 0.26 0.05 bd 0.02 0.08 0.21 bd 0.15

D9

na na na

0.01 0.12 11.4 1108 4.8 6.2 62.0 29.0

0.15 na 8.53 1.3 90.1 2.49 1.26

0.78 bdc 0.11 0.04 0.30 bd nad 0.22 0.61 0.91 0.04 0.08 0.23 0.12 0.06 bd 1.23 0.02 0.01 0.02 0.09 bd bd bd bd bd bd 0.43

D10f

na na na

0.78 1.63 11.1 1093 22.1 12.1 64.4 23.7

0.20 na 21.75 8.8 110.9 0.56 0.96

3.57 0.10 0.42 0.14 1.74 0.04 na 1.21 4.81 16.30 0.15 0.82 4.37 2.15 0.78 0.11 1.07 0.48 0.18 0.10 0.56 0.03 bd bd bd bd bd 3.67

D10

D14

na na na

0.69 1.46 11.3 1051 8.6 6.5 25.0 78.0

0.16 na 15.76 9.5 82.4 0.54 1.07

5.87 0.29 0.66 0.22 1.28 0.07 na 1.09 4.23 8.77 0.14 0.65 1.36 0.72 0.60 0.22 0.32 0.34 0.19 0.10 1.46 0.07 bd bd bd bd bd 1.57

Mean Values of Exhaust Emission Data Calculated per Engine Make and Fuel

TABLE 4

na na na

0.56 1.49 11.6 1064 22.9 19.0 50.0 12.0

0.16 na 27.24 13.3 91.1 0.31 1.21

2.79 0.22 0.52 0.18 0.79 0.06 na 0.83 3.85 11.54 0.14 0.60 2.40 1.14 0.57 0.16 0.24 0.30 0.11 0.06 0.65 0.03 bd bd bd bd bd 2.17

D15

24.2 17.2 55.1

1.39 6.96 15.6 1100 58.0 38.5 na 63.2

0.47 0.08 40.20 15.8 342.0 na 1.60

5.47 0.71 0.36 0.29 2.56 2.96 3.19 2.91 11.53 11.72 0.35 0.20 0.88 0.53 0.84 0.53 0.39 0.65 0.22 0.39 0.19 0.30 bd bd 0.32 0.32 bd 0.32

D1

35.8 17.0 38.7

1.31 7.15 15.5 1119 68.0 25.0 5.0 49.0

0.50 0.09 44.90 20.2 329.0 na 2.10

9.76 1.41 0.05 0.02 3.46 4.21 2.93 2.83 18.38 19.60 0.30 0.30 0.49 0.44 1.98 0.62 0.44 0.83 0.05 0.35 0.15 0.31 bd 0.01 0.03 0.02 bd 1.03

D2

23.2 13.1 44.2

1.42 8.12 16.4 1143 101.0 30.0 na 29.9

0.59 0.10 43.50 17.0 363.0 14.60 3.80

3.89 0.67 bd 0.25 1.70 1.96 2.06 1.74 14.29 10.38 0.44 0.31 0.75 0.46 2.12 0.90 1.00 1.69 0.46 0.84 0.45 0.45 0.10 0.20 0.06 0.31 0.01 7.19

D4

30.0 16.3 54.2

1.39 7.92 16.1 1119 92.0 75.0 143.3 34.6

0.56 0.11 51.40 37.0 373.0 1.30 2.60

18.46 2.63 bd bd 9.95 11.66 10.66 10.07 21.01 20.79 0.84 1.04 2.06 1.43 1.69 1.39 1.07 1.34 bd 0.43 0.24 0.57 bd bd bd bd bd 1.51

D5

32.3 17.6 45.7

1.52 6.53 16.9 1128 134.5 128.0 77.5 53.2

0.40 0.13 38.30 23.9 275.0 1.60 2.30

4.53 0.97 bd 0.07 3.93 4.48 5.24 4.89 11.92 11.93 0.75 0.68 1.54 1.08 1.02 2.20 0.79 1.03 0.13 0.37 0.30 0.13 0.11 bd bd bd bd 0.57

D7

26.8 16.0 66.5

1.22 5.95 15.3 1141 74.0 106.5 58.6 42.4

0.46 0.10 42.40 24.6 339.0 na 1.90

7.48 1.12 bd bd 4.24 5.87 8.21 6.75 21.81 21.15 0.62 0.66 0.92 0.55 1.94 1.56 1.32 1.19 0.32 1.10 0.76 0.67 bd 0.02 0.02 0.07 0.02 0.45

D8

D9

24.9 22.0 49.8

1.15 6.64 15.8 1115 65.5 118.5 66.2 48.2

0.56 0.10 49.40 31.2 345.0 na 0.80

17.86 1.59 bd bd 23.31 25.51 9.65 10.81 28.25 45.26 0.23 0.70 1.45 0.83 4.13 1.81 1.23 3.03 0.59 0.93 0.15 0.09 bd 0.05 0.19 0.21 bd 2.56

engine make 2

na na na

0.46 1.92 14.9 1259 124.1 85.3 100.0 25.0

0.03 na 7.94 1.5 4.4 0.37 0.55

0.13 0.10 bd bd 0.07 bd na 0.10 0.37 0.53 0.07 0.07 0.13 0.07 0.10 bd 0.53 0.07 0.04 bd bd bd bd bd bd bd bd 0.16

D10f

na na na

0.73 3.03 15.3 1272 123.8 87.9 130.0 23.0

0.40 na 29.38 45.7 231.5 1.30 1.97

6.57 2.40 bd 0.10 3.10 0.63 na 2.87 6.00 24.33 0.30 0.93 5.77 2.87 0.90 0.33 0.23 0.47 0.10 0.07 0.10 bd bd bd bd bd bd 0.45

D10

na na na

0.62 3.14 15.0 1259 105.4 67.3 47.4 24.2

0.41 na 19.32 33.5 247.2 3.68 2.14

6.40 0.93 bd 0.14 1.70 0.37 na 2.23 5.80 10.83 0.23 1.10 1.03 0.50 0.60 0.23 0.23 0.40 0.07 bd 0.04 bd bd bd bd bd bd 0.32

D14

na na na

0.54 2.53 13.8 1179 na na 65.0 8.0

0.43 na 34.75 34.8 231.1 2.71 1.95

5.80 1.00 bd 0.20 1.70 0.40 na 2.20 6.20 21.90 0.20 0.80 4.30 1.90 0.90 0.30 0.20 0.50 bd 0.10 0.10 0.10 bd bd bd bd bd 0.48

D15b

4.74 9.15 6.02 3.39 0.24 9.37 5.44 12.48 7.20 125.0 21.3

3.57 7.39 5.01 2.78 0.20 8.74 4.04 6.18 5.42 64.0 11.5

4.06 8.00 5.10 3.70 0.93 9.24 3.98 9.73 6.00 72.0 17.8

na na na na na 51.50 34.50 6.50 na na na

na na na na na 28.67 11.00 bd na 69.9 na

na na na na na 34.00 5.33 bd na 78.0 na

na na na na na 43.00 16.00 4.00 na na na

T and U. The relation between the X and Y matrices can be graphically presented by plotting these vectors. In the present study, PCA and PLS analyzes were applied on autoscaled data, which means that each variable was mean centered, and its variance was set to 1. PCs and latent variables were extracted using the NIPALS (non-linear iterative partial least-squares) algorithm with cross-validation. The level of statistical significance was 0.05. All PCA and PLS calculations were performed using version 5.1 of the SIMCA-S for Windows software, developed by UMETRI AB, Umea˚, Sweden. In this study, three PCA analyzes were made, based on the two individual engine makes, as well as an analysis with the two separate engine makes combined to one data set. Also, two PLS analyses were performed using the fuel data as a common X matrix and the two matrices of emissions from the engine makes, respectively, as Y matrices.

na, not analyzed. d

One sample analyzed. c bd, below detection level. b

Two samples included in statistical analysis. a

ec52 ec53 ec54 ec55 ec57 ec58 ec59 ec60 ec61 ec62 ec63

mg/km mg/km mg/km mg/km mg/km mg/km mg/km mg/km mg/km mg/km mg/km

3.18 5.37 4.05 2.65 0.40 8.36 9.15 4.80 3.25 na na

2.23 4.73 3.96 3.26 0.16 5.11 4.75 6.74 3.59 na na

1.85 4.05 3.43 2.51 0.17 5.67 2.84 5.38 2.78 31.5 9.7

2.42 4.98 4.26 2.55 0.15 5.99 5.80 8.40 5.81 40.0 11.1

1.88 4.32 3.26 2.09 0.12 4.29 2.51 8.66 3.68 39.5 12.2

1.91 5.00 3.62 3.31 0.17 4.39 1.89 13.89 6.36 41.0 7.4

2.24 5.25 2.91 1.69 0.18 5.27 2.64 7.00 3.96 33.0 7.7

na na na na na na na na na 60.0 60.0

na na na na na na na na na 71.0 60.0

na na na na na na na na na 30.0 30.0

na na na na na na na na na 46.5 40.0

3.77 7.03 4.27 5.03 0.70 8.25 4.87 9.28 6.26 91.5 9.6

3.96 7.83 5.19 3.31 0.68 10.00 4.28 5.60 6.09 69.0 10.4

3.68 7.88 4.96 3.01 0.36 8.01 4.56 3.91 4.46 80.0 6.8

4.51 9.09 5.48 3.39 0.30 9.46 6.31 8.33 6.78 102.0 14.6

Results and Discussion PCA, Engine Make 1. With five PCs, 71.8% of the total matrix variation (expressed as sum of squares, SSX) could be explained (Table 5). The two least significant PCs did not describe fuel effects but rather explained variations between individual samples and will therefore not be discussed. With this engine make, one exhaust sample from fuel D7 was recognized as an outlier, exhibiting extreme levels of PAC in general and specifically benzo[a]pyrene, perylene, and picene. The amounts of these compounds were more than 20 times higher in this individual sample as compared to other samples from D7. It was therefore omitted from further PCA analysis. For this engine make, the loadings and scores of the first three PCs are shown in Figure 1 panels a-d, respectively. It should be remembered that panels a and c are superimposable, as are panels b and d. The first PC accounted for 33.6% of the matrix sum of squares (SSX) (Table 5). As seen in Figure 1a, the most descriptive variables included fluoranthene (ec9), 1-methylphenanthrene (ec8), particle carbon contents (ec37), and benzo[e]pyrene (ec20). CO2 (ec44) as well as the regulated pollutants (ec31, ec41-43) also contributed to this direction of the first PC. Elevated emissions of these pollutants are indicated in samples derived from fuels D5, D9, and D7 (Figure 1c). On the other extreme, samples from D10f, D10, D14, D15, and D1 had high amounts of propylene (ec63) and methanol (ec47) but lower levels of PAC in general (Figure 1, panels a and c. The emission of CO2 (ec44) is a measure mainly of fuel consumption (on mass basis). Accordingly, the exhaust emission of CO2 was higher from the higher density fuels (D5-D9), as seen in Table 4. There are, however, other aspects of CO2 exhaust emission that will be further discussed below. Other features seen in this PC include the negative correlation between 1-nitropyrene (ec30) and the correlated pair of NOx (ec43) and pyrene (ec10). Hence, the results indicate the formation of 1-nitropyrene from the reaction between nitrous gases and pyrene (8, 10,12). It has also been shown that nitro derivation of PAH by NO2 may occur on filters (39). The second PC explained 14.8% SSX. It was mainly contributed to by sulfates (ec38), nitrate (ec40), and oilderived soluble organic fraction (O-SOF, ec36), as exemplified by samples from D4 and D2 (Figure 1, panels a and c). These variables were negatively correlated to 1-meth-

VOL. 30, NO. 1, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

43

TABLE 5 2

Fractions of Sums of Squares (s ) and Variances after Correction for Missing Data (s2 adj) for PCA Models engine make 1

engine make 2

both engine makes

PC no. s2 (%)a s2 (adj) (%)b s2 (%) s2 (adj) (%) s2 (%) s2 (adj) (%) 1 2 3 4 5 totalc a

33.6 14.8 8.9 7.3 7.1

29.7 12.3 6.9 5.9 6.5

25.1 13.3 9.6 8.3 6.7

20.7 10.1 7.1 6.5 5.3

25.0 14.2 9.4 7.0 6.3

22.1 12.2 7.9 5.9 5.5

71.8

61.3

63.0

49.7

61.2

53.6

s2

The denotes the fraction of sum of squares explained per PC dimension. b The s2 adj denotes the matrix variance after correction for missing values. c After five principal components.

ylpyrene (ec14) and 2-methylpyrene (ec13). Except for these pyrenes, the impact from PAC was of low relevance in this PC. The D10 samples showed higher levels of the methylated pyrenes, whereas the D10f samples contained more nitrate and sulfates. This may be due to postcombustion reactions in the exhaust aftertreatment device. In PC 3 (8.9% SSX), the D1 samples (Figure 1d) contributed with high levels of methacrolein (ec52), toluene (ec59), and benzene (ec58) (Figure 1b). One D14 sample clustered with the D1 samples (Figure 1d). This was, however, mainly due to very low levels of ethylene (ec62) and benz[a]anthracene (ec17). The D10f samples and one D9 sample were, in contrast, high regarding benz[a]anthracene (ec17). The two least significant PCs explained 7.9% and 7.1% of the SSX, respectively, but were mainly contributed to by individual samples. Thus, these components mainly reflected variations in running of engines, sampling, and chemical analysis procedures. PCA, Engine Make 2. Using five PCs, 63.0% of the SSX was explained with this engine make (Table 5). Loadings and scores are presented graphically in Figure 2, panels a-d. The first PC (25.1% SSX) was mainly described by the impacts of CO2 (ec44), benzene (ec58), and toluene (ec59). It was mostly contributed to by D10f samples. On the other extreme, fuel-derived soluble organic fraction (F-SOF, ec35), fluoranthene (ec9), the regulated pollutants, HC (ec41), and CO (ec42), were contributed to mainly by samples from fuel D9. Within this cluster of variables, there were also some PAC variables, e.g., 1-methylphenanthrene (ec8) (Figure 2a). In a previous PCA study performed on these fuels (23), similar correlation patterns of PAC in the fuels were found, although no clear negative correlation to toluene and benzene could be observed. In this PCA, the exhaust emission of CO2 (ec44) was negatively correlated to the emissions of the regulated compounds HC (ec41), CO (ec42), and particulates (ec31). The ratio of CO2 versus these pollutants reflects the efficiency by which the fuels are combusted. In the ideal case, i.e., a complete combustion, all fuel carbon is fully oxidized into CO2. The nature of CO2 emissions will be further discussed below. The second PC (13.3% SSX) mainly separated samples from D4 and to a lesser extent D1 based on the difference in emitted amounts of sulfates (ec38), indeno[1,2,3-cd]pyrene (ec24), and 3-methyl dibenzothiophene (ec4). For

44

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 30, NO. 1, 1996

samples from D5, D9, and D10, the contents of these variables were low, but high with respect to methanol (ec47). In the third PC (9.6% SSX), samples from D9 were separated (Figure 2b) from the others by their increased levels of isovaleraldehyde (ec57). The D10f samples (Figure 2d) clustered with the D9 samples but were low in particulate nitrate (ec40), which was negatively correlated to isovaleraldehyde (ec57). The D10 samples were, in contrast, high regarding the methylated pyrenes (ec13 and ec14). As with engine make 1, the results from the fourth (8.3% SSX) and fifth (6.7% SSX) PC reflected variances derived from engine running, sampling, and chemical analysis and so will not be further discussed. Comparisons of PC Models for the Two Engine Makes. When PC analyses for both engine makes were compared, some common patterns were observed (Figure 3, panels a-d). The first principal component was mainly similar for the two engine makes (Figure 3, panels a and c). The best corresponding variables of the first PC were the particle carbon content (ec37) and fluoranthene (ec9). The first components separated the fuels with respect to regulated pollutants (ec41-43) and the contents of PAC, in general. The exhaust emissions of propylene (ec53) and methanol (ec47), in engine make 1, and toluene (ec59) and benzene (ec58), in engine make 2, were negatively correlated with the exhaust emissions of PAC. A discrepancy between the engines was observed (Figure 3a) regarding the emission of CO2 (ec44). Possible causes of this discrepancy will be discussed below. The second PC was also similar for the two engine makes. In both models, sulfate (ec38), 1-methylpyrene (ec13), 2-methylpyrene (ec14), and methanol were (ec47) important (Figure 3b). Exhaust samples that were well described in both engine makes (Figure 3d) included the D4 samples with high sulfates (ec38) and exhaust samples from D5 and D10 with high 1-methylpyrene (ec13) and 2-methyl pyrene (ec14) contents (Table 2). The impact from sulfates (ec38) in the D4 samples was a reflection of a significantly higher level of sulfur in the fuel as compared to the other fuels (cf. PLS analysis below). Similarly, the D5 samples were derived from a fuel that was very high with respect to 1-methylpyrene (ec13) and 2-methylpyrene (ec14) (cf. ref 16). Fuel D10, in contrast, contained below detection levels of these compounds (22), although the emissions of them were high. The cause for formation and emission of the methylated pyrenes (ec13 and ec14) is yet not clarified. Interestingly, with exhaust aftertreatment (particulate trap), levels of the these compounds were reduced 1 order of magnitude. This indicates that reactions occur within the exhaust aftertreatment device. Also the second PC showed a discrepancy between engine makes. The oil-derived soluble organic fraction (ec36) was well correlated with sulfate (ec38) and nitrate (ec40) in engine make 1, with high emissions in samples from D2 and D4. In engine make 2, it was correlated with the methylated pyrenes (ec13 and ec14) and methanol (ec47) (Figure 3, panels b and d). Samples that in this engine contributed with high emissions of these compounds were derived from fuels D5, D7, and D10. The nature of this interaction can, however, not be assessed in this study, although the emission of oil-derived soluble organic fraction (ec36) will be discussed briefly below.

FIGURE 2. Loadings (a and b) and scores (c and d) plots for the first three PCs of exhaust emissions from engine make 2. Notations of the variable points are presented in Table 3, whereas sample notations are consistent with the denotation of the fuel from which the samples emanate.

FIGURE 3. Loadings (a and b) and scores (c and d) plots for the first two PCs of exhaust emissions from engine make 1 in comparison with engine make 2. Notations of the variable points are presented in Table 3, whereas sample notations are consistent with the denotation of the fuel from which the samples emanate. Rectangles show the ranges of score data from engine make 1 (horizontal spread) and engine make 2 (vertical spread).

From the three last PCs, the PCA results show that there were variations in samples that did not originate from the fuels alone. These influences included slight variations in driving the engines, sampling of exhaust emission fractions, and chemical analysis. Although the ‘error’ in each step may be minimal, the net result is the sum of deviations,

and it may therefore be difficult to detail these influences. PC Model of Both Engine Makes. An additional PC analysis of the engine makes combined (Figure 5) showed that exhaust emission samples from the two engine makes were discriminated by the two first principal components (Figure 5b), which together explained 39% SSX (Table 5).

VOL. 30, NO. 1, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

45

FIGURE 4. Relation between CO2 emissions and fuel consumption from the two engine makes. (-) denotes samples from engine make 1. (+) denotes samples from engine make 2.

FIGURE 5. Loadings (a) and scores (b) plots for the first two PCs of exhaust emissions from both engine makes. Notations of the variable points are presented in Table 3, whereas sample notations are consistent with the denotation of the fuel from which the samples emanate.

Interestingly, there was an impact of oil-derived soluble organic fraction (SOF, ec36) for engine make 1. This was because engine make 2 had been adjusted to minimize oil leakage. Engine make 2, however, emitted a higher amount of PAC in general and of particulate carbon (ec37). The same engine also emitted more methanol as seen from PC 2 (Figure 5) as well as more benzene (ec58) and toluene (ec59). The higher exhaust emission of oil-derived SOF from engine make 1 seems to have no impact on the emission of the regulated pollutants. In contrast, these emissions were higher with engine make 2. The discrepancy of CO2 emissions seen for the engines was reflected in the first two PCs, where this compound had low impact on the model (loadings near origo). This will be discussed further below. A comparison of the D10f samples versus the unfiltered particulate samples from D10 indicated the efficiency of the exhaust aftertreatment device. With both engine makes,

46

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 30, NO. 1, 1996

PAC and particle emissions were reduced as compared to the unfiltered samples. In general, the effects of the exhaust aftertreatment device was more clearly observed with engine make 2 than with engine make 1. The reduction of DEP and of most PAC was in agreement with results previously shown for this type of engine but with fuels containing >50 times the PAC levels as compared to D10 (17, 18). It can therefore be concluded that the use of an exhaust aftertreatment system effectively reduces the exhaust emissions of DEP for a wide range of low sulfur diesel fuels. PLS Analysis. Since the difference in engine make and design introduced exhaust emission variations that were not inherent from the fuels, separate PLS models were derived for the two engine makes. The main object of this PLS study was, however, to find exhaust emission characteristics relative to fuel composition. Hence, results for the two individual PLS models will be jointly discussed and illustrated. For the two first dimensions, the patterns of correlations and plausible causality were very similar for the two PLS models (Figure 6). The proportions of explained matrix variances are shown in Table 6. As seen in Figure 6a, the exhaust emission variables that were best described by the fuel data, in both models, were those related to PAC and particulates. The sums of both 14 PAH (ec28) and of 27 PAC (ec29) were well correlated with the corresponding variables of the fuel (fc28 and fc29). Also, the exhaust emissions of particles (ec31), particle carbon contents (ec37), and unburnt hydrocarbons (ec43) seem directly related to the fuel contents of PAC and fuel aromaticity (fc38). These exhaust emissions were, as also shown in the PCA analysis, well correlated with the other regulated pollutants; NOx (ec43) and CO (ec42). The fuels giving rise to exhaust samples with high levels of the above variables were the more aromatic and higher density fuels D5, D7, D8, and D9 (Figure 6d). In contrast to these fuels, D10, D14, and D15 all gave rise to exhaust samples with very low PAC, particle, and hydrocarbon emissions, but higher emissions of oxygenates and monoaromatics, i.e., benzene (ec58) and toluene (ec59). It has been demonstrated that PAC present in the exhaust emissions may be derived from pyrosynthesis in the combustion chamber (40), although a major proportion originates from unburnt fuel PAC (22, 40, 41). With this study, the close correlation of PAC exhaust emissions with the fuel contents of PAC substantiates the notion that PAC retrieved from the particulate exhaust emissions are unburnt fuel residues of PAC. The negative correlation in the exhaust emissions of benzene and toluene with PAC is most likely due to the negative correlation of benzene in the fuel (ec44) with PAC in the fuel. It can, however, not be ruled out that the monoaromatics, benzene and toluene, may be pyrosynthesized from the decomposition of larger size hydrocarbons in the combustion process (10). It has previously been stated that an increase in diesel fuel aromaticity has several effects on the exhaust emissions (7-9, 11, 12). With an increase of fuel aromaticity, the carbon to hydrogen ratio increases, giving a lower yield of heating energy per quantum of carbon. The higher contents of aromatics may also lead to a slightly higher density (14, 23). Thus, in higher aromatic fuels the specific energy per kilogram of fuel is lower, although the specific energy per

FIGURE 6. Loadings (a and b) and scores (c and d) plots for the first two PLS components (latent variables) of fuel data versus exhaust emissions from engine make 1 in comparison with engine make 2. Notations of the variable points are presented in Tables 1 and 3, whereas sample notations are consistent with the denotation of the fuel from which the samples emanate. TABLE 6

Fractions of Explained X (Fuel) Data and Y (Exhaust Emission) Data for PLS Models engine make 1 X PLS no.

s2

(%)b

s2

engine make 2 Y

adj

(%)c

s2

(%)

s2

X adj (%)

s2

(%)

s2

both engine makesa

Y adj (%)

s2

(%)

s2

adj (%)

X

s2

adj (%)

Y s2 adj (%)

1 2 3 4 5

24.8 20.9 14.5 17.4 7.7

18.6 17.5 12.8 19.6 8.8

35.6 14.7 11.1 5.6 6.3

33.0 13.1 10.2 4.6 5.9

18.6 24.3 17.4 15.4 7.7

11.7 20.9 16.2 16.8 8.5

27.1 10.7 7.5 6.4 6.9

24.2 8.5 5.6 4.8 5.9

16.8 19.3 19.7 14.4 4.0

17.8 6.4 1.9 2.8 1.4

totalc

85.3

77.3

73.3

66.8

83.4

74.1

58.6

49.0

74.2

30.3

a

Not discussed. b The s2 denotes the fraction of sum of squares explained per PC dimension. c The s2 adj denotes the matrix variance after correction for missing values. d Latent variables or PLS components. e After five PLS components.

liter of fuel may increase (23). High aromatic fuels generally have lower volatility, higher flash point, and lower cetane numbers as compared to low aromatic diesel fuels (7, 14, 23). The lower ignitability of these fuels may lead to a delayed ignition in the combustion chamber, which may result in a less complete combustion (7). A possible decrease of temperature in the combustion chamber in turn leads to a higher exhaust emission of soluble organic fraction (10). Hence, the effects observed for higher aromaticity fuels may be due both to an increase in fuel carbon consumption and to a less complete fuel combustion.

of formation, by the intermediate levels of PAC in the combustion chamber (40). Therefore, a fuel of higher aromaticity would yield more particles than would a low aromaticity fuel. Also, a higher combustion temperature, e.g., due to a higher engine load, is assumed to favor the formation of carbonaceous particle cores (10). This would also increase the formation of CO and NOx. Lower combustion chamber temperatures, in contrast, promote the formation and exhaust emissions of PAC and SOF (10). Hence, factors that influence the formation and composition of DEP include fuel aromaticity as well as combustion and exhaust temperatures.

Studies have shown that the particle to gas phase distribution of hydrocarbons in the exhaust emissions is dependent on their molecular weight and on the temperature of the exhaust emissions (ref 8 and references cited therein). It has also been suggested that the carbonaceous core of DEP is positively influenced, by similar mechanisms

In a PLS analysis (unpublished results) of published diesel fuel and exhaust parameters (7), a positive although statistically not significant correlation between aromaticity and exhaust temperature was observed. The ratio of air to fuel was however negatively correlated to the exhaust temperature. This may be due to an increase in fuel density

VOL. 30, NO. 1, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

47

and viscosity with increasing aromaticity (cf. ref 7). In the present analysis, the simultaneous increase of the regulated pollutants and DEP with higher aromaticity fuels indicate that these were affected primarily by the fuel’s contents of aromatics and PAC. Considering the present results, it cannot be excluded that the formation of DEP is a consequence also of PAC-promoted formation of particle cores, possibly due to an increase in average combustion temperature (not measured). As was previously shown in the PCA, the factor that deviated the most regarding the two PLS models was the exhaust emission of CO2. Theoretically, the formation of CO2 is influenced by (a) the input of fuel carbon (fuel consumption, in mass); (b) the energy yield per carbon atom in the fuel; and (c) the efficiency by which the fuel is combusted. Thus, for engine make 1 the emission of CO2 was a reflection of increased fuel consumption (by mass) of lower hydrogen/carbon ratio and possibly less complete combustion. This effect was not seen in engine make 2. Instead, the exhaust emissions of CO2 were higher from the lighter fuels. Since the fuels were exactly the same on the two engine makes, this discrepancy can only be explained by interactions of fuel and engine. The determination of the engine factors that contribute to differences in CO2 exhaust emission patterns does however not lie within the scope of the present investigation. The second PLS component demonstrated the direct relation of sulfur in the fuel (fc35) to the exhaust emissions of water-soluble sulfate (ec38) (Figure 6b). This was in good agreement with previous findings (12, 13). The particle-bound exhaust emissions of nitrate (ec40) were well correlated with sulfate (ec38). This indicates that the formation of sulfate has effects on either the formation of nitrates in the combustion process or on the condensation of nitrates on the particles (12). Also, the flash point (fp24) was well described by this PLS dimension. This may however be due to the negative correlation with paraffins in the fuel (fc42). The exhaust emission samples that contributed the most to this PLS component were derived from fuel D4 (Figure 6d). In contrast, exhaust emission samples from fuel D10 for example were low with respect to both sulfates and nitrates (cf. Table 4). The factors that were most different in the two PLS models, PLS component no. 2, were the emission of oilderived soluble organic fraction (ec36, Figure 6d). As discussed above, this discrepancy was due to the optimization of engine make 2 to minimize oil leakage. To conclude, many relations that have previously been demonstrated in other studies have been corroborated. It should however be noted that the present data matrix has been generated by running the engines according to a transient driving cycle that more closely resembles a reallife exposure situation. Also, the present study is based on a set of data obtained from 10 fuels, some of which were intended for environmental classification. The use of two engine makes also facilitates the discrimination of fuel effects and engine effects on the patterns of exhaust emissions. It is important to verify which of the exhaust emission components contributes most to specific adverse effects. It is well known that some individual PAH of exhaust emissions are carcinogenic. Since this study showed that there was a good correlation of fuel and emission PAC, an expected ‘good’ fuel would have low amounts of PAC. A

48

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 30, NO. 1, 1996

decrease in fuel sulfur is expected to reduce the amounts of sulfates and sulfur-containing PAH in the exhaust emission. The correlation of fuel sulfur to 1-nitropyrene further indicates that a reduction of sulfur would lead to particulate exhaust emissions of a lower biologically adverse potency. If other adverse effects, such as asthma, acid rain, and corrosion, were to be considered, other factors may influence the composition of the “ideal” diesel fuel. At present, these effects do not lie within the scope of this ongoing investigation. In the continuation of this work, however, genotoxic effects of diesel exhaust emissions have been investigated (24).

Acknowledgments The authors wish to thank Lena Elvfer for skillful assistance in the laboratory. Karl-Erik Egeba¨ck is acknowledged for valuable discussions. This work was financed by the Swedish Environmental Protection Agency.

Literature Cited (1) Westerholm, R. Inorganic and organic compounds in emissions from diesel powered vehicles. A literature survey. SNV Report 3389; Swedish National Environmental Protection Board: Stockholm, Sweden, 1987. (2) Tuominen, J.; Salomaa, S.; Pyysalo, H.; Skytta, E.; Tikkanen, L.; Nuromela, T.; Sorsa, M.; Pohjola, V.; Sauri, M.; Himber, K. Environ. Sci. Technol. 1988, 22, 1228-1234. (3) The Evaluation of the Carcinogenic Risk to Humans. IARC Monogr. Suppl. No. 7, 1987. (4) The Evaluation of the Carcinogenic Risk to Humans: Diesel and gasoline engine exhausts and some nitro-PAH. IARC Monogr. No. 46, 1989. (5) Code of Federal Regulations. Parts 81-99, Revised July 1, 1987; GPO: Washington, 1987. (6) World Health Organization. International Programme on chemical safety. Environmental helth criteria: Diesel fuel and exhaust emissions; Mercier, M., Ed.; World Health Organization: Geneva, 1994. (7) Mills, G. A.; Howarth, J. S.; Howard, A. G. J. Inst. Energy 1984, 57, 273-286. (8) Schuetzle, D.; Frazier, J. A. In Carcinogenic and mutagenic effects of diesel engine exhaust; Ishinishi, N., Koizumi, A., McLellan, R. O., Sto¨ber, W., Eds.; Elsevier Science Publications: Amsterdam, 1986; pp 41-63. (9) Williams, P. T.; Bartle, K. D.; Andrews, G. E. Fuel 1986, 65, 11501158. (10) Barale, R.; Bulleri, M.; Cornetti, G.; Loprieno, N.; Wachter, W. F. SAE Tech. Pap. Ser. 1992, No. 920397. (11) McCarthy, C. I.; Slodowske, W. J.; Sienicki, E. J.; Jass, R. E. SAE Tech. Pap. Ser. 1992, No. 922267. (12) Scheepers, P. T. J.; Bos, R. P. Int. Arch. Occup. Environ. Health 1992, 64, 149-161. (13) Baranescu, R. SAE Tech. Pap. Ser. 1988, No. 881174. (14) Eide, I.; Johansson, E. Chemom. Intell. Lab. Syst. 1994, 22, 7785. (15) McCreath, C. G. Combust. Flame 1971, 17, 359-366. (16) Rasmussen, R. E.; Devillez, G.; Smith, L. R. J Appl. Toxicol. 1989, 9, 159-168. (17) Westerholm, R.; Egeba¨ck, K.-E. Environ. Health Perspect. 1994, 102, 13-23. (18) Westerholm, R.; Alsberg, T.; Strandell, M.; Frommeli, A° .; Grigoriadis, V.; Hantzaridou, A.; Maitra, G.; Winquist, L.; Rannug, U.; Egeba¨ck, K.-E.; Bertilsson, T. SAE Tech. Pap. Ser. 1986, No. 860014. (19) Egeba¨ck, K.-E.; Bertilsson, B.-M. Chemical and biological characterisation of exhaust emissions from vehicles fuelled with gasoline, alcohol, LPG, and diesel. SNV PM 1635; National Swedish Environmental Protection Board: Stockholm, Sweden, 1983. (20) Westerholm, R.; Egeba¨ck, K.-E. Impact of Fuels on Diesel Exhaust Emissions: A Chemical and Biological characterization. Swedish Environmental Protection Agency Report 3968; Swedish EPA: Solna, Sweden, 1991. (21) Li, H.; Sjo¨gren, M.; Rannug, U.; Westerholm, R. In Polycyclic Aromatic Compounds: Synthesis, properties, analytical measurements, occurrence and biological effects; Garrigues, P., Lamotte,

(22) (23) (24)

(25) (26) (27) (28) (29) (30)

(31)

M., Eds.; Gordon and Breach Science Publishers: New York, 1993; pp 647-654. Westerholm, R.; Li, H. Environ. Sci. Technol. 1994, 28, 965. Sjo¨gren, M.; Hang, L.; Rannug, U.; Westerholm, R. Fuel 1995, 74, 983-989. Sjo¨gren, M.; Hang, L.; Banner, C.; Rafter, J.; Westerholm, R.; Rannug, U. Influence of physical and chemical characteristics of diesel fuels and exhaust emissions on biological effects of particle extracts: A multivariate statistical analysis of ten diesel fuels. Chem. Res. Toxicol., in press. Westerholm, R.; Alme´n, J.; Li, H.; Rannug, J.; Egeba¨ck, K.-E.; Gra¨gg, K. Environ. Sci. Technol. 1991, 25, 332-338. Williams, P. T.; Andrews, G. E.; Bartle, K. D. SAE Tech. Pap. Ser. 1987, No. 872084. Abass, M. K.; Williams, P. T.; Andrews, G. E.; Bartle, K. D. SAE Tech. Pap. Ser. 1987, No. 872085. Gra¨gg, K. Investigation of three diesel fuels regarding the impact of boiling point on the exhaust emissions. MTC 9109B; Motortestcenter: Haninge, Sweden, 1991 (in Swedish). Norde´n, B.; Edlund, U.; Wold, S. Acta Chem. Scand. 1978, 32, 602-608. Albano, C.; Blomquist, G.; Coomans, D.; Dunn, W. J.; Edlund. U.; Eliasson, B.; Hellberg, S.; Johansson, E.; Norde´n, B.; Johnels, D.; Sjo¨stro¨m, M.; So¨derstro¨m, B.; Wold, H.; Wold, S. In Pattern recognition by means of disjoint principal components models (SIMCA). Philosophy and methods; Ho¨skuldsson, A., Esbensen, K., Eds.; Proceedings of the Symposium on Applied Statistics; RECKU: Copenhagen, Denmark, 1981. Sharaf, M. A.; Illman, D. L.; Kowalski, B. R Chemometrics; Wiley: New York, 1986.

(32) Wold, S.; Albano, C.; Dunn, W. J.; Edlund, U.; Esbensen, K.; Geladi, P.; Hellberg, S.; Johansson, E.; Lindberg, W.; Sjo¨stro¨m, M. In Chemometrics; Kowalski, B. R., Ed; Cozensa, Italy, 1983. (33) Wold, H. In Systems under indirect observation; Jo¨reskog, H., Wold, H., Eds.; North-Holland: Amsterdam, 1982; Vol. 2, pp 1-54. (34) Wold, S.; Geladi, P.; Esbensen, K.; O ¨ hman, J. J Chemom. 1987, 1, 41-56. (35) Wold, S.; Albano, C.; Dunn, W. J.; Esbensen, K.; Hellberg, S.; Johansson, E.; Lindberg, W.; Sjo¨stro¨m, M. Analusis 1984, 10, 477-485. (36) Wold, S. Technometrics 1978, 20, 397-405. (37) Wold, S. Quant. Struct-Act. Relat. 1991, 10, 191-193. (38) Geladi, P.; Kowalski, B. R. Anal. Chim. Acta 1986, 185, 1-17. (39) Pitts, J. R.; van Cauwenberghe, K. A.; Grosjean, D.; Schmid, J. P.; Fitz, D. R.; Belser, W. L.; Knudson, G. B.; Hynds, P. M. Science 1978, 202, 515-519. (40) Abass, M. K.; Andrews, G. E.; Williams, P. T.; Bartle, K. D.; Davies, I. L.; Tanui, L. K. SAE Tech. Pap. Ser. 1988, No. 880345. (41) Andrews, G. E.; Iheozor-Ejiofor, I. E.; Pang, S. W.; Oeapipatanakul, S. I. Mech. Eng. (Marcel Dekker) 1983, No. c73/83.

Received for review December 23, 1994. Revised manuscript received July 25, 1995. Accepted July 26, 1995.X ES940772T X

Abstract published in Advance ACS Abstracts, October 1, 1995.

VOL. 30, NO. 1, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

49