360
Anal. Chem. 1904, 56,368-373
(22) Tomer, K. B.; Crow, F. W.; Gross, M. L. J. Am. Chem. SOC. 1983, 105, 5487-5488. (23) Gross, M. L.; McCrery, D.; Crow, F.; Torner, K. B.; Pope, M. R.; Ciuffetti, L. M.; Knoche, H. W.; Daly, J. M.; Dunkle, L. D. Tetrahedron Left. 1982, 2 3 , 5381-5384. (24) Tomer, K. 8.; Crow, F. W.; Knoche, H. W.; Gross, M. L. Anal. Chem. 1983, 5 5 , 1033-1036. (25) Beynon, J. H.; Cooks, R. G.; Amy, J. W.; Baltinger, W. E.;Ridley, T. Y. Anal. Chem. 1973, 4 5 , 1023A-1031A. (26) Wachs, T.; Bente, P. F., 111, McLafferty, F. W. Int. J. M a s s Spectrom. Ion Phys. 1972, 9 , 333-311. (27) Cooks, R. G.; Beynon, J. H.; Caprioli, R. M.; Lester, G. R. "Metastable Ions"; Elsevier: Amsterdam, 1973. (28) McLucky, S. A.; Schoen, A. E.; Cooks, R. G. J. Am. Cbem. SOC. 1982, 104, 848-850. (29) Cooks, R. G.; Kruger, T. L. J. Am. Chem. SOC. 1977, 9 9 , 1279-1281. (30) McLuckey, S. A.; Cameron, D.; Cooks, R. G. J. Am. Chem. Sot. 1981, 103, 1313-1317.
(31) Libson, K.; Barnett, B. L.; Deutsch, E. Inorg. Chem. 1983, 2 2 , 1695-1704. (32) Cohen, A. I.; Glaven, K. A.; Kronauge, J. F. Biomed. M a s s Spectrom. 1983, 10, 287-291. (33) Clarke, M. J.; Fackler, P. H. I n "Topics of Inorganic and Physical Chemistry: Structure and Bonding 50"; Springer-Veriag: Heldelberg, 1982. (34) Chlotellis, E.; Stassinopoulou, C. I.;Varvarigou, A,; Vavourakl, H. d . M e d . Chem. 1982, 2 5 , 1370-1374. (35) Glish, G. L.; Todd, P. J. Anal. Chem. 1982, 5 4 , 842-843. (36) Lacey, M. J.; Macdonald, C. G. Org. M a s s Spectrom. 1978, 13, 243-247. (37) DeStefano, A. J.; Keough, T. "1983 Abstracts", 1983 Pittsburgh Conference on Analytical Chemlstry and Applied Spectroscopy, Atlantic Clty, NJ, March 1983, Abstract 367.
for review August 1, 1983. Accepted NCNedXr 28,
1983.
Analysis of Smoke Aerosols from Nonflaming Combustion by Pyrolysis/Mass Spectrometry with Pattern Recognition Rushung Tsao and Kent J. Voorhees* Department of Chemistry and Geochemistry, Colorado School of Mines, Golden, Colorado 80401
Pyrolysis/mass spectrometry (Py/MS) combined with pattern recognition procedures has been applied to smoke aerosols to differentiate the materials involved in nonfiaming combustion. The nonvolatile compounds associated with aerosols were found to have characteristic Py/MS patterns which ailowed for identification of the initial fuels. The characteristic peaks were used as a fingerprint for pattern recognition calculations and eventual identification of polymeric source materiais during combustion of a set of mixtures.
The combustion of polymeric material in a flame is a complex process involving many sequential parallel physical and chemical reactions. A simplified mechanism for this process can be hypothesized as occurring in two primary steps:
-
polymer pyrolysates
polymer fragments
+ pyrolysates
stable products (low a n d high molecular weight)
+ soot
The initial pyrolysis step to form complex pyrolysates has been extensively studied for many polymer systems (1). Usually the reported volatile compounds account for only a portion of the degradation weight loss. The material comprising the remainder of the weight loss is most often observed as an aerosol. Characterization of the aerosol material has been done in a number of cases. For example, the yellow oily solid material from urethane smoke has been shown by infrared spectrometry to have a similar structure to the starting polymer ( 2 , 3 ) . This would indicate that much of the aerosol was composed of portions of the original polymer plus lower molecular weight (degraded) species with essentially the same repeating structure as the original polymer. The combusion of the pyrolysates in the flame is postulated as producing stable low molecular weight species plus soot. As an example of the identification of stable compounds surviving the combustion zone, Liao and Browner reported on poly(viny1 chloride) (PVC) degradation (4). These in0003-2700/84/0356-0368$0 1.50/0
Table I. Materials Used in Nonflarning Combustion
material Douglas fir plywood red oak cotton wool
ignition temp,
material
"C
465 480 480
525 650
ignition temp, "C
polyurethane GM-21 polyurethane GM-29 polystyrene poly(viny1 chloride)
395 550 490 600
ABS
575
vestigators, primarily interested in polycyclic aromatic hydrocarbons (PAH) produced from polymer combustion, identified 26 aromatic species in PVC smoke. Large quantities of literature on the formation of soot are available (5). In general, this literature summarizes the formation of soot as a variable process depending on such parameters as flame type (6), flame temperature ( I ) , and fuel structure (7,B). Surprisingly, very little structural difference has been observed in soot produced from various pure gases (6). Soot is usually carbon with less than 1% hydrogen or oxygen. In addition to the previously mentioned flame processes occurring for the pyrolysates in step 2, the polymer fragments produced in the initial pyrolysis step are also speculated to have some further thermal degradation; however, because of the nature of the polymer backbone, a great deal of the original polymer structure remains unchanged. Pyrolysis/mass spectrometry (Py/MS) has been shown to be useful to differentiate, to classify, and sometimes to characterize nonvolatile macromolecules (9-12). When combined with modern pattern recognition procedures, the technique offers a rapid method for fingerprinting followed by classification. It was speculated, therefore, that the technique should be useful in classifying aerosol materials from combustion. Since in flaming and nonflaming combustion the conditions are considerably different, it was expected that different aerosol materials should be produced from the two modes. The nonflaming mode, because of the postulated simplicity 0 1984 American Chemical Society
ANALYTICAL CHEMISTRY, VOL. 56, INT.
QUADRUPOLE
11 1
COIL *SOURCE
Flgure 1. Curie-point pyrolysis mass spectrometry system.
of the products, was chosen for an initial study. The results of a study of 10 polymeric materials in a nonflaming mode are discussed.
EXPERIMENTAL SECTION A total of five natural and five synthetic polymers were selected as the major materials for this study. Table I summarizes these materials along with their measured ignition temperature. These materials were decomposed individually and as mixtures in a 1.5 f t X 2 ft x 2 f t Plexiglas chamber using the Potts furnace (13). Pure materials were decomposed 25 "C below their ignition temperature for nonflaming degradation. Mixtures were decomposed at about 390 "C for those containing polyurethane GM-21 and 450 "C for the remainder. In order to investigate the effect of decomposition temperature in the nonflaming mode, Douglas fir was also decomposed at two temperatures, 390 "C and 415 "C, while polyurethane GM-29 was decomposed at 440 "C and 500 "C. An iron-Constantan thermocouple was used to monitor the furnace cup temperature. Particles were removed and collected from the chamber with an Emerson pump (flow rate, 10 L/min) connected to a Millipore filter holder containing a Gelman 47-mm glass fiber fiter (pore size 0.4 Mm). The Fiters had been previously extracted with methanol and hexane prior to use, Filters laden with particulate material were initially treated to remove volatiles by three separate procedures. The f i t method was extraction with methanol (6 14.5). Most of the polymeric aerosol material from decomposition of the individual polymers dissolved in methanol. Other filters were extracted with hexane (6 7.24). It appeared that only minimal material dissolved using this approach. The third procedure involved exposing the unextracted Titers to vacuum at 55 "C for various lengths of time. Vacuum treatment using an exposure time of 48 h was eventually chosen as the preferred method. The top layer of the vacuum-treated filters was removed and transferred to an agate mortar and pestle followed by extensive grinding. After the filters were ground, the sample containing the aerosol was transferred to an Eppendorf tube and stored under nitrogen. At the time of analysis, a 2-mg portion was transferred to another Eppendorf tube and 1mL of water added. If the water suspensions were stable after manual shaking for 5 s, the suspension was considered satisfactory. Rotating Curie-point wires (14) were coated with three 5-ML aliquob. After evaporation of the water by a gentle heat stream, a uniform layer of the sample was coated on the wire. The sample on the wire was pyrolyzed by using a Fisher Curie-point pyrolyzer (1.5 kW, 1.1MHz power supply) in conjunction with an Extranuclear SpectrEL computerized quadrupole mass spectrometer system. A schematic of the system is shown in Figure 1. This instrument was designed for high sample throughput without serious ion source contamination. The Curie-point pyrolyzer provides reproducible final temperatures and heating ramps plus a disposable heating element. The open ion source and liquid nitrogen shroud is necessary to minimize the background noise level. The advantages of the Curie-point Py/MS system have been extensively discussed (15). Curie-point wires composed of
NO. 3, MARCH 1984 389
Fe, Ni, and Co (510 "C) were used throughout the study. Low energy electron ionization (15 eV) was used for all work in this study. Meuzelaar and co-workew have shown that 13-15 eV electron ionization produces peaks which are more diagnostic for structural correlation as well as better suited for multivariate statistical analysis than conventional 70-eV ionization (16, 17). This group also established that long-term reproducibility can be obtained by using a low-energy ionization (16,18). A typical scan range of 10-300 amu at 700 amu/s is common. Each of 14 collected samples were analyzed in triplicate and each mass spectrum was collected and stored on a computer disk. Data Analysis. Original raw mass spectral data usually containing 80-100 features were first pattern scaled where each peak was expressed in terms of a percentage of the sum of all ions. Next, the data were autoscaled such that all standardized peaks had a mean value of zero and a standard deviation of one. This transformation scaled each peak with a constant variation; therefore, both the large and small peaks contributed similarly for differentiation between samples. For further data refinement, principal factor analysis (PFA) was applied (19). With PFA mutually orthogonal eigenvectors were consecutively calculated so as to minimize the residual error in each step. Experimental error in the data inevitably leads to a greater number of eigenvectors than the true number of factors. Therefore, only primary eigenvectors which contributed to 85% of the total variance were selected for a regenerated data set. These refined data contained less variability than the original data set. The ARTHUR computer program (20), obtained from Infometrix Inc., Seattle, WA, was used for the multivariate data analysis. From a geometrical point of view, each mass spectrum could be represented as a point in a multidimensional space. To evaluate interspectra variation, a hierarchical clustering method (21) was used. The similarity value, calculated based on Euclidean distance between each possible pair of points, was used as the scale for the dendrogram. The larger this value, the more similar the spectral patterns. Another mathematical method called nonlinear mapping (22) was used for projection of each data point from multidimensional space to two dimensions. A simple model for nonlinear mapping involves connecting springs between all points in multidimensional space and then collapsing the model to two dimensions such that the stress on the springs was minimized. The nonlinear mapping method attempts to conserve interpoint distances and minimize the distortion which occurs due to projecting data from high-dimensional space to a space of lower dimension. Two major mathematical methods were used to quantitatively analyze mixture spectra. If enough pyrolysis data for the mixtures with different compositions were available, factor analysis was used on the assumption that factors were composed of linear combination of measurements. If the identity of the components in a mixture were known, reference spectra were available for each component and regression analysis was applied for mixture resolution. Principal component analysis yields only an abstract solution of which the factors have no real physical or chemical meaning. However, through a graphical rotation procedure (23) by rotating two factor axes at a time, the chemical information concerning the composition of mixtures can be resolved in some cases. The rotated factor axes in this study were evaluated against reference spectra in each step. The method of regression analysis used in this study has been described by Fausett and Weber (24). The computer program fits, through a constrained least-squares procedure, reference or library spectra to a mixture mass spectrum giving both qualitative and quantitative information. The program is valid provided that the assumption that spectral intensities of the mixture are the linear sum of the corresponding intensitiesof the pure components weighted by their compmition. Before running the Fausett-Weber program, a feature selection procedure was used to minimize the redundancy in the data. Only the 20 peaks with the highest variance weights (20) were chosen to run the regression program.
RESULTS AND DISCUSSION Ten selected pyrolysis mass spectra of aerosols from nonflaming conditions are given in Figure 2. Visual inspection of these Py/MS spectra shows some differences and simi-
ANALYTICAL CHEMISTRY, VOL. 56, NO. 3, MARCH 1984
370
loo! 80
P'
'"1
174
/I4*
144
80
B
~
u 40'
DOUGLAS FIR 150
57
M/Z
lool
119
144
PLYWOOD
l.,b M/Z
'7i l144 80
804
1
1
/'05120
,
b , , .
I
loo1 80
POLYSTYRENE
80-
PVC d
COTTON
- BO-
40
20 M/Z
MIZ
WOOL
ABS
-
a
40
20
0
Figure 2. Py/MS
M/Z
spectra of aerosols collected from nonflaming combustion.
larities between samples. Materials of common composition such as vegetative derived materials are in most cases quite similar. The spectra for the aerosols from Douglas fir, red oak, plywood, and cotton contain a large number of peaks which seem to be nearly identical. The higher molecular weight species, principally m / z 124, 138, 150, 152, and 164, from lignin materials can be used to differentiate the individual wood samples. The two urethane spectra show significant differences between each other. GM-29, a rigid urethane, shows a strong peak at m/z 93 which is derived from the isocyanate decomposition. Peaks at m/z 107,119, and 133 are major fragments from the direct cleavage of polyurethane linkage. These peaks were basically absent in GM-21, a flexible foam containing a different isocyanate system, where strong peaks were observed at m / z 122, 148, and 174. A majority of the remaining spectra were significantly different. Visual differentiation of styrene from PVC, for example, can easily be made. One important characteristic of the PVC data was the fact that the mlz 36 and 38 peaks, representing hydrogen chloride, were generated with the correct isotopic ratio in all PVC samples investigated. Figure 3 shows a hierarchical clustering dendrogram of the 10 nonflaming combustion samples. Except for the poly(acrylonitrile-butadiene-styrene) (ABS) aerosol, the high similarity value (about 0.95) for each sample in triplicate analyses illustrates high reproducibility in the experimental
SIMILARITY VALUES
.so
1.00
A
A
=
+
.eo
.70
.BO
-
l
l
'
.50
'
'
.40
l
3 0
l
20
'
3I
8 RED OAK PLYWOOD
h
1-q
COTTON
WOOL
POLYURETHANE OW21 POLYSTYRENE
ABS
4
1.00
.SO
.EO
.70
.BO
.50
.40
.30
.20
Figure 3. Hierarchical clustering dendrogram of the 10 nonflaming combustion samples. Douglas fir was decomposed at (A) 390 OC, (B) 415 OC, and (C) 440 OC; polyurethane GM-29 was decomposed at (A') 525 OC, (B') 500 OC, and (C') 400 'C.
data. The poor reproducibility for the ABS data was probably due to the difficulty in grinding the ABS into fine particles.
ANALYTICAL CHEMISTRY, VOL. 56, NO. 3, MARCH 1984 371
c7 POLYURETHANE GM-21 (SOO'C)&
PLYWOOD
')
OAK
(390&RED
[%DOUGLAS
@!440
(525 C ) P O L Y U R E T H A N E GM-29
a
&
FIR
POLYSTYRENE (440 C)
n
PVC
Figure 4. Nonlinear map of PylMS data from aerosols collected from
the 10 materials. This affected the coating of the Curie-point wire. As expected from visual examination, the three wood samples clustered together with like similarity values. However, the cotton spectra showed differences from the three wood samples. This is expected based on differences in lignin content. The two polyurethanes which were quite different in their spectra patterns showed a rather low similarity value of 0.53. Temperature alternation in the combustion furnace did not affect significantly the spectra of Douglas fir and polyurethane GM-29, although the Douglas fir seemed to show more temperature effect than the polyurethane sample. The nonlinear map derived from the distance matrix is shown in Figure 4. The triangles were formed by connecting all replica data points. The four vegetative samples are located on the left side of the map. One Douglas fir replica fell into the red oak area indicating that within experiment error, the two spectra sets were very similar. The greater separation occurred between the Douglas fir over the urethane GM-29. The spacing of the samples run at different temperatures provided the same conclusion that Douglas fir was more susceptible to temperature changes during combustion than GM-29. Two sets of mixtures have been decomposed in the Potts chamber in the nonflaming mode. In one suite, five materials (Douglas fir, cotton, polystyrene, and two polyurethanes) were formulated with three or four components in each mixture. In the other suite, Douglas fir, wool, urethane GM-29, PVC, and ABS were mixed with three components in each of 10 mixtures. A multidimensional distance calculation and nonlinear mapping were conducted on the mixture Py/MS data. On the basis of the differences observed for the pyrolysis spectra of the pure polymers, it was anticipated that the overall spectra of the mixtures should be different. These facts were observed in the nonlinear map represented in Figure 5. The only structural overlap occurred between mixtures 7 and 10. Both mixtures contained two common components, GM-21 and polystyrene; however, the major factor contributing to the overlap was probably due to the poor reproducibility of the data from mixture 10. This is indicated by a large area encompassed in the triangle for sample mixture 10. Graphic rotation of the factors from principal component analysis was performed on the second set of mixtures mentioned above. By rotation of factors 1 and 2 by 135O, a factor spectrum containing mainly the GM-29 characteristic peaks was generated and is shown in the upper part of Figure 6. Peaks below m / z 60 which are common for most reference spectra are not shown here. The lower part of the spectrum is comprised of portions of the remaining components of the mixture and is generally not used for identification. Similarly,
Flgure 5. Nonlinear map of Py/MS data from aerosols collected from the first 10 mixtures. 1
50
70
90
110
130
150
170
190 2 1 0
110
130
150
170
190 210
1
1
230 1
(
03
I
50
70
90
23C
MI2
Flgure 6.
Factor spectrum after graphical rotation.
wool, PVC, and ABS have been identified in the mixtures with about the same quality of spectra. A factor spectrum of mainly Douglas fir was generated, a 35O rotation of factors 2 and 3 and then a 45O rotation of factors 1and 2 (upper portion of Figure 7 ) . Although most of the peaks can be assigned to Douglas fir, the direct comparison is more difficult than a majority of the other polymers studied. This represents a worst case situation. Least-squares analysis was also conducted on the Py/MS data. These results are summarized for the first set of 10 nonflaming mixtures in Table 11. Because of the lack of literature information concerning the polymer weight to aerosol weight efficiency, it is difficult to assess the quantitative results. However, the results can be used to predict the presence of a particular component involved in the combustion. The overall correct prediction was over 88%. Most incorrect assignments were made between the two urethanes. Results of the other set of 10 mixtures are summarized in Table 111. The number of correct (present - absent) assignments in this data set were comparable to the first experiment. In attempting to evaluate the quantitative results with the goal of understanding the efficiency of converting polymer to aerosol, we note some very interesting trends. PVC in all 10 mixtures had a higher calculated percentage in the aerosol than the original mixture. This would seem that the aerosol formation is very favorable for this polymer. The
372
ANALYTICAL CHEMISTRY, VOL. 56, NO. 3, MARCH 1984 110 130 1 5 0 170 190 210
90
230
Table 111. Least-Squares Results of the Composition of Mixture Aerosols mix- Douglas ture fir
wool
1
2
3.4
24.9 (29.9) 23.7 (32.7) 39.4 (52.3)
3 4
31.2
5
48.9 (50.5) 15.5 (24.1) 45.7 (29.6) 20.4 (48.5) 39.9 (29.8) 14.4 (26.3)
6 7 8
9 30
50
70
110 130 150
90
170 190 210 2 3 0
10
M/Z
Flgure 7. Factor spectrum after graphical rotation.
a
Table 11. Least-Squares Results of the Composition of Mixture Aerosols mixture
GM-21
1
2 3 4
(12.6) 64.0 (40.6)
7
8
(15.6) 72.8 (67.2) I
9 10
Polystyrene
11.5 (5.8) 21.6 (20.5)
6.7 (14.5) 3.5
1.8
14.5 (19.7)
13.2 (11.6) 42.6 (29.2)
5 6
GM-29
46.8
38.9 (19.1) 39.9 (34.0) 7.2 (15.1) 20.2 (23.2) .
,
1.1
cotton 31.7 (10.0)
27.5 (53.5) 53.1 17.9) 44.3 29.2) 46.5 10.8)
2.2 (21.9) 8.2 (20.2) 9.3 (17.7) 14.8' (41.8) 12.9
4.4
\
18.5 (14.1) 30.3
46.4 (20.9)
' Weight percentage of mixtures before combustion. remaining four materials were observed to vary both higher and lower than in the original polymer mixture. The observed scatter could be postulated as a potential weakness in the sample preparation (vacuum treatment). In a real combustion, a large number of materials could be potentially in the original mixture. Instead of taking all reference spectra into least-squares calculation or a targeted factor analysis, prescreening mixture spectra with a series of characteristic peaks is necessary in order to find suspected materials. Table IV lists the major characteristic peaks of each material in our study. Since vegetative materials show almost identical characteristic peaks, this entire class of compounds is placed into one category. The work conducted for this paper clearly shows that the aerosol pyrolysis products are unique for a given polymer. Pyrolysis/mass spectra in conjunction with pattern recognition procedures have been successfully used on smoke aerosols from laboratory generated mixtures to determine the number of components and the identity of the components of the mixture. Based on the approach of classifying the components of the aerosol mixtures by the characteristic ions followed by
ABS
PVC
6.7 (29.4)
5.8 (25.8) 17.0 (26.6)
87.5 (44.8)' 54.5 (43.5) 53.9 (24.9)
22.4 (42.4) 24.7 (23.5) 5.6 (26.8) 33.0 (46.9)
27.0 (25.6) 22.4 (23.3) 46.6 (45.7)
10.3 (24.2) 20.7 (25.3) 2.1 21.3 (23.5) 3.1 13.7 (46.9)
30.4 (24.2) 76.8 (49.1) 49.5 (25.9) 24.0
18.8 20.2 (28.0) Weight percentage of mixtures before combustion.
Table IV. Characteristic Peaks from Nonflaming Combustion
Douglas fir 50.0 (67.7)' 47.3 (25.9) 17.3 (50.8) 13.0 (41.6) 13.5 (57.5) 31.7 (37.5) 47.5 (30.2)
GM-29
material
characteristic peaks
vegetative samples
85, 96, 110, 124, 138, 150, 152, 164 wool 64, 69, 96, 108, 122, 136, 150, 159 polyurethane GM-21 59, 87, 101, 103, 117, 148, 174 polyurethane GM-29 66, 93, 106, 107, 119, 133 polystyrene 78, 92, 104, 105, 120, 122, 134, 148
PVC ABS
3 6 3 8 , 84, 104, 142, 156, 166, 170 78, 91, 94, 104, 118, 130, 144, 156, 170
pattern recognition, successful identification of the fuels in the nonflaming combustion has been greater than 88%. Further work to improve this success rate as well as the investigation of the procedure in flaming conditions are in progress.
ACKNOWLEDGMENT A sincere thanks is extended to Steven Durfee (CSM) and Maya Paabo (NBS) for their help. Registry No. PVC, 9002-86-2; ABS, 9003-56-9;polystyrene, 9003-53-6. LITERATURE CITED (1) Madorsky, S. "Thermal Degradation of Organic Polymers"; Intersclence: New York, 1964. (2) Hileman, F. D.; Voorhees, K. J.; Wojcik, L. H.; Birky, M. M.; Ryan, P. W.; Einhorn, I. N. J. fo/ym. Sei., fo/ym. Chem. Ed. 1975, 13, 571-584. (3) Wooley, W. D. Br. fo/ymn. J. 1972, 4 4 , 27-43. (4) Liao, J. C.; Browner, R. F. Anal. Chem. 1978, 5 0 , 1683-1686. (5) Glassman, I."Phenomenological Models of Soot Processes in Combustion Systems", Report prepared for Air Force Office of Scientific Research, Project #F-49610-78-C-004, July 1979. (6) Palmer, H. B.; Cuills, H. F. I n "The Chemistry and Physics of Carbon"; Marcel Dekker: New York, 1965; Vol. 1, p 265. (7) Clark, A. E.; Hauber, T. G.; Garner, F. H. J. Inst. Pet. Techno/. 1948, 32, 627-635. (8) Schalla, R. L.; Clark, T. P.; McDonald, G. E. Natl. Advis. Comm. Aeronat., Rep. 1954, 1186. (9) Meuzeiaar, H. L. C.; Kistemaker, P. G.; Posthumus, M. A. Biomed. Mass Spectrom. 1974, 1 , 312-319. (10) Meuzelaar, H. L. C. Proceedings of the 26th Annual Conference on Mass Spectrometry and Allied Topics, St. Louis, MO, 1978, p 29. (1 1) Meuzeiaar, H. L. C.; Wood, R. "Rapid Characterization of Coals and Shales by Computerized Py-MS"; Rocky Mountain Fuel Society Meeting, Salt Lake City, UT, Feb 1980. (12) Voorhees, K. J.; Hileman, F. D. J . Anal. Appl. fyro/ysis 1981, 3 , 15 1-1 60.
373
Anal. Chem. 1984, 56,373-376 (13) Levin, B. C.; Foweii, A. J.; Birky, M. M.; Paabo, M. A,; Stoke, A.; Malek, D. “Further Development of a Test Method for the Assessment of Acute Toxicity of Combustion Products”; US. Deptartment of Commerce, National Bureau of Standards: Washington, DC, 1982; NBSIR 82-2532. (14) Giacobbo, H.; Simon, W. Pharm. Acta Heh. 1864, 39, 162-167. (15) Meuzelaar, H. L. C.; Haverkamp, J.; Hileman, F. D. “Pyrolysis Mass Spectrometry of Recent and Fossil Biomaterials; Compendum and Atlas”; Eisevler: Amsterdam, 1982. (18) Meuzelaar, H. L. C.; Posthumus, M. A.; Kistemaker, P. G.; Kistemaker, J. Anal. Chem. 1873, 45, 1546-1549. (17) Posthumus, M. A.; Boerboom, A. J. H.; Meuzeiaar, H. L. C. I n “Advances in Mass Spectrometry”; West, A. R., Ed.; Heyden: London, 1974, VOI. 6. pp 397-402. (18) Windlg, W.; Kistemaker, P. G.; Haverkamp, J.; Meuzeiaar. H. L. C. d . Anal. Appl. Pyrolysis 1878, I , 39-52. (19) Malinowski, E. R.; Howery, D. G. “Factor Analysis in Chemistry”; Wiley: New York, 1980.
(20) Harper, A. M.; Duewer, D. L.; Kowaiski, 8. R.; Fasching, J. L. I n “Chemometrics: Theory and Applications”; Kowalski, 8. R., Ed.; American Chemical Society: Washington, DC, 1977; ACS Symp. Ser. No. 52. (21) Kowaiski, B. R.; Bender, C. F. J . Am. Chem. SOC. 1972, 9 4 , 5632-5639. (22) Kowaiski, B. R.; Bender, C. F. J . A m . Chem. SOC. 1873, 95, 686-693. (23) Rummei, R. J. “Applied Factor Analysis”; Northwestern University Press: Evanston, IL, 1970. (24) Fausett, D. W.; Weber, J. A. Anal. Chem. 1878, 5 0 , 722-731.
RECEIVED for review, June 10,1983. Accepted December 5, 1983. This work has been supported by the Center for Fire Research of the National Bureau of Standards, Grant No. NBSlNADA2020.
Computer-Assisted Determination of Masses in High-Resolution Mass Spectrometry with Selected Ion Monitoring Yves Tondeur, J. Ronald Hass,* Donald J. Harvan, and Philip W. Albro National Institute of Environmental Health Sciences, Laboratory of Environmental Chemistry, P.O. Box 12233, Research Triangle Park, North Carolina 27709
The measurement of accurate masses by hlgh-resolutlon selected Ion monltorlng wlth a computer-controlled peak matching system Is descrlbed. The Influence of various factors on the accuracy of the determlnatlons Is evaluated. Mass accuracles of better than 5 ppm are obtained at hlgh resolutlon and, provlded some minor modlflcatlons are made to standard Instruments, thls analytlcal approach can be used for solvlng speclflc problems requlrlng hlgh resolvlng power from the mass spectrometer In comblnatlon wlth ceplllary gas chromatographic columns. Appllcatlon to samples of tetrachlorodlbenzodloxln at 3-100 pg levels gave mass measurement errors of 0.3-5.9 ppm, wlth standard devlatlons of approximately 2 ppm.
The computer-assisted determination of exact masses by high-resolution mass spectrometry usually involves scanning the magnet over a wide mass range a t some appropriate scan rate. For an exponential scan rate of 10 s/decade of mass at 10000 resolution with a digitization rate of 40000 samples/s, the peak profile of a signal is defined by a certain number of data points ( I ) . In this particular case and assuming a reasonably abundant signal (-100 ions at the collector), the number of data points between the 5% levels is about 17. During an overwhelming portion of the time (>95% typically), no data are being stored by the computer. A widely used method for improving the signal/noise ( S I N ) for the measurement of only selected signals involves stepping the mass spectrometer between these peaks so that dwell time can be increased. Recently this has been extended so that the measurement of exact masses of selected ions at high resolution is possible by recording peak profile data ( 2 , 3 ) . Thus a given peak can be defined by more than 3000 sample resulting in a substantial improvement in the definition of the profile and improvements in SIN. The present study is designed to evaluate this technique of computer-controlled peak matching in terms of precision and accuracy of the determinations.
EXPERIMENTAL SECTION A VG Micromass ZAB-2F mass spectrometer and a digital multiple ion detector unit (DIGMID) coupled to a Finnigan-Incos 2300 data system was used for acquisition and processing of the data. The DIGMID unit enables automatic peak matching of up to seven ions with respect to a reference compound. The reference compound serves as a “lock mass”, because on each cycle its position in the sampled window is checked. If it is not centered, a h5 ppm correction is applied to recenter the peak. The operation of the DIGMID under computer control has been reported elsewhere (3). The resolution of the mass spectrometer was set by use of the 5% crossover definition, and the size of the mass window scanned by the accelerating voltage was usually twice the peak width at 5% height. The scan period was 0.5 s for each monitored ion except for the lock mass ion, which was 0.2 s. The mass spectrometer was operated at 8-kV accelerating voltage with 70-eV E1 ionization energy.
RESULTS The histograms depicted in Figure 1show the accuracy of the exact mass determinations for a given ion at various resolving powers with similar SIN. For instance, a t 5000 resolving power (using a mass window of 400 ppm) one gets an accuracy ranging from 5 to 10 ppm; i.e. for an ion at m / z 350, its exact mass is measured under these conditions within 2 to 4 amu. Examination of the distribution curves shows that increasing the resolution results in an improvement of both the accuracy and the precision. However, in the present conditions there is a limitation regarding the improvement in accuracy with resolving power, which results from the f 5 ppm correction applied by the MID unit on the accelerating voltage each time the lock mass is not in the center of the mass window when sampled. The measurement of the exact masses involves a manual determination of the position of the centroid of the signal. Several factors can affect the measurements, the peak shape being one of them. The distortion of the shape of a signal can be the product of an instrumental error or due to the presence of one or more chemical interferences. It can also originate from statistical fluctuations in the ion beam due to small ion currents.
This article not subject to US. Copyright. Published 1984 by the American Chemical Society