Document OzkvO0Y6MMvNqKnY219NXJJkX

i vi '4! Characterizing and Discriminating the Shape of Asbestos Particles H. G. Siegrist, Jr., and A. G. Wylie Department of Geology, University of Maryland, College Park, Maryland 20742 Received January 29, 1980 I t The lengths and widths of approximately 1000 particles from each of four asbestos sam ples, two nonfibrous amphiboles, and one talc-serpentine sample were measured utilizing ')$ the SEM. The asbestos samples are commercially available and include chrysotile from Quebec, chrysotile from California, amosite from South Africa, and crocidolite from South Africa. The tremolite and talc-serpentine are from New York, and the riebeckite is from California. Beneficiation, including milling of amosite, crocidolite, and tremolite and air classifications of the two chrysotile samples, was done under commercial conditions; the <.-^j riebeckite was milled in the laboratory. For comparison, the same measurements were made >i?l| on the California chrysotile using TEM. Shape characterizations of the samples are com j|J)pared using regression techniques. The usefulness of the various shape definitions including length, width, and aspect ratio (length/width) in characterizing and discriminating between '' samples is explored and evaluated. Significant results include; (i) Frequency distribution of log length, log width, and log aspect ratio show very apparent differences between asbestos and nonasbestos populations, (ii) Dimensional differences and accurate classification ac- . cording to dimensions are enhanced by regressing log width and/or log aspect ratio against log length, (iii) Discriminant function analysis is able to quantify the distinction between asbestos and nonasbestos particle dimensions such that over 95% of the population assign ments are correct, (iv) Log width is a more efficient classifier than log aspect ratio using either linear regression of discriminant function analysis for these particular samples, (v), _ The choice of instrumentation, i.e., TEM vs SEM, may affect the sample characterization,' ;|| (vi) Quantitative descriptions of the dimensions of small particles may be related to the .''Jr habit of the mineral and the structure of the mineral groups to which the particles belong^-ism However, such mineralogical distinctions are probably not valid for particles whose longest1-*'*'* dimension is less than about I p.m. INTRODUCTION " Morphology has often been used to characterize, define, and differentiate as|M bestos minerals. For example, the Occupational Safety and Health Administratiofei defines an asbestos fiber as any particle of anthophyllite, tremolite, actinolite)| chrysotile, amosite, or crocidolite longer than 5 /xm with an aspect ratio (lengtljp width) of 3 or greater (NIOSH, 1972). This morphological standard originated liftf England as the result of an air-monitoring program in an asbestos textile factory/if The 5-fj.m length was chosen as a lower limit because the optical microscope was)! the instrument being used for monitoring. This choice can be justified based on!j reproducibility studies by Addingley (1966) and Lynch et al. (1970) which shoW/| that the counting of less than 5-/zm fibers can lead to imprecise results. The choice); of an aspect ratio of 3, however, was arbitrary, and not based on any systematic) study. It has yet to be shown that this aspect ratio has any relationship to biologicv cal activity or disease, nor does such an aspect ratio uniquely define asbestos!^' Many materials break, cleave, or crystallize with aspect ratios in excess of 3. 0013-9351/80/060348- 14S02.00/0 Copyright 1980 by Academic Press, Inc. AH rights of reproduction in any form reserved. 348 PLAINTIFF'S EXHIBIT SA-5I7 jjjhpugh asbestos p (Unary tensile strc pscopically. Dimer jjjpestion remains, |tos and distinguis I of the same mine ||(iftiple Descriptiot samples of con p were examined i jjfiy.be used by the [ oral ingestion st> jf^80). The two c ifiolite were air jet Trail mill, all under | .laboratory in a W pf/contaminants. F pay alone was gei spite and talc-ser: j|tjifee minerals. E Hgrpentine based .... whenever me ji'also be noted tl ^qonsisted of ap l^rocidolite is t W'ff Preparati0 K.pl'M examinatio |||ping liquid was lllpfcrsed by hand solution resulted. If,solution was prei |p,:l-p,m Nucleopor Mi.O ml) was dispc ||and poured down pter using a hand n I'lfvith distilled wa Hpjn completely ds St'and mounted on on an SEM S{ P'Sjpheres was plac V. The specimen st WTEM examinati /.*>' /'/TV' m hape of nrytand 20742 lour asbestos sammeasured utilizing Je chrysotile from idolite from South riebeckite is from . tremolite and air ial conditions: the ements were made samples are comfinitions including iminating between icy distribution of between asbestos classification ac'pect ratio against stinction between lopulation assignispect ratio using ular samples, (v) characterization, be related to the particles belong, .es whose longest d differentiate as-v^j th Administrationi'M nolite,' actinolite,'ijj| oect ratio (length/3Si .lard originated in is textile factory. : 1 microscope was ustified based on 970) which show suits. The choice. n any systematic mship to bioiogidefine asbestos, n excess of 3. CHARACTERIZING ASBESTOS 349 though asbestos possesses many unique properties, among them being ex binary tensile strength and flexibility, most are very difficult to measure jscopically. Dimensions, on the other hand, are readily measured. Therefore, jmestion remains, can shape and size uniquely and quantitatively describe l&os and distinguish it from other minerals, especially the more massive var|iof the same mineral? This study was undertaken to address this question. MATERIALS AND METHODS ^Sample Description Sur samples of commerically available asbestos and three samples of nonasjps were examined in this study (Table 1). All the samples except the riebeckite |io be used by the National Institute of Environmental Health Sciences in pal oral ingestion studies. They are more completely described by Campbell et f|980). The two chrysotile samples were air classified, the amosite and ficlolite were air jet milled, and the tremolite and talc-serpentine were ground j>all mill, all under commercial conditions. The riebeckite sample was ground : laboratory in a Wiley mill. None of the samples, except the riebeckite, were |jf contaminants. For the chrysotile, amosite, and crocidolite, however, mormgy alone was generally sufficient to identify the asbestos; in the case of |jilite and talc-serpentine, however, the sample provided was a mixture of (three minerals. Each particle examined was identified either as tremolite or fr,serpentine based on an energy-dispersive X-ray analysis (EDXA). EDXA f|||sed whenever morphology was insufficient in the other samples as well. It |ijid also be noted that amosite is properly a commerical term. This particular le consisted of approximately 95% grunerite asbestos and 5% actinolite as||i. Crocidolite is the variety name for riebeckite asbestos. ff%ttiple Preparation |r SEM examination of the samples, a solution of 4 ml distilled water and 1 ml gashing liquid was mixed in a beaker. To this, 0.5 mg of sample was added feiispersed by hand swirling the beaker. A mixture of 0.1 mg of sample per 1.0 |jf solution resulted. From this, a dilute solution of 0.005 mg of sample per 1.0 of solution was prepared. p0.1-/xm Nucleopore filter was placed in a filtering apparatus. The dilute solu|(1.0 ml) was dispersed in distilled water of sufficient quantity to cover the |',and poured down a glass rod onto the filter. The water was drawn through iter using a hand vacuum pump. After filtration, the sample was rinsed three pi with distilled water to remove the soap. hen completely dry, a square of approximately 25 mm2 was cut from the If and mounted on a carbon disk with double-sided tape. This was in turn lifted on an SEM specimen tab. A small drop of solution containing 1.099-/xm k spheres was placed on one comer of the filter square; these were used for The specimen stub was then copper coated. >r TEM examination, a mixture of 20 ml of water and 0.5 mg of sample was y-.'l ' ,E \ `Hvi;i> !% Vji *'( *, a, SIEGRIST AND WYLIE ed. The mix; iximately 10 n \ Collodion-co. irate. After e\ lian due to st; jay be that sc iy.by breaking ures which jjjfUrasonificatk rsing agent; aningful lengl jtata collection o SEMs were `fyute of Physic; I' Maryland, a |Uurgy Resea iped with ED Ibfthe data co ju.m). TEM s>lengths and ' (CRT) ol |ps. Five sphe ' `to. A new sc instruments v [cation was ent. Widths p|to be repre [jnent setting r ail samples ieasured were length and |RT. This api ithat give the ib words, a lai |ed the same, lured from pi also generate ele numbers. k The lengths were record iefitages.1 'his approach ac> (/.for chrysotile bu Courtesy USBM, Sample No. COF25. *' Courtesy USBM , Sample No. CP80-CP89. '' Courtesy USBM, Sample No. CPM2A-G. 'C ourtesy USBM, Sample No. CP117, 119, 121, 123, 125, 127, 129. CHARACTERIZING ASBESTOS 351 red. The mixture was placed in a sealed jar and ultrasonically dispersed for Ipximately 10 min. After dispersal, several drops of the mixture were placed !; Collodion-coated, 200-mesh copper grid, and the water was allowed to |rate. After evaporation, a fairly even distribution of particles adhered to the Jian due to static charge. iay be that some of these preparation techniques could alter the population |y by breaking up a few fibers or fiber bundles. In designing the experiments, jures which would be more likely to alter the population characteristics, iltrasonification, were kept to a minimum. However, without the addition of jrsing agent and some agitation, clumping of the fibers made the collection jKjaningful length and width data impossible. fata collection jo SEMs were employed in this study: a Cambridge Mark Ila located at the ute of Physical Sciences and Technology, University of Maryland, College ^Maryland, and an AMR 1400 located at the Bureau of Mines, Avondale jjlurgy Research Center, Avondale, Maryland. Both instruments were |ijed with EDXA. SEM rather than TEM was chosen as the technique for jpfthe data collection because of the extremely long lengths of some particles i yum). TEM was utilized only for the short-fiber chrysotile. lengths and widths of all particles were measured directly on the cathode f|ib (CRT) of the SEM using a scale based on the apparent size of latex is. Five spheres were measured and the average diameter was taken as 1.1 iijjh. A new scale was established for every specimen tab and magnification, itstruments were operated at approximately 10,000x to 15,000x but the Ration was increased to approximately 20,000x for accurate width meaftent. Widths were measured perpendicular to length at a point the operator jjplito be representative of the particle. Scale was established on the TEM by iment setting previously calibrated to an NBS standard. Jr all samples studied by SEM, except the short-fiber chrysotile, particles to Measured were chosen by moving the specimen tab in increments and recordlie length and width of the particle whose center fell closest to the center of pfRT. This approach effectively reduces all particles to points and generates |$that give the distributions of particle sizes according to particle numbers. In 1 -words, a large particle and a small particle are each one particle and are Uid the same. For the short-fiber chrysotile, TEM, all lengths and widths were |ured from photographs; all particles falling in a given area were measured. jjf|dso generates data that give distributions of particle dimensions according to ffcle numbers. The short-fiber chrysotile examined by SEM was treated differThe lengths and widths of every particle that crossed a set of evenly spaced Were recorded. This approach generated data that closely reflect volume Stages.1 |This approach accurately reflects volume only when the widths are constant. This is not strictly Hfor chrysotile but the deviation is small and probably does not significantly affect the data. IP K>| : it ' "A ../j . Mil; m If ji 1V* v' s/% ,$ . c'.'a 352 SIEGRIST AND WYLIE RESULTS (a) Frequency Distributions Histograms showing the frequency distributions of log length, log width, and lj aspect ratio are shown in Figs. 1 through 6. They were tested for normality byit X2 goodness-of-fit test2 as summarized in Table 2. While it is visually evident ( most of the distributions are unimodai and approximately symmetrical, many! the computed x2 values are significant at the 95% level of probability; i.e., many the sample frequency distributions imply nonnonnal population distribution. ThtjJ departures from normality could be real or reflect nonrandom sampling of l parent population of minerals, and such departures could present difficulties in til 'llinterpretation of statistical techniques where normality of the distribution is critical assumption. However, differentiation of asbestos from nonasbestos linear regression and/or discriminant function analysis is not sensitive to small! departures from normality. The summary statistics characterizing the frequency distributions in Tablesi|| strongly point to the conclusion that asbestos minerals have a significantly smallf but more uniform log width than do nonasbestos minerals; that they have long^j but more varied log lengths; and that they have over twice the log aspect ratios) nonasbestos, but with significantly higher variation. Statistical techniques con paring variances and means (i.e., analysis of variance, multiple comparison te$j of means) assume normal distributions, a condition not met by our data, so theap 2 The x2 statistic can be used to make statistical comparisons between a set of observed frequencies in each of k categories and a set of theoretical frequencies (in this case from a normal distributip^)|| F,. The statistic computed is TM x2 = (f - F,m- TrHUt H N' 61$ e ;ct 20 30 || H 192 20 U 1i. l 0I l m 1^ Xlebsckfte J IJi ... H Jk .J *m `UHl. *M 8i L*cid i | 2 LR(tk i/um) Fio. I. Frequency distributions of log length for tremolite, talc-serpentine, riebeckite, and short-fiber chrysotile. 80 | lW ft *11 tol 1m ttol p, 2. Frequency distn jite, and crocidolite. I'l |lusions are only * [id. Wiiear Regressio jltt least-squares gmX) are plotted ^fcere significant ^f:ri the correl Iprd 'errors of t Ratios around t gps a descriptiv phtions of log a Regression line. L jti> explain the da Ispect ratio comp Squally impressiv to be the simpl WtQ' ` Kile assume that within pSftlt for the variation fehtion coefficient, su pie correlation coeff m, 1973). An r of 1 f feopship. The statistic Ipsed from tables appc It'Should be noted the It CHARACTERIZING ASBESTOS 353 , log width, and l<j >r normality by i sually evident u, metrical, many^ ulity; i.e., manyn distribution. The n sampling of ,, t difficulties in . distribution is n nonasbestos byi j sensitive to smattl'4 utions in Table'll mificantly smalic , t they have longer^ >g aspect ratios ofti, techniques com4m$ comparison tests'., (i iur data, so theseVl bserved frequenciesit\| normal distribution),' c. riebeckite. and SF Chrysotlle TEM H=I0ET TM 0I l Anosite N on jlj Lf Chrysotile N11348 LC|Q Letth (yum) LX|q Leafth (jum) j. 2. Frequency distributions of log length for short-fiber chrysotile (TEM), long-fiber chrysotile, ijte, and cfocidolite. 1'fit fusions are only visual, albeit the differences are considered strong enough to '.lid. pilear Regressions jght least-squares linear regressions3 of log aspect ratio (Y) versus log particle 3ji (X) are plotted in Fig. 7 and described in Table 3. In each case the regresi'-were significant at the 95% level of probability as indicated by the tabulated S.i of r, the correlation coefficient.4 Also listed in Table 3 are the respective |dard errors of the estimates," which are the standard deviations of the log $t ratios around the computed regression line. The importance of the standard as a descriptive statistic lies in the fact that approximately 95% of the ^rvations of log aspect ratio should be within 2 standard error estimates of Jfegression line. Large standard errors reflect the failure of the least-squares jpto explain the data (Chayes, 1971). |spect ratio comparisons may be considered valid by regulatory agencies, but Squally impressive morphologic characterization and discrimination tool ap is to be the simple linear regression of log width (T) on log length (AT) (Fig. 8).5 'e assume that within the size range measured, a linear model would describe the data, i.e., would iunt for the variation in the dependent variable, log aspect ratio. The significant values of r, the lation coefficient, support the linear model. me correlation coefficient, r. is an expression of the linear relationship between two variables Vis, 1973). An r of I indicates a perfect direct relationship: an r of -1 indicates a perfect inverse ijbnship. The statistical significance of a given value of r for any number of observations can be ;Bsed from tables appearing in most elementary statistics texts. it should be noted that this is not an independent relationship as shown by the following: log aspect ratio = M log length B. log length - log width = M log length B, log width = (1 - M) log length - B. 354 S1EGRIST AND WYLIE TREMOUTE 100 50 180 >o wa 120 O' ee 60 RIEBECKITE JL.il L0G|g WIDTH Fig. 3. Frequency distributions of log width for tremolite, talc-serpentine, riebeckite, and shortf| fiber chrysotile. Stated more strongly, ratios do not add any appreciable information that is dli already obvious from the dimensions themselves. On the contrary, and there i|j extensive examples of this in the literature (Davis, 1973; Chayes, 1971), takitj ratios may either distort or at best cloud the true relationship. The regression linjS of log width on log length are described in Table 4. They are all statisticafif significant. (c) Discriminant Function Analysis f~ In discriminant function analysis, the problem is to find a linear function variables, in this case log length, log width, and log aspect ratio, which producsi 300 u. aool 100 200 uj 100 10G|Q WIDTHl*m> L0G|q WIDTH C""> Fig. 4. Frequency distributions of log width for short-fiber chrysotile (TEM), long-fiber chrysotile, amosite, and crocidolite. , 'Fig. 6. Frequency distributions of log aspect ratio for short-fiber chrysotile (TEM), long-fiber sotile, amosite, and crocidolite. 356 S1EGRIST AND WYLIE TABLE 2 X2 Goodness-of-Fit Values of Log Length, Log Width, and Log Aspect Ratio Sample Tremolite Talc-serpentine Riebeckite Short-fiber chrysotile Short-fiber chrysotile (TEM) Long-fiber chrysotile Amosite Crocidolite Log length 58.4 93.8" 72.1" 47.9 21.6 106.9" 31.8 61.4" Log width 72.4" 60.4" 53.0 101.1" 94.8" 140.3" 29.2 112.1" Log aspect ratio |j 125.2 144.8" 112.5" 24.3 18.1 68.5" 139.7" 10.4 ~11 .J M :1 ,\1 " Significant at 95% level of probability (Davis, 1973). The technique transforms an original set of measurements into a single discrima inant "score" or transformed variable. In this work, the functions were calc&| lated as follows, >1$ Ya = -2.57491 - 5.89958 (log width), . ( Y,, = -1.27804 + 0.87779 (log width), where Ya is the discriminant score for asbestos particles and Yn is the respectiiil score for nonasbestos particles using width as the sole criterion for discriminatidMl If log aspect ratio is introduced as an additional discriminating variable, the digff criminant functions become Ya = -6.92419 - 5.79358 (log width) + 6.12433 (log aspect ratio), Y,, = -2.05436 + 0.91638 (log width) + 2.22936 (log aspect ratio) (ffl M (41 jM Regressi Sample Mmolite jjjjiJ- serpentine |)eckite ifift-Ftber chrysotit irt-fiber chrysotih flng-fiber chrysotile tiosite j^focidolite correlation coef Sfe = SD of observe m jgje functions ca f& .the coefficie % ft ` T = 2.7501 ^plugging in tl >>used to allot pjhe correctne ^particle camt "llhayes, 1971 ittion 74.7% o lfuovs,t.ing th e crite m if' '?! m <;ia mL:v-. 'y? 5E 0 Fig. 7. Least-squares linear regression plots of log aspect ratio as a function of log length. Fig. 8. Least--. g aspect ratio 125.2 144.8" 112.5" 24.3 18.1 68.5" 139.7" 10.4 single discrinwjf is were calcitKj the respective!! liscriminatiotip iable, the digfcji CHARACTERIZING ASBESTOS TABLE 3 Regression Equations, Correlation Coefficients and Standard Error of Log Aspect Ratio vs Log Length hi C Sample Regression equation Log (Aspect, = /(Log Length) ratio) r" jlppiolite Sic-serpentine Hpheckite Mrt-fiber chrysotile (SEM) p6rt-fiber chrysotile (TEM) png-ftber chrysotile Urtiosite afdcidolite Y = 0.334S: + 0.192 Y = 0.254X + 0.134 Y = 0.436* + 0.336 Y = 0.858* + 0.971 Y = 0.763* + 1.234 Y = 0.985* + 0.819 Y = 0.816* + 0.559 Y = 0.858* + 0.709 0.537 0.401 0.565 0.757 0.873 0.959 0.923 0.891 correlation coefficient. |E = SD of observations around the regression line. 357 SE` 0.181 0.165 0.335 0.251 0.220 0.181 0.201 0.193 functions can be reduced and rotated geometrically to a single function the coefficients are independent, the so-called canonical form. It is found Y = 2.75011 + 2.16673 (log width) - 1.25773 (log aspect ratio). (5) ^plugging in the dimensions of a particular particle, this canonical function Mp, used to allocate that particle to either an asbestos or nonasbestos populaKipfjte correctness of these allocations is given in Table 5. For example, given m\particle came from a nonasbestos population, using this discriminant func- *||^yes, 1971) we can correctly allocate that particle to the nonasbestos ition 74.7%_ of the time using the sole criterion of log width or 87.9% of the lasing the criteria combination of log width and log aspect ratio. The correct .M t log length Fig. 8. Least-squares linear regression plots of log width as a function of log length. F 358 S1EUR1SI AND WYLIE TABLE 4 Regression Equations, Correlation Coefficients and Standard Error of Log Width vs. Log Length Sample Regression equation Log width = /(Log length) r" Tremolite Talc-serpentine Riebeckite Short-fiber chrysotile (SEM) Short-fiber chrysotile (TEM) Long-fiber chrysotile Amosite Crocidolite Y = 0.666* - 0.192 T = 0.746* - 0.134 Y = 0.564* - 0.336 Y = 0.142* - 0.971 T = 0.237* - 1.234 Y = 0.015* - 0.820 T = 0.184* - 0.560 Y = 0.142* - 0.709 0.785 0.790 0.663 0.189 0.487 0.050 0.477 0.309 SE6 fl 0.1813 0.165/1 0.335 $1 0.251 v| 0.220 (<f 0.181 ^ 0.201 0.193 . i However, ev fcongly suppor ted lengths, th than do nor |tl excellent lin yell as betwee || samples can (jssion equatic pfebestos, base ^singly obvio iired to diffen Ustos and nona ' r = correlation coefficient jiese populatio ' SE = SD of observations around the regression line especially v. allocations are respectively 95.9 and 97.8% if the particle is originally drawn fronjjf an asbestos population. Jl'ave lengths al! Hips would be ghe discriminar ', DISCUSSION The first approach to morphology characterization, classification, and discrimjfl nation was attempted by comparison of the observed univariate frequency disfl tributions and the summary statistics, i.e., means, standard deviations, an$|f ranges, of log length, log width, and log aspect ratio. While from a visual comparfllp son of these frequency distributions the differences among the populations rTM evident, the nonnormality of many of the sample distributions makes quantitative descriptions such as analyses of variance and probability statements inappropfi fsifier of a sai Elusion is stro 4/,- |rently, to co Id only measui |fv log aspect ftlog length is Jbt ratio; it a pjtnbered that Hijsamples. Ex Bply be done fllfficiently, t TABLE 5 Classification Efficiency of Discriminant Functions ppipumerical ||samples are (a) Criterion: width Discriminant function: Y = 2.75011 + 2.16673 (log width)" llbr, less vari. Icture of asbe Group Percentage correct Number of cases classified Nonasbestos Asbestos led sheets; in Mlh grow in bt Jj|r crystallogr Nonasbestos Asbestos 74.7 95.9 450 5412 (b) Criteria: width and aspect ratio Discriminant function: Y = 2.75011 + 2.16673 (log width) - 1.25773 (log aspect ratio)" Idi The fibril ttyr more vari; '^^^fe;that amosit |j|';less flexibil rlVwidths are Group Percentage correct Number of cases classified Nonasbestos Asbestos Sable widths Rhe-regressioi |ihguish asbe^ Nonasbestos Asbestos 87.9 97.8 Ke slope is in Jfed an index " Equation (5). BO vs log len; pos. Talc-sei ICIENTS AND Length i r" 0.785 0.790 0.663 0.189 0.487 0.050 0.477 0.309 SE6 4 o.isTil 0.165 1 0.335 '1 0.251 ,| 0.220 % 0.181 '% 0.201 v 0.193 ' le is originally drawn from?! lassification, and discrimi-v univariate frequency disvi standard deviations, and! hiVe from a visual compari^l mong the populations arijt| mtions makes quantitative ity statements inappropri?| Functions mber of cases classified bestos 12 13 Asbestos 450 5412 -vfl 773 (log aspect ratio)" nber of cases classified iestos Asbestos 216 5523 CHARACTERIZING ASBESTOS 359 However, even without these statistical tools, we think that the data in Table 3ngly support the conclusions that asbestos particles have longer and more lid lengths, thinner and less varied widths, and greater, more varied aspect |s than do nonasbestos particles. (.excellent linear relationship exists between log aspect ratio and log length, fell as between log width and log length for all samples studied. Each of the ^'samples can be described by unique regression equations. Furthermore, the lesion equations describing the samples fall into two groups, asbestos and Sbestos, based upon slope. The differences between these two groups are ^singly obvious as the length increases, implying that fewer observations are lired to differentiate, either on the basis of aspect ratio or width, between fstos and nonasbestos when the particles are long. Conversely, differentiation pese populations in the shorter lengths demands a larger number of observap, especially when aspect ratio is used. Indeed if a population of particles were iWe lengths all less than approximately 1 (im, differentiation between the two filps..would be very difficult if not impossible, pe discriminant function analysis suggests that log width is a more sensitive S|ifier of a sample particle than either log aspect ratio or log length. This Illusion is stronger if the population being sampled is, in fact, asbestos. Stated Krently, to correctly categorize an asbestos fiber as such, one theoretically l.only measure log width; to correctly classify a nonasbestos cleavage frag ility log aspect ratio should be measured as well. It appears from our analysis fe:|og length is a redundant measurement, except as needed to calculate the pit ratio; it contains almost no unique classifying information. It should be iSiftbered that these conclusions are based on classification of particles from psamples. Extrapolation to the classification of air- or waterborne particles gfpiy be done with extreme caution because these two agents can sort particles efficiently, thereby altering the parent population, peinumerical relationships describing the morphology of the particles in these |t(samples are readily related to the structure and habit of these minerals. The ller, less variable widths of all the asbestos samples can be related to the fibril Jcture of asbestos. In chrysotile, the fibrils are individual tubes produced by fed sheets; in amphibole asbestos the fibrils are single or twinned crystals ncti grow in bundles sharing a common C axis but which have no order in the r crystallographic directions. In both cases the fibrils are small and similarly The fibril structure is best developed in chrysotile. The fibrils are larger and W more variability in the amphiboles, especially amosite. It is interesting to jSYthat amosite is commonly described as being more "harsh" and its fibers i less flexibility than other types of asbestos. The data suggest that the larger tr. iwidths are probably the reason. Cleavage fragments have much larger more l^ble widths than does asbestos and possesses no flexibility, lie regression lines relating log aspect ratio or log width to log length readily Ijjnguish asbestos from nonasbestos based on the slope ofthese lines. The value he slope is in part a measure of the degree of development of fibers and may be pfed an index of fibrosity (Wylie, 1979). For example, considering log aspect lo vs log length, a slope of zero indicates all particles have identical aspect ios. Talc-serpentine particles most closely approximate this shape. They be- /V -Vi 360 SIEGRIST AND WYLIE long to the sheet silicate group and have similar crystallographic dimensions p allel to the sheets, producing nearly equidimensional cleavage fragments wh< crushed. At the other extreme is chrysotile with slopes of the regression lin$ relating log aspect ratio to log width that approach one. This implies a ne; constant particle width, a consequence of chrysotile's fibrillar structure. The oth8 types of asbestos also have by this analysis very high slopes, also reflecting thj fibril structure. In between are the amphibole cleavage fragments. These cleava^ fragments have an elongated nature due to their double-chain structure. Althoi the longer particles tend to have higher aspect ratios, they do not approach dimensions of asbestos and any population of such cleavage fragments should distinguishable from asbestos provided some of their lengths are greater tl approximately 1 /i.m. There is some indication that the frequency distribution of log aspect ratio amosite is bimodal. The amosite sample contains cleavage fragments of grunerit^l which were not distinguished from grunerite asbestos, the major component of th$ amosite sample in this study. In addition, the amosite sample contains about actinolite asbestos which was included and counted as part of the amosite populilj tion. The apparent bimodal distributions may be due to either or both of thejj| "contaminants." The fact that mineral samples are generally not pure must not overlooked in sample population descriptions and great care must be exercised i$| attempting this type of analysis in other studies. Finally, it is appealing to assign the obvious differences between the short-fibl^ chrysotile as characterized by SEM and TEM to the differences in instrument tion. However, the techniques used for gathering the data were different: the TE| was used to collect data based on particle number while the SEM was used|| collect databased on particle volume. Therefore the frequency of log length^ TEM should show a larger number of short particles. The regression equation relating log aspect ratio and/or log width to length are very similar,and cannotijt used ta.distinguish significantly the effects of instrumentation. The comparison'! the frequencies of log width, however, presents a different case. For asbestojjj width is practically independent of length. Therefore, width frequencies shoul not be affected by the nature of the data collection, i.e., number vs volume. It || evident that the frequency of log width by TEM shows many more particles b|| smaller widths than does the SEM characterization of the same sample. This i|| probably a real instrumentation effect. Two fibrils side by side would be seenjf measured and recorded as two fibers by TEM but by SEM they would appear an$ be counted as one thicker fiber. . In summary, it appears from this study that populations can be classified as'i either asbestos or nonasbestos based on morphological considerations. However,1 there may be many mineral particle populations that may not be so clearly distin-. . guished as those we studied. Moreover, the effects of grinding, sampling, im strumentation, and mineralogical characteristics on small particle morphology/ need to be more clearly described by future experiments. ACKNOWLEDGMENTS We gratefully acknowledge the assistance of the personnel at the University of Maryland Computer Science Center and the use of their statistical program library (University of Maryland, 1978). We are ' :;v -'' M traphic dimensions par^J avage fragments wheal of the regression line . This implies a neaijf lar structure. The othfc >es, also reflecting theij gments. These cleava ain structure. Althoug -y do not approach th$| ige fragments should b| ngths are greater than?! ''I n of log aspect ratio ofl fragments of grunerit&l major component of tliejf aple contains about 59m t of the amosite popula*.? either or both of these,')! Ily not pure must not! re must be exercised if] between the short-fiber'J :rences in instruments ^ere different: the TEJ the SEM was used uency of log length by e regression equations / similar and cannot I ion. The comparison of1% nt case. For asbestos, dth frequencies should umber vs volume. It ist. nany more particles of % : same sample. This is '*' >y side would be seen, they would appearand as can be classified as isiderations. However, lot be so clearly distintrinding, sampling, in11 particle morphology CHARACTERIZING ASBESTOS 361 bted to the following persons for the gathering and compiling of the data: T. Barr, J. Bergen, P. |fe, M. Eisner, T. Gore, L. Johnson, M. Kempa, D. Kightlinger, J. Lowry, B. Schrieber, P. fteitzer, R. Reichlin, E. Steel, D. Vroblesky, B. Virta, and P. Wheeless. Our thanks also to G. j>r, Institute of Physical Sciences and Technology, for his assistance in all aspects of the electron ^scopy. This work was supported by the Geology Department, University of Maryland, College |j' and grant from the Bureau of Mines to the Geology Department. REFERENCES iihgley, C. F. (1966). Asbestos dust and its measurement. Ann. Occup. Hyg. 9, 73-82. Spbell, W., Huggins, C., and Wylie, A. (1980). Chemical and physical characterization of amosite, H'bhrysotiie, crocidolite, and nonfibrous tremolite for oral ingestion studies by the National InstiKtute of Environmental Health Sciences. U.S. Bur. Mines Rep. Invest. #8452, 63 pages, fees, F. (1971). "Ratio correlation," pp. 1-98. Univ. of Chicago Press, Chicago. ` Wis, J. C. (1973). "Statistics and Data Analysis in Geology. Chap. 3, pp. 64-123, Chap. 5, pp. g,I92-203. Wiley, New York. Jjh, J. R., Ayer, H. E. and Johnson, D. L. (1970). The interrelationships of selected asbestos Sexposure indices. Amer. Ind. Hyg. J. 31, No. 5, 598-604. ifclal Institute for Occupational Safety and Health (NIOSH) (1972). Criteria for a Recommended Standard for Occupational Exposure to Asbestos. Federal Register jHefsity of Maryland (1978). "Basic Programs for Univac 1108," Computer Science Center, Uni|j|yersity of Maryland, College Park., Md. (U0M*REGRESSION: UM*CHISQUARE.) Bfe', A. G., (1979). Fiber length and aspect ratio of some selected asbestos samples. Proceeding of gpVorkshop #1, Conference of the Scientific Basis for the Public Control of Environmental Health hazards, June 26, 1978. Ann. N.Y. Acad. Sci. 330, 611 iM'ty of Maryland Computer of Maryland. 1978). We are 1980 lide........................ Arinobu Ishiza- Renal Dysfunc- lanoma Mortality clor 1242 and Rehynchos). . . . ALD C. Mauns. s in the Brain of ND J. L. Epler. Chemical Muta- 'D E. Morrison. rption.................. Zaidi. Experi- i of Thyroid and 'ho Salmon {On- if Inhaled Endo- li Poisoning and ies of the Heart. DERICK J. Mil lary Susceptibil- of Manganese dENZEL. Stimres by Pollutant -trialfy Exposed late-Foam Fac- 1 13 24 29 35 42 54 68 77 87 98 110 121 137