-^"^ "'JA^i- % 0"^ . ' • o, "^o • o"^ s • . , ^ »'"' v^J-^ ^ * ..s* A .^^"V •v ^ 0^ s.., ^. ""'' . °^ *'-• >^' ':^o^ ^5 ("^ .»v '«• :^ /«: %^^ ^.Mk^ \/ «l%^ %.^^ :m '"''^ """ A^ "V ''- «» »^ bV 'o V *.r." x:->:^-rvP^v< j^ia^ --o.*^ ;^-. --^0^ A-, %,.^ « %/ ,^, ^.^/ ..^, V„.- - ^O-v-. „ 3P'^1,, .^^°- ^0" v^*„^o- <^' ^o^^^^-^/ ^^^TfTo'^^^" ^^'rff^^'jP \ ^^ c*^ * ^^^' "^^^^-^Z '^^^^^!^\/ ^o.^^5*,o'^ \'^ ."J^^ ^oV" i-i'" .i 'bV" ^'=.v^„ C .r. IC 8968 Bureau of Mines information Circular/1984 r Performance Evaluation of a Real-Time Aerosol Monitor By K. L. Williams and R. J. Timko ??^ UNITED STATES DEPARTMENT OF THE INTERIOR Information Circular 8968 Performance Evaluation of a Real-Time Aerosol Monitor By K. L. Williams and R. J. Timko UNITED STATES DEPARTMENT OF THE INTERIOR William P. Clark, Secretary BUREAU OF MINES Robert C. Norton, Director {\0' Library of Congress Cataloging in Publication Data: Williams, Kenneth L., 1952- Performance evaluation of a real-time aerosol monitor. (Information circular; 8968) Bibliography: p. 16. Supt. of Docs, no.: I 28.27:8968. 1, Mine dusts— Measurement— Instruments. 2. Aerosols— Measure- ment— Instruments. I. Timko, Robert J. II. Title. III. Title: Perform- ance evaluation of the RAM-1. IV. Series: Information circular (United States. Bureau of Mines) ; 8968. TN295.U4 [TN312] 622s [622'.8] 83-600363 ^. CONTENTS U^ Page § b Abstract 1 ^ Introduction 2 ^ Equipment and procedure 2 Test chamber 2 "^(^y^ Llppmann-type sampler arrangement 3 ^ Gravimetric samplers i 3 ^, The RAM-1 3 Tubing dust losses 5 ^\^~" Test procedures 6 Results and discussion 6 Uncertainty In the gravimetric measurement 7 Linearity of RAM-1 response 9 Effect of particle characteristics 10 Size or density 11 Shape factor 12 Surface properties 13 Calibration 13 RAM-1 precision 13 Conclusions 15 Recommendations 16 References. 16 Appendix A. — Dust losses In tubing 17 Appendix B, — Recommendations on the calibration of the RAM-1 20 ILLUSTRATIONS 1. Test setup 3 2. Llppmann-type sampler 3 3. Real-time aerosol monitor 4 4 . Standard and modified cyclone arrangements 5 5. Comparison of resplrable dust concentrations measured by RAM-1 with con- centrations measured gravlmetrlcally for coal 1 7 6. Comparison of resplrable dust concentrations measured by RAM-1 with con- centrations measured gravlmetrlcally for coal 2 8 7. Comparison of resplrable dust concentrations measured by RAM-1 with con- centrations measured gravlmetrlcally for Arizona Road Dust 8 8. Comparison of resplrable dust concentrations measured by RAM-1 with con- centrations measured gravlmetrlcally for limestone 1 9 9. Comparison of resplrable dust concentrations measured by RAM-1 with con- centrations measured gravlmetrlcally for limestone 2 9 Response of RAM-1 to coal 1, coal 2, limestone 1, limestone 2, and Arl- ~^ zona Road Dust....... 11 11. Scanning electron microscope micrograph of coal 1 12 -^ 12. Scanning electron microscope micrograph of coal 2 13 c^A-1. Comparison of resplrable dust concentrations measured with filters Inside _d the dust chamber with concentrations measured with filters outside the dust chamber 17 A-2. Comparison of resplrable coal dust concentrations measured by RAM-1 with concentrations measured gravlmetrlcally inside and outside the dust chamber 19 iio. ii TABLES Page 1. Regression statistics 10 2. Dust parameters 12 3. Ninety-five-percent confidence intervals at Yc = 2.0 mg/m^ 15 UNIT OF MEASURE ABBREVIATIONS USED IN THIS REPORT cm centimeter m^/min cubic meter per minute ft foot mg milligram ft3/cm cubic foot per minute mg/m^ milligram per cubic meter ga gauge min minute gal gallon mm millimeter g/cm3 gram per cubic centimeter ym micrometer h hour pet percent L/min liter per minute yr year PERFORMANCE EVALUATION OF A REAL-TIME AEROSOL MONITOR By K. L. Williams ^ and R. J. Timko2 ABSTRACT The Bureau of Mines laboratory tested the response of a commercially available real-time aerosol monitor (GCA RAM-1) to various dusts. Mon- itor measurements were recorded, averaged, and compared with simultane- ous gravimetric measurements of each test dust. Tests usually lasted several hours. The test dusts of various particle size distributions used included coal, limestone, and a commercially available test dust. For each particular dust, the monitor response was linear and corre- lated well with mass concentration over the range of about 0.5 to 10 mg/m^ . The monitor can estimate a 2.0-mg/m^ respirable coal dust concentra- tion within as little as ±6 pet with 95 pet confidence. The monitor must, however, be calibrated with the dust to be measured because the instrxament response is affected by the type of dust particle. The average monitor response to a mass concentration of coal dust was ap- proximately twice the average monitor response to the same mass concen- tration of limestone dust. 1 Physicist, ^physical scientist. Pittsburgh Research Center, Bureau of Mines, Pittsburgh, PA. INTRODUCTION In 1978, GCACorp., Technology Div. , Bedford, MA, designed and fabricated the RAM-1 for the Bureau of Mines under con- tract H0377092, "Improved Light Scatter- ing Dust Monitor." This report describes the procedure and results of laboratory tests performed by the Bureau to evaluate the response of the RAM-1 to various dusts. The RAM-1 measures resplrable dust con- centrations In the air almost Instantane- ously by using a light-scattering tech- nique. A sampling pump draws air Into a sensing chamber where It passes through the path of a light beam. Dust In the air scatters some of the light to a de- tector that produces an electrical signal proportional to the Intensity of the light. The Intensity of light scattered from dust particles Is a function of the par- ticle size, shape, refractive Index, wavelength of the light, and the angle from which the light Is detected. It Is not a function of the density of the par- ticles , and thus Is not a direct function of the mass of the particles. dust. For that reason, GCA tried to de- sign the RAM-1 to measure the mass con- centration of dust, regardless of parti- cle characteristics such as size, shape, or refractive Index, GCA selected the design based on a dust monitor built ear- lier In the Federal Republic of Germany with similar characteristics. That moni- tor (the Tyndallometer TM-Dlgltal) (O^ successfully reduced the dependence of light scattering on the characteristics of particles In coal mines. The RAM-1, because of Its fast re- sponse, could be used In many mining ap- plications, especially for dust monitor- ing for control of resplrable coal mine dust (2^). Thus, the Bureau tested the response behavior of the Instrument when measuring various test dusts. The RAM-1 was treated as a "black box;" no prior assumptions were made about response be- havior. The questions were — 1. Is the Instrument response with mass concentration? linear 2. Does the Instrument respond differ- ently to different dusts? However, the health hazards posed by various dusts In mines are related to the mass concentration of the particular 3. What Is the estimate for a given type of dust? of precision EQUIPMENT AND PROCEDURE Both the RAM-1 's and gravimetric dust sampling devices were exposed to various concentrations and particle size distri- butions of coal dust, limestone dust,'* and Arizona Road Dust^ (ARD).^ A de- scription of the equipment, test proce- dures, and rationale follows. TEST CHAMBER Figure 1 Illustrates the test setup. The test dust was dispersed by feeding ■^Underlined numbers in parentheses re- fer to items in the list of references preceding the appendixes. ^Typically called "rock dust," lime- stone dust is mixed with dust on surfaces inside coal mines to reduce f lammability. the dry dust Into a 61- by 91- by 122-cm plywood chamber using either a Wright dust feeder or a TSI model 3400 fluld- Ized bed aerosol generator (FBAG) O). Although the FBAG characteristically pro- duced a more constant dust concentration In the chamber, the Wright dust feeder was occasionally needed to obtain higher concentrations. Inside the chamber, a small fan was used to mix the dust and air. ^Reference to specific products does not imply endorsement by the Bureau of Mines. ^ARD is a carefully sized, commercial test dust used primarily to test the ef- ficiency of air filters for internal com- bustion engines. V\\ VVV\V\VVVVVlV\lkkV ; J= 61 cm TOP VIEW SIDE VIEW ^To vacuum source <-- — lO-port, 2-L/min critical orifice monifold Stack sampler to 14.2-L/mm vacuum source RAM-Is FIGURE 1, . Test setup. LIPPMANN-TYPE SAMPLER ARRANGEMENT Experience had indicated that the small fan ensured that no severe concentra- tion gradients existed in the chamber — at least when measurements were averaged over several hours. Nevertheless, to as- sure that all dust sampling devices were exposed to as uniform a dust cloud as possible, a device patterned after the Lippmann sampler (4^) was used. The Lippmann-type sampler (fig. 2) was fabricated from a 5-gal metal can. Ap- proximately 18 cm from the top, a round plexiglass platform was mounted in the can with 20 holes drilled equidistant at a 10-cm radius. Dorr-Oliver cyclones (5- 6^) (sampling heads) were placed in those holes with the inlets facing the center. A lid with a 5-cm-diam hole cut in the center was placed on top of the can. 3! cm Sampling positions SIDE VIEW TOP VIEW FIGURE 2. • Lippmann-type sampler. Dusty air in the chamber entered the device through the hole in the lid and was drawn vertically downward toward the center of the plexiglass platform. Be- cause all cyclones mounted in the plat- form operated at 2-L/min flow rate, the dust cloud was distributed evenly along each sampling cyclone. With this ar- rangement, the coefficient of variation among the 10 gravimetric sampling device measurements was almost always less than 10 pet. GRAVIMETRIC SAMPLERS Ten gravimetric sampling devices were used to make a reference measurement of the mass concentration of dust. Each gravimetric device consisted of a Mine Safety Appliance Co. (MSA) sampling head connected to a critical orifice (7^) con- trolled airflow system. MSA sampling heads consist of Dorr-Oliver 10-mm-diam nylon cyclone particle size classifiers and preweighed 37-mm polyvinylchloride membrane filters in plastic filter cas- settes. The cyclones retain the large, nonrespirable particles and allow respi- rable particles to pass through to the filters. For the tests, critical orifices (CO's) were fabricated from 20-ga hypodermic needles. The orifices were calibrated to 2±0.01 L/min by shortening the needle or modifying the opening. A wet test meter was used to measure airflow through the CO's. Initially, flow rates were checked before and after each test. Later, be- cause the CO's were quite stable, spot checks were relied on to assure a con- stant 2-L/min flow rate through the grav- imetric devices. The components for each gravimetric sampling device were labeled and always used together. That procedure helped in the location and correction of any equip- ment problems. THE RAM-1 The RAM-1 (fig. 3) measures and dis- plays current respirable dust levels in FIGURE 3. - Real-time aerosol monitor (RAM-1). the air. The dusty air is drawn through a lO-mm-diam Dorr-Oliver nylon cyclone, Respirable dust (8^) , the portion of the sampled dust that passes through the cy- clone, enters a light-scattering chamber. Here the instrument detects light scat- tered from the particles at an angle of 70°±25°, The detector converts the light into an electrical signal propor- tional to the amount of dust present in the airstream, A detailed description of the instriiment design has been given in other reports U, 9-10). Circuitry is temperature compensated and protected against humidity. The Bu- reau extensively tested the electronic characteristics of the RAM-1 and found the instrument to be extremely stable. Those test results are discussed in the contract final report (9^) . The RAM-1 's also sampled from inside the Lippmann-type sampler. For the RAM- I'S, MSA sampling heads were modified as follows (fig, 4), The fitting from the cyclone holding bracket normally used to connect the flexible tubing to the exit side of the filter cassette was removed, A short copper sleeve was inserted as a spacer between the top of the bracket and the top of the nylon cyclone. The spacer kept the cyclone vortex finder properly aligned and sealed to the cyclone body. Flexible tubing used to connect each RAM-1 to its sampling head was inserted ji^^Or^^^f^ r^T^ Adjustable knob Flexible plastic tubing To 2-L/min vacuum source Adjustable knob Cyclone Standard Modified FIGURE 4. - Standard (left) and modified (right) cyclone arrangements. through the top of the bracket, passed through the copper sleeve, and then was connected to the exit port of the cyclone. Resplrable dust passing through the cy- clone flowed through the flexible tubing to the entry port of the RAM-1 located outside the chamber. Tubing between the sampling head and the RAM-1 units was limited to 3-ft lengths to minimize dust losses in the tubing. TUBING DUST LOSSES Dust losses can occur when transport- ing dusty air through tubing. Mounting of the cyclone on the RAM-1 inlet and placement of the entire instrument with- in the test chamber could reduce these losses. This was not done in the Bu- reau's tests; however, the test setup used was justified in two ways: First, such an arrangement allowed use of the Lippmann-type sampler so that all sam- pling heads were exposed to the same dust cloud^ and, second, dust losses in the ^Because of the volume occupied by t±ie bodies of the RAM-1 's, the cyclone inlets could not be placed near enough to each other to ensure each instrument sampled the same dust atmosphere. tubing were assximed to be about the same In all cases. A short series of tests was performed to investigate the magnitude and constan- cy of dust losses in the tubing. A dis- cussion of these tests and the results are given in appendix A. Briefly, dust losses in the tubing were less than 10 pet. Because the purpose of the evalua- tion was to compare responses of the RAM-1 to various dusts, not to calibrate the instruments to indicate an absolute value, that small constant bias was not troublesome. However, dust losses were not constant. Variations in dust losses appear as random error of the RAM-1 measurements, causing the RAM-1 measurements to appear to be less precise. If the RAM-1 was operated with the cyclone mounted direct- ly on the inlet, variable tubing losses would be eliminated, and the RAM-1 preci- sion would be slightly higher. TEST PROCEDURES The RAM-1 units were operated from their battery chargers to avoid problems with battery discharge or failure. With the RAM-1 reference scatterer inserted into the light beam, the gain of each RAM-1 unit was adjusted so that the in- strument indicated the calibration value recommended by the manufacturer. Be- fore each test, the zero and gain were checked. It was found that adjustments to gain and zero were rarely needed. Once the RAM-1 units and gravimetric sampling units were prepared, the dust generation system was started and allowed to stabilize. Determination of the dust concentration stabilization (usually af- ter 1 h) was made by observing readings on the RAM-1 units. The flow system for the gravimetric sampling devices was then turned on and recording of the electrical output signal from the RAM-1 units on a strip chart recorder began. Tests lasted approximately 4 h, depending on the con- centration of the test dust. The objective when deciding test dura- tion was to sample long enough to collect at least 1 mg of dust on the filters of the gravimetric sampling devices. The precision of the analytical balance for a single weighing was ±0.01 mg. Weighing precision, when considering preweighing and postweighing necessary to determine the mass of the collected dust, was then 0.014 mg. With a dust mass of at least 1 mg, relative weighing error attributa- ble to the balance, neglecting any error introduced by the operator, was limited to 1.4 pet. After the test, the area under the curve of each RAM-1 recorded output trace was calculated to determine the average reading over the test period. That read- ing was compared with the mean of the 10 gravimetric concentration measurements. The gravimetric concentration (Cone.) was determined using the following equation: Cone, (mg/m^) = 3^ = Am (0.002) (t) (1) where Am = mass of the dust collected on the filter, mg. and t = sampling time, min. The constant, 0.002, is the flow rate of the samplers in cubic meters per minute. Periodically throughout the series of tests, the size distribution of the test dust was measured with an Andersen Mark III stack sampler. The Mark III is an eight-stage cascade impactor with stage size cutoffs ranging from 13.6 to 0.54 pm when operated at 14.2 L/min (0.5 ft^/ min) . RESULTS AND DISCUSSION The intensity of light scattered from dust particles is not a direct function of the mass of the particles. Light- scattering theory (12) states that the intensity of the light scattered by a particle will depend on such things as the wavelength of the source light , the angle between the incident and the detected light, particle index of refrac- tion, and particle size. It does not depend on particle density. Light- scattering behavior is fairly predictable for ideal spherical particles. However, in most situations, particle shape fac- tors complicate the matter. Thus, the theory does not predict a direct func- tional relationship between the intensity of scattered light and the mass of the particle. Any correlation between the intensity of the scattered light and the mass concentration of dust is statistical rather than functional. No exact mathe- matical relationship exists that relates scattered light intensity to particle mass in all cases. For a more complete discussion of statistical versus func- tional relationships, see reference 13. UNCERTAINTY IN THE GRAVIMETRIC MEASUREMENT Figures 5 through 9 show the scatter plots and linear regressions of RAM-1 readings versus gravimetrically deter- mined respirable dust concentrations. Each data pair (x, y) represents the RAM-1 reading averaged over the time of one test (y value) and the mean of 10 gravimetric measurements obtained during the same time period (x value) . The root-mean-square estimated relative standard devi ati on of the gravimetric measurements (RSD) for all tests is 0.11. This value was calculated as follows: E RAM-1 -- 1.08 gravimetric +0.55 .o 12 3 4 GRAVIMETRIC MEAN, mg/m^ FIGURE 5, - Comparison of respirable dust concen- trations measured by RAM-1 with concentrations mea- sured gravimetrically for coal 1. expected uncertainty in a single measure- ment made by one gravimetric sai]q>ler. RSD = [^I(S,/X,)2] ' ', (2) where N = the number of tests, S| = the estimated standard devi- ation of the gravimetric measurements for the ith test. The independent variable plotted in figures 5 through 9, however, is not a single gravimetric measurement. Instead, the mean (X) of 10 such measurements was used to estimate the mass concentration. To determine how well this mean estimates the true concentration of dust (mq^ ^^ measured by gravimetric samplers, the following expression is used: and X| = the mean of the gravimetric measurements for the ith test. Obviously, not every test exhibits a RSD of exactly 0.11; however, this root-mean- square average could roughly indicate the = X ± ^(v,a) ^* (3) where v = n-1 = degrees of freedom, n = number of samples. m ^ 8 a> E (S 6 Q < q: 4 3 2- < a: "^glO ' 1 1 1 RAM-1 = 2. 14 gravimetric r=0.89 -Sy.x =0.77 O o 1 -0.01 o O y^ O/^ O 1 c O Oy/^ X — o >^ v° - % ~ ^^ go '^ \ 1 1 1 1 1 1 - E 8- 5 6 a: CD 4 ^ 2 I ^ to 8 E 1 1 1 1 1 1 1 - RAIV1-1= 2.50 gravimetric + 0.20 y" r=0.95 o ^5^ Sy.x =0.66 o/^ ° X °o/^° - <9^^ 5^ - - ^^ 8 - -""^ 1 - ,- 6- z Q < cr ^ o 5 2- < ^ r RAM-l = 1.73 gravimetric + 1.14 r = 0.93 o o Sy.x = 0.48 o I 2 3 ^ GRAVIMETRIC MEAN, mg/m^ FIGURE 6. - Comparison of respirable dust concen- trations measured by RAM-1 with concentrations mea- sured gravimetrically for coal 2, and 1-a = confidence level, S = estimated standard deviation of the samples. In this case, n = 10 and v = 9. From a table of values for t(^^c(), it is found E eT z Q < UJ q: 10 8 6 4 2 < ro E CD Z Q < UJ CC O Z < — I 1 1 1 I I I r RAM-1 = 0.94 gravitnetric +0.57 ^ r = 0.95 ■Sy.x=0.65 q ! I I I I I I J L 10 8 6 4 2 — I 1 1 1 1 1 r RAM-1 = 0.82 gravimetric + 0.88 r= 0.97 •Syx =0.50 _L I 23456789 10 GRAVIMETRIC MEAN, mg/m^ FIGLHRE 7, - Comparison of respirable dust concen- trations measured by RAM-1 with concentrations mea- sured gravimetrically for Arizona Road Dust, that for a confidence level of 95 pet, t(9, 0.05) = 2.262. As an example, calculate the 95-pct confidence interval for the true concen- tration using equation 3 when the sample mean (X) is equal to 2.0 mg/m^. If RSD = 0.11, then a likely estimated standard deviation (S) would be 0.22, since S = (RSD)X = (0.11)(2.0). Substituting in- to equation 3 shows Pq ~ 2.0±0.16 mg/m^ ~ 2 mg/m^ ±8 pet. In other words, if the mean of 10 gravimetric mass measurements for a particular test is 2.0 and the es- timated standard deviation of one mea- surement is 0.22, then a 95-pct assurance exists that the true concentration in the en E o 4- < LU cr < 5 2- 3 I < 1 1 1 RAM-1= 0.75 gravimetric r = 0.82 - Sy.x =0.66 1 + 0.56 o o 1 ^ ^ - o o ^^ y^ o %^ o o o - y^ ^ o> h E o 5 z n < 4 UJ a: < 6 H Z 7 Z) -H s \ < a: ro 7 E 1 1 1 1 1 RAM- 1 = 0.99 gravimetric - 0.04 -r=0.96 . y^- _Sy.x = 0.49 o^ - y^ o - - ° y'^'^ - cr o - y^ 1 1 1 1 1 1 2 3 4 5 6 7 GRAVIMETRIC MEAN, mg/m^ FIGURE 9. - Comparison of respirable dust concen- trations measured by RAM-1 v/ith concentrations mea- sured gravimetrically for limestone 2, test chamber as measured by the reference method is within the interval of 2 mg/m^ ±8 pet. Normally, the independent variable in a regression analysis is defined to be without uncertainty. From the preceding discussion it can_be seen that if the gravimetric mean (X) is used to estimate the dust concentration in the test cham- ber, the uncertainty is not excessive. Therefore, it is assumed that using X as the independent variable in the regres- sion analysis is at least reasonable. Note, however, that any estimate of un- certainty in the RAM-1 measurements will be inflated by the uncertainty in the gravimetric reference measurement and re- sult in an underestimation of the true precision of the RAM-1. LINEARITY OF RAM-1 RESPONSE By visually reviewing the plots in fig- ures 5 through 9, it was concluded that the response of the RAM-1 is linear with respect to mass concentration for each test dust. To verify that conclusion, a statistical randomness test was applied to examine linearity (11) . Briefly, the sequence of signs (+ or -) of the devia- tions of the measured y values (RAM-1 responses) from the corresponding fitted regression line values in order of in- creasing X values (mean gravimetric mea- surement) , was considered. The following were determined with this information: (a) the number of "+" signs, (b) the num- ber of "-" signs, and (c) the number of runs. A run is defined as an uninter- rupted series of one or more of the same sign. Values a and b were used as 10 parameters in a table of critical values for runs. If the observed number of runs was within the two limiting table values, then the null hypothesis of linearity could not be rejected at the significance level of the table elected for use. The randomness test for linearity was performed on each of the plots in figures 5 through 9. In no instance, at a sig- nificance level of 0.05, could the null hypothesis of linearity be rejected; that is, it could not be denied that the re- sponse of the RAM-1 to the various test dusts was linear. There are better tests for linearity than the above randomness test; however, most require that there be more than one observed RAM-1 reading (y value) for each corresponding dust level (x value) . Un- fortunately the dust generation system used in this evaluation could not exactly repeat dust concentrations from day to day to allow repeated RAM-1 readings to be made of identical dust concentrations. where y = RAM-1 reading, mg/m^ , X = mean gravimetric reading, mg/m3 , m = regression slope, and b = y axis intercept, mg/m^, TABLE 1. - Regression statistics RAM-1 Coal 1 ARD Limestone REGRESSION SLOPES A 0.90 0) 1.08 2.14 2.50 1.73 0.94 0) .82 0.75 0) .95 0.99 B 1.08 C 0) CORRELATION COEFFICIENTS A 0.71 0) .81 0.89 .95 .93 0.95 0) .97 0.82 0) .88 0.96 B .96 C (1) STANDARD DEVIATION OF REGRESSION A 0.49 0) .42 0.77 .66 .48 0.65 (1) .50 0.66 0) .64 0.49 B .52 C 0) ^Unit was not available. EFFECT OF PARTICLE CHARACTERISTICS The next question addressed is whether or not the RAM-1 responds differently to different dusts. To do so, the values for the slope of the regression of the RAM-1 versus gravimetric readings for each type of dust are compared. The slope characterizes the instrximent re- sponse well if the response is linear and if the correlation between RAM-1 readings and gravimetric measurements is high. It has already been demonstrated that the RAM-1 response is linear with mass con- centration; now the degree of correlation between the RAM-1 and gravimetric read- ings will be examined. Table 1 shows the values for the slopes, correlation coefficients, and standard deviation of regressions (stan- dard error of estimate) from the linear regression analyses for figures 5 through 9. The regression equation is y = mx + b (4) The sample correlation coefficient (r) is an estimate of the true correla- tion coefficient (p) which is the degree of association between the y and x val- ues in a statistical relationship. The values of r in table 1 range from 0.71 (a fair correlation) to 0.97 (a high cor- relation) , with a test case average of r = 0.89. In general then, the light- scattering signal from the RAM-1 corre- lates well with mass concentration — at least when dust parameters such as size, index of refraction, density, and shape are held constant. Because the RAM-1 response is linear and correlates well with gravimetric mea- surements of mass concentration, the slopes of the regression lines can be reasonably interpreted as the response of the RAM-1 to a particular dust. The slopes can be used to compare the RAM-1 response to different test dusts. Table 1 lists the regression slopes for each RAM-1 unit and for each test dust. Ex- cept for coal 2 cases, the regression 11 slopes center around 0.9, ranging from 0.75 to 1.08. The regression slopes for the tests with coal 2, however, are no- ticeably higher — roughly by a factor of 2. This difference is shown in figure 10. Because of data scatter, regression analysis provides only an estimate of the true regression; that is, some uncertain- ty exists about the true slope and inter- cept of the regression line. The uncer- tainty is a function of the extent of the scatter. Each regression in figure 10 has an associated uncertainty. Does the data scatter around the regression lines in figure 10 nullify the apparent differ- ences among the slopes? The standard es- timate of error (Sy^j^) for the sany>le is a measure of the data scatter around the calculated regression line. If the cal- culated regression is a good estimate of the true relationship between the RAM-1 values and the mass concentration values, then Sy^j^ is a good estimate of Oy.^* the true standard estimate of error. By def- inition, 68 pet of data pairs should lie within the values of ±ay,x» In figure 10, the gray shaded intervals surrounding the regression lines repre- sent S„ ^. For test dusts other than y .X coal 2, individual Sy^^^ intervals overlap to a great extent; the larger shaded area in figure 10, about the regression lines, represent the outermost bounds of all the overlapping S ^^ intervals. The Sy^^ interval for coal 2 does not overlap the others for concentrations greater than 1 or 2 mg/m^. Although more rigorous tests could show the same results mathematical- ly, this simple analysis graphically il- lustrates that the response of the RAM-1 to coal 2 is different than its response to the other dusts. The results for coal 2 are real: The behavior was verified by repeating tests. Efforts were then made to discover why the RAM-1 responded so differently to coal 2 than to coal 1 and the other dusts. What was different about coal 2? The characteristics to be considered are size, density, shape, and surface properties. Size or Density Table 2 lists size distribution and density data for the test dusts. The values for density (pq) were not mea- sured, but were taken from the litera- ture. The mass median aerodynamic diam- eters (MMAD) and geometric standard 4- e CD ■z. Q 3 < LU a: < b 2 < 1 \ / ' ' .yj A I y\f I // *^ / ~ / //x- / / / / //fy^ / / //^ / A "^ I /A r KEY - / /W Coal 1 t/// Coal 2 »p4 h) RAM-1 measurements is equal to or better than that of gravimetric personal samplers, and that once calibrated, the instrument read within ±10 pet of the gravimetrical- ly determined mass concentration. Work of that type should continue to determine the limitations of the instrument when used in the field. REFERENCES 1. Breur, H. Das Feins taub-Streu- lichtphotometer TM Digital (The Fine-Dust Light Scattering Photometer TM Digital). Staub-Reinhalt Luft, v. 36, Jan. 1976, pp. 6-10. 2. Williams, K. L. , and G. H. Schnakenberg, Jr. Direct Measurement of Respirable Dust. Paper in Proceedings of the Fifth WVU Conference on Coal Mine Elect rot echnology , July 30-31, August 1, 1980 (WV Univ., Morgantown, WV) . BuMines OFR 82-81, 1980, pp. 18-1—18-13. 3. Marple, V. A., B. Y. H. Liu, and K. L. Rubow. A Dust Generator for Labo- ratory Use. Am. Ind. Hyg. Assoc. J., v. 39, 1978, pp. 26-32. 4. Blackman, M. W. , and M. Lippmann. Performance Characteristics of the Multi- cyclone Aerosol Sampler. Am. Ind. Hyg. Assoc. J., V. 35, 1974, pp. 311-326. 5. Caplan, K. J., L. J. Doemeny, and S. D. Sorenson. Performance Characteris- tics of the 10-mm Respirable Mass Sam- pler: Pt. I — Monodisperse Studies. Am. Ind. Hyg, Assoc. J., v. 38, 1977, pp. 83- 95. 6. . Performance Characteris- tics of the 10-mm Respirable Mass Sam- pler: Pt. Il-Coal Dust Studies. Am. Ind. Hyg. Assoc. J., v. 38, 1977, pp. 162-173. 7. Corn, M. , and W. Bell, A Tech- nique for Construction of Predictable Low-Capacity Critical Orifices, Am, Ind, Hyg, Assoc, J,, V, 24, 1963, pp, 502- 504, 8. Aerosol Technology Committee. Guide to Respirable Mass Sampling. Am. Ind. Hyg. Assoc. J., v. 31, 1970, pp. 133-137. 9. Lilienfeld, P. Improved Light Scattering Dust Monitor (contract H0377092, GCA Corp.). BuMines OFR 90-79, 1979, 44 pp.; NTIS PB 299 938. 10. Tomb, T. F., H. N. Treaftis, and A. J. Gero. Instantaneous Dust Exposure Monitors. Environ. Int., v, 5, 1981, pp, 85-96, 11. Crow, E. L. , F. A. Davis, and M. W. Maxfield. Statistics Manual. Dover Publications, Inc., New York, 1960, 288 pp. 12. Mie, G. Considerations on the Optics of Turbid Media, Especially Col- loidal Metal Solutions. Ann. Phys. (Leip- zig), V. 25, 1908, pp. 376-445. 13. Natrella, M. G. Experimental Sta- tistics. U.S. Department of Commerce, National Bureau of Standards, Handbook, 1963, p. 91. 14. Tomb, T. F,, and A. J. Gero. De- velopment of a Machine-Mounted Respirable Coal Mine Dust Monitor, Ch, in Aerosols in the Mining and Industrial Work En- vironments, ed, by V. A. Marple and B. Y. H, Liu, Ann Arbor Science, Ann Ar- bor, MI, V. 3, 1983, pp. 647-663. 17 APPENDIX A. —DUST LOSSES IN TUBING The standard procedure used for deter- mining the response of the RAM-1 to vari- ous dusts involved comparing RAM-1 mea- surements with gravimetric measurements. For the gravimetric samples, the respira- ble dust that exits the cyclone deposits immediately on a filter. For the RAM-1, however, respirable dust that exits the cyclone must travel through approximately 3 ft of flexible tubing before reaching the light-scattering sensing chamber. The following tests were performed to ex- amine possible dust losses in the flexi- ble tubing during transport. EXPERIMENTAL PROCEDURE Of the 10 gravimetric samplers normally used to determine the mean mass concen- tration of dust inside the dust chamber, five were modified to be like the ones used in the RAM-1 sampling train (see figure 4 in the main text). The filter cassette immediately atop the cyclone was removed and the flexible tube was at- tached directly to the exit port of the cyclone. The filter was reinserted into the sampling line approximately 3 ft downstream. In these modified gravimet- ric systems, therefore, respirable dust traveled through the same length of flexible tubing as in the RAM-1 sampling train before being collected on the filter. Measurements made using the standard gravimetric sampling trains were desig- nated as "inside" measurements since the filter cassette was located inside the Lippmann-type sampling arrangement with the cyclone. Measurements made using the modified gravimetric sampling trains were designated as "outside" measurements since the filter cassette was located outside the test chamber. EFFECT ON BIAS The mean of the five outside gravimet- ric measurements were compared with the mean of the five inside gravimetric mea- surements. The scatter plot and linear regression are shown in figure A-1. At to F ^ en E 4 r- lU Q !^ ?> (- 3 O fh 2 2 < liJ 2 O I a: t- LlI s > < a: o Out = 1.09 (ln)-0.23 r= 0.90 - Sy.x = 0.28 ± first glance, the slope of 1.09 is dis- turbing since this would imply that dust is not lost in the tubing, but rather is created! However, a regression line is only an estimate of the true relation- ship. One can test the significance of the results in the following way (7^).^ A null hypothesis that the true slope of regression (M) is equal to 1 (that is, no dust is lost in the tubing) is stated. A value, t, can be calculated using the following equation: t = m - M (A-1) where m 1 2 3 4 GRAVIMETRIC MEAN, INSIDE, mg/m^ FIGURE A-1,- Comparison of respirable dust con- centrations measured with filters inside the dust cham- ber (no tubing) with concentrations measured with fil- ters outside the dust chamber (approximately 3 ft of tubing). and M = S„ = the slope of the regression estimate, the slope of the true regression, estimated standard deviation of the value m. ^ Underlined numbers in parentheses re- fer to items in the list of references preceding this appendix. 18 From the regression analysis, m = 1,09, Sn, = 0.086, and M = 1 is selected arbi- trarily. Substituting into equation A-1, t = 1.05. This t value is compared with a t distribution table value t(Qj/2, n-2) where a is the significance level and n is the number of tests. In the re- ported case, a = 0.05 and n = 39 were selected. The table value t(o,o25 37) = 2.03. Now since t > t(Qt/2, n-2) > the null hypothesis cannot be rejected at the 0.05 significance level; that is, there is no statistically based reason to conclude that M is not equal to 1. Based on the data, it cannot be stated with any certainty that dust is lost in the tubing. by The confidence interval for M is given M = m ± t(a/2, n-2)(Sm). (A-2) Substituting into equation A-2, M = 1.09 ±0,17 or 0,91 < M < 1,26, From the physics of the situation, M > 1 would not be expected because the tubing cannot create dust. However, dust losses in the tubing could be expected to be less than 10 pet, 95 times out of 100, The preceding discussion deals with dust losses, or biases (systematic er- rors) in the comparison of RAM-1 readings with gravimetric readings. Such biases could be important when calibrating the RAM-1 to indicate mass concentration as determined by gravimetric devices. How- ever, since the tests were to examine re- sponse behavior, this bias, if consist- ent, is not significant. What must be determined, however, is whether trans- porting the dust through the tubing in- troduces more random error. Since the same lengths of tubing are used in the RAM-1 sampling trains, any random error introduced by dust losses in the tubing would appear as less precision in the RAM-1 measurement. EFFECT ON PRECISION In figure A-2, the amount of data scat- ter about the estimated regression line is higher when the RAM-1 measurements are compared with gravimetric measurements made through a length of tubing (outside) than when the RAM-1 measurements are com- pared with gravimetric measurements made immediately after the cyclone (inside). Sy^x is an estimate of the data scatter about the regression line. For RAM-1, unit C, Sy^x increased from 0.42 to 0.55, or by 31 pet. For RAM-1, unit A, Sy^x increased from 0.49 to 0,58, or by 18 pet. On the average, the random error introduced by drawing the dust through the tubing caused a 24-pct increase in the value of Sy^x* Therefore, one can safely assume that some of the scatter about estimated regression lines for RAM-1 measurements (all drawn through tubing) compared with gravimetric mea- surements made inside the dust chamber (no tubing) is due to random dust losses in the tubing to the RAM-1, in addition to that random measurement error inherent in the RAM-1. If this indeed is true, the estimates of the ability of the RAM-1 to reproducibly predict the true mass concentration are conservative. 19 CO E o z o < 0:4 < I- 1 1 1 T" 1 ^ RAM-1 = 0.90 gravimetric + 0.34 ° 1 y^ y - r = 07l ° 0^ y^ _Sy.x=0.49 ^^ „ S'^ o°X^ yb ~ /^ - - / Q^ - 1 1 1 1 1 A 1 - to e o» E cT z Q < UJ o I 1 1 1 1 1 1 RAM-l=l.08gravimetric+0.55 ^ r = 0.8l ° ^oy^ / -Q =042 ° ^^ - ° y/^° - B 1 1 1 1 1 1 1 - => 4 < 1 1 1 1 1 1 r RAM-l=0.72gravimetric+l.53 ^ ° r=0.64 o o ■Sy.x=0.55 ° o o ± ± _l_ 3 4 I 2 3^ GRAVIMETRIC MEAN, mg/m3 FIGURE A-2. - Comparison of respirable coal dust concentrations measured by RAM-1 with concen- trations measured gravimetrically inside {A^ B) and outside {C, D) the dust chamber. 20 APPENDIX B. —RECOMMENDATIONS ON THE CALIBRATION OF THE RAM-1 Previous Bureau work showed that the zero and gain of the RAM-1 are exception- ally stable (9^). These tests have shown that for a given test dust, the RAM-1 re- sponse is linear and correlates well with mass concentration. Based on these find- ings , one can reasonably expect that — 1, The RAM-1 can be calibrated to in- dicate respirable mass concentrations di- rectly in milligrams per cubic meter. 2. The RAM-1 must be calibrated for each type of dust to be measured. Once calibrated, the reliability of sub- sequent measurements will then depend on the variability of the properties of the particles in the dust cloud. To calibrate the RAM-1, the user must compare RAM-1 and gravimetric measure- ments as was done in these tests. The user can then either adjust the gain of the RAM-1 so that the instrument displays the proper concentration values , or develop a calibration curve to convert displayed values to the true mass concentration. TUBING LOSSES In either case, the user should be aware that some dust losses can occur if the dust is transported through long lengths of flexible tubing (see appen- dix A) . Since the intent of this eval- uation was only to observe response behavior, systematic losses were not important. However, if the user wishes to use the RAM-1 to obtain absolute val- ues of dust concentrations, he or she should calibrate the instrument in the configuration in which it will later be used. In other words, if the application of the instrument will require that the sampled dust be drawn through tubing to the RAM-1 sensor, then the instrument should be calibrated using tubing of the same approximate length and material. That procedure will at least reduce bias in the measurements although the preci- sion may still be measurably reduced. DEVELOPING A CALIBRATION CURVE In examining the behavior of the RAM-1, target concentrations of 1, 2, 4, and 10 mg/m^ were arbitrarily selected. The conclusions were based on regressions for data in that concentration range. Ex- trapolating these regressions to very high concentrations could conceivably lead to very large errors. If the user wishes to make measurements at such high concentrations, he or she should make some comparative measurements at those levels during the calibration procedure. Since the RAM-1 was tested at concen- trations above nominally 1 mg/m^ , little or no meaning was attached to the y in- tercept of the regression equations. A functional relationship to be used for calibration, however, would be greatly simplified if the y intercept were zero. Therefore, the user, when developing a calibration curve, may wish to perform the regression analysis to force the line through the origin. Many statistics books, including Natrella (13) , offer procedures for such a regression. Before using such a procedure, however, one should be sure that the RAM-1 does indeed indicate zero in dust-free air. Manufacturer data and earlier Bureau data (9^) indicate that the RAM-1 's zero indi- cation, once adjusted in a dust-free environment, is not affected by tempera- ture, humidity, etc. Background scatter- ing resulting from dust contamination of the optics would cause a zero shift. However, it was found that the clean-air sheath (9^) over the optical surfaces suc- cessfully prevents deposition of dust. A properly adjusted RAM-1 should, there- fore, read mg/m^ when no dust is pres- ent in the air. Although forcing the regression through zero can be justified, the user should nevertheless make comparison measurements at concentrations near mg/m^ . These measurements would (a) establish linear- ity in the low concentration range and (b) estimate the precision of the RAM-1 at low concentrations. INT.-BU.OF MINES, PGH., PA. 27347 H 2«9 «A .^ \.-^-^*/ V^^,***" V^'^**/ %''i^--.*^ -o .-i^^ ^^'%. " -^^^^^ < "ov" » .0 ^ *bv" • ^^' .A^-^^ v^ . t ' » f -^ :''^I'^"'^ / ''&,^''^--^^^ '^^'^^^-^^ -'M,^^^-'-''^ ^ V i9^. - ^^^.^"' ■^J^M^^^ ^^Mc^ ^i^^B^". ^o-^ ^^^^- '^^^ ^'^^M\ ^O'^ "oV ^AO^ :< -n.-o^ r He <^ 'J".-;* .G^ \5 "^7^^ .A. <;;^ 'o.»* G^ "^ ^:n