Document baEyRE3V8O77KvXB4ndmwnZ1y

182 March-April, 1966 Computer Output The basic computer output consists of the mean concentration, standard deviation, num ber of analyses recorded, and percentage of time the concentration was above preselected levels, all at each location for each shift dur ing the survey. Shift or daily averages can also be obtained over any selected time inter val such as a week or a month. These data, further identified as to department, air con taminant, etc., are permanently stored on magnetic tape. The computer has been programed to re ject poor data such as that caused by a stick ing punch or by a faulty encoder. If poor data are being obtained, this fact is often signaled first by an unexplained decrease in Furthermore, assume that, over the time period of interest, concentrations at these data locations behaved in this manner: % of Time Concentration Exceeds Selected Values at Specified Locations Concentration Data Locauon Number greater than 1 7 4 12 25 ppm 50 100 250 500 26.63% 28.68% 93.44% 31.55' 16.39 22.13 20.08 23.77 6.55 10.65 17.62 11.47 1.63 2.04 16.39 3.68 0.00 0.00 0.00 0.00 The time-weighted percentage of time spent by operators in concentrations above 25 ppm will then be: .(12J) (26.63) + (31.3) (28.63) + (18.8) (93.44) + (25.C) (31.35) + (12.4) (0) 100 38.92 the number of analyses used by the computer This means that on the average, during the to obtain the shift averages. This has become time period of interest, men in the opera the normal signal for nonroutine maintenance of the digitizing equipment. In addition to the basic output, the com puter calculates time-weighted averages of two kinds. The first kind is the "usual" timeweighted average concentration to which men are exposed. It is obtained by combining the air analysis data with "job analysis" infor tors classification encountered concentrations above 25 ppm 38.92% of the time. Similar calculations show that these men encountered concentrations above 50 ppm 28.24% of the time; above 100 ppm 10.95% of the time; above 250 ppm 4.00% of the time; and above 500 ppm 0.00% of the time. mation on the percentage of time spent by men in the vicinity of specific analyzer probes. Job analysis information is also combined with the data on the percentage of time dur ing which concentrations exceed the prese lected levels. This results in a second kind of average which is the "time-weighted percen tage of time" exposures exceeded the prese lected levels. An example of how the time-weighted per centage of time is calculated is given below. Assume that the following information is true Data Analysis Automatic air monitoring reveals that in an industrial situation the variation of con centration with time can be large despite the use of a rather large time base. Figure l is a plot of daily time-weighted average concen trations to which operators were exposed on each shift over one week. Each point on the graph is the mean of several hundred deter minations spaced equally over an eight-hour period. These are actual plant data. for an operator in a plant: Data Location Percentage of Time Spent Number at That Location 1 12.5 7 31.3 4 18.8 On the second shift there was little varia tion during the first four days, but over the last three days of this period the shift average varied bv a factor of almost four. If the threshold limit value (TLV1 for this mate rial were 50 ppm, an industrial hvgienist tak 12 Unexposcd 25.0 ing air samples during the first three days on 12.4 the second shift would probably have de clared that the hazard to health was low or 100.0 nonexistent. On the other hand, if he had 151 American Industrial Hygiene Association Journal 183 Figure 1. Daily time-weighted average exposures for each shift from January 30 to February 5. Time-weighted average for the week (all shifts) was 32.32 ppm. sampled on the last day of this period (on the second shift! he could have experienced a strong inclination to "push the panic button." Variation of concentration between shifts can also be striking, as illustrated by a com parison of the third shift with the first or second shifts during the first three days. Figure 2 places Figure 1 in a broader con text, Data for Figure 2 are also time-weighted average concentrations to which operators were exposed. The rather extreme variation shown in Figure 1 for a time basis of one shift can be repeated for a time basis of one week. The information contained in Figure 2 is based upon hundreds of thousands of indi vidual air sampies. Average concentrations and time-weighted average concentrations are not the onlv data summaries provided by the computer: aver age concentrations are accompanied by the standard deviation of the data. Having the standard deviation and using it quantitatively are two different things, however. We do not yet know how to use such information effi ciently. On the other hand, the spread or varia bility of the data is indicated in a more mean ingful manner by the percentage of time the concentration exceeded certain levels. Figure 3 is a plot of the time-weighted percentage of time above these levels on log-probability paper. The interval during which data were gathered and the operational classification arc identical to those in Figure 2. This graph shows that the median concentration to which men on all shifts in this classification were exposed was 40 ppm and that they were exposed to 500 ppm or higher 2Co of the time. The conventional plot (Figure 2) shows a maximum concentration of about 135 ppm. but it is one of seven-day averages, whereas Figure 3 is a summation of instantaneous values. Both kinds of graph have advantages. A conventional plot illustrates better how exposutes varv with time, and anv trends be come readily apparent. With this kind ot graph, however, the oniv usable index of ex posure is the time-weighted average, a num ber of limited utility because it cannot reflect 152  'if'V-' I V Reprinted from AMERICAN INDUSTRIAL HYGIENE ASSOCIATION JOURNAL Volume 27. March-Apnl. 1966 The Application of Computer Science to Industrial Hygiene J. E. PETERSON, H. R. HOYLE, and E. J. SCHNEIDER The Biochemical Research Laboratory, The Dow Chemical Company, Midland, Michigan Automatic sampling and analysis of environmental atmospheres can result in voluminous amounts of data describing exposures to chemicals. The use of a digital computer to process such data and the advantages and disadvantages of the technique are presented. Introduction A UTOMATIC AIR monitoring equipment is expensive. Nevertheless, if it is used to signal the need for action by operating per sonnel to eliminate leaks and to effect needed repairs of equipment, its cost can easily be jusdfied in terms of reduced exposures, re duced hazards, and reduced loss of process material. Furthermore, if sample locations are properly chosen, it provides data on the concentrations of air contaminants to which men are exposed. That information, handled properly, can be used to show long-term trends of exposure which, correlated with plant operations, can result in still lower ex posures. In addition, such data can be used to determine the exposures of workmen in what is. in effect, a continuing industrial hygiene survey. These data, in turn, could be correlated with medical information on effects of the exposures to confirm or denv present standards or to suggest new ones. At The Dow Chemical Company, auto matic air sampling and analysis .vere first used in 1950 to monitor air concentrations of carbon tetrachloride in a production plant. At that time no method was available for handling the mass of data generated bv the air monitoring equipment. Even though the instrument was successfully rmploved hv plant Presented at the \nmiiil Mverimr of the American Confer ence of (invernmemai industrial ilv<iemsu, Houston. Texas, Mav supervision to control exposures, full use of the data to estimate the average concentra tion to which men were exposed was not feasible. To illustrate: The instrument rec orded air concentrations at the rate of one every six seconds; this is 432,000 times per month. Because the first continuous air monitor was a success at day-to-day control, others were installed in several plants during the 1950's. As each monitor was put into use. even more information on workmen s ex posures was "going to waste" simply because there was too much of it. The advent of digital computers suggested a solution. Tak ing advantage of the fact that the Computa tions Research Laboratory had a Burroughs B220 computer, the Environmental Research Laboratory in 1961 acquired a machine for translating air concentration data from a recorder to punched paper tape. Continuous air concentration data could then be com bined with computer analysis to yield a much more complete description of inhalation ex posures than had heretofore been possible. This paper recounts some of the problems involved, from the installation of air moni toring equipment to the interpretation of the data as summarized by the computer. Actual data arc used, but only for illustrative pur poses, so plant, process, and even the air con taminant are immaterial. This is not a report ot an environmental survey: it is a primer in the application of computer science to indus trial hygiene. ISO ddv" 153 American Industrial Hygiene Association Journal 181 Obtaining Air Samples day number, hour, and minute. A switch An automatic air monitoring system must be sensitive enough to detect concentrations of significance to health. Sensitivity must range from concentrations that may be im mediately hazardous to those that have little or no significance even for prolonged, re peated exposures. The instrument should be selective; it must respond only to the mate rial (s'! of interest. It must be stable; it should be unaffected by minor (or even major i variadons of temperature, humidity, vibration, and line voltage. The instrument must be capable of operating unattended for relatively long periods of time. Daily atten tion by the operator, minor service weekly by an instrument man, and occasional perform ance testing should be adequate to keep it in operation. Finally, because a permanent rec ord of air concentrations can be important, concentration data must be automatically re corded. Any continuous analyzer that satisfies these specifications can be used. So far we have used only combustion-conductivity in struments and long-path gas cell infrared spectrometers; both types produce data suit able for computer analysis. Single-point monitoring in a production plant is seldom economical. Usually the analyzer is positioned at a central location in selects a "type of data" digit (from 0 to 9) which is incorporated into the digitizer out put. An extra "type of data" (for example, from an infrared spectrometer or from a com bustion-conductivity anaivzer! digit can allow the computer to reject data that obviously do nots^long with that being processed. To minimize costs, punched paper tape is used to transfer data from the digitizer to the computer. For each datum the digitizer punches two "words" on the tape. The first "word" con tains the day number, hour, and minute, as weil as the "type of data" identification num ber. The second "word" contains the probe number and the actual datum which, in this case, is a number from 0 to 999, propor tional to the concentration of the air con taminant. This information is obtained at a rate that may vary from two times a minute to once in 2.5 minutes. Because collecting each datum generated is not always neces sary, the digitizer can be programed to skip the collection of some data. In every case, once data from probe No. 1 are punched onto the tape, data from the other probes are obtained in serial order, but the digitizer can ignore ail but every second, fifth, or tenth sets of analyses. the plant, and air samples are brought con tinuously to it through probes of some sort. Programing the Computer Tubing used for these probes must not react We decided that the minimum time in the with, or sorb, the air contaminant of interest. data summary should be an eight-hour shift. For this period the Dow Computations Re Preparing Data for Analysis search Laboratory programed the computer to calculate the mean concentration at each There is no method presently available for reading results automatically from multiplepoint chart paper into a computer. This means that a device must be used to "digi location, the standard deviation of these data, the percentage of time that the concentration was above several preselected levels, and the appropriate time-weighted averages. tize" the data as they are obtained. At the minimum, the digitizer must translate the analog data (usually the position of a wiper on the slide wire of a potentiometer) to digi tal form (for example, numbers) that can be handled by the computer. Computers are versatile. For instance, if one probe is located to sample air outside the building, the computer can be programed to correct automatically for deviations in the "background," It can take into account peculiarities in the calibration curve of the Because data on the time of exposure can instrument used for air analysis and can re minimize programing difficulties, the digitizer cognize several kinds of errors in the data includes a digital clock with an output of the and use only "good" data for calculations. oov 1 184 March-April, 1966 Figure 2. Weekly tfme-'.vetjhted average exposures for each shift from January 16 to Aoni 10, Thne-weiihtea a\erase tor tile whole period (all shiftsi was 79.09 ppm. `These points are expanded in Figure 1. concentration variations. A log-probability plot such as that shown in Figure 3 is not very useful as a trend indi cator, but the fact that time-weighted a/wjutc data plot as a straight line on this graph paper oilers exciting possibilities. In our limited experience all time-weighted percent age of time data have been best fitted by a straight line on log-probabilitv paper. That Figure .1. Dam lor the period ut time shown in Figure 2 fall shuts'. I he .ur monitor was not beinst used bv oneratine personnel to assist in the control of leaks. straight line is a complete summary of ex posure information, including measures both of the average concentration and of the con centration variability.1 The data in Figure 3 were obtained by a survey using the equipment described. How are these data to be interpreted? If we still assume a TLV of 50 ppm, is the exposure represented by Figure 3 a hazardous one or not? These data are among the most com plete e\er gathered for men industrially ex posed to a potentially hazardous vapor; what kind ut animal experiments should be con ducted to interpret these exposures? We cannut yet answer these questions, but we know iliac they must be answered eventually. Despite this void in our knowledge, the data trom this type of survey have several uses. Any intormation about the intensity of exposures t.cips to intetpret medical findings, liven without complete interpretation, rank ing ut exposures is possible. From graphs such as Figtue 3 one can judge if the exposuie is obtiously hazardous or nonhazar(ious. For instance, in Figure 4 the concentrations to which the men were ex posed ate shown to exceed 15 ppm oO'T ot 155