Document wDQ1z1Kkp0wyvn4azjL37aOa4

iTlTi i A. R* Webber PVC Plant Managers C. Scheap A. M. Falrlle MBLO NOtNT OH /MUON OBPOHTMBNT * M.OO. NO. nets setwT os ammo** oveTMcwT * si or,, mo. Cleveland YOUH Uttint OA't THIS lTf|S May 2, 1974 Azbient Vinyl Chloride Analysis In Fedrlcktown's Polymerization Buildings: A Study of Staple Point Locations and a Way to Cover the Build ing Environment More Thoroughly When Using a 6-Point Bendix Sampling System OBJECTIVES 1. Be able to predict floor and building averages with a standard error of less than 10Z (for aaintalnlng an accurate record of VC1 monomer exposure levels). 2. Accomplish the first objective with a six-probe continuous sampling system* 3. Maintain --sensitivity to problems In the building so that plant personnel would be alerted to ''excursions1' of high VC1 monomer concen trations as soon as they occur* INTRODUCTION In February, a study was made of over 100 sets of OVA readings from both the Mass and Pearl-Paste polymerisation buildings at Pedrlcktown* Readings vere taken at 16 points in the Mass Building and at 27 points In the PearlPaste Building* The results of this study indicated the following. If the best 6 points are selected for each building. 1 Bullding Mass Pearl-Paate Points Selected 4,8,9,10,13,15 1,8,12,16,21,27 Standard Error For Predicting Averages 8-3 to 23.7X 2 9.2 to 15.3X 1. Figures 1 and 2 diagram the location of sampling points for the build ings. Tables 1 and 2 summarize the results of the overall statistical analyses* 2. With a standard error of 23.7%, the 95Z confidence limits for predict ing a floor average are: True Value 10 ppm 50 ppm Renge of that Would be Estimated 6.8. 14 7 ppm 34 - 73 ppm With a standard error of 9.7Z, the 95Z confidence limits for predict ing a floor average are: True Value 10 ppm 50 ppm Range of Valuea that Would be Estimated 8.4 - 12 ppm 42 - 60 ppm cv '**< IN BFG31741 2ttilC2 S&2 Bendlx Scapling System May 2, 1974 Fge 2 This range of error Is too great -- and It represents the best set of 6 points that could be found. The "optima" seta of points for each building were: Building Maas Pearl-Paste Ho. of Points Required9 10 9 10 Points Selected 1,4,6,8,9,10,11*13,15 1,5,8,12,14,16,20,21,25,27 Standard Error For Predicting Averages .5.3 to 9.72 6.8 to 9.22 Neither of these solutions aeet all three objectives. A possible solution is to place tees and snail rotameters on the Bendlx probes, (see Figure 3) using them to sample pairs of locations Instead of single points. Working with Pedricktown data (prior to February 4, 1974), the computer was used to simulate paired sampling. This note deals with an evaluation of the results. (Pedricktown has run a similar evaluation of paired sampling for their paste-pearl building -- leading to paired sampling for the Bendlx. This report attempts to evaluate alternatives and objections to paired sampling In more detail.) SUttiAKT Faired sampling with one Bendlx 6-polnt analyser enables the user to predict floor and building averages with almost the same accuracy as would be possible with two Bendlx analysers looking at Individual points. At the same time, all 79 single point excursions above 50 ppm were Identified by using 35 ppm as an action point. Thus, paired sampling makes possible a such more thorough and accurate coverage of ambient vinyl chloride levels in a building without In creasing the investment to two sampling systems. 24523003 RECOMIENDAIIONS 1. All plants should evaluate the "paired" sampling technique (through simulation) and Install It wherever greater accuracy and/or more com plete coverage of work areas Is needed. 2. Plants should use process capability studies and control chart theory to: a. Establish ambient VC1 levels below 50 ppm at which action should be taken to seek out and eliminate sources of high VC1. b. Evaluate point, floor, and building capabilities for maintaining selected levels of ambient VC1. c. Contrast process capabilities of various parts of buildings. As the obvious problem areas come under better control, process capa bility theory can be used to select less obvious problem spots for special work. BFG31742 Bendlx Sapling System May 2, 1974 Fig< 3 3. Don't use exhaust fan data aa the only source for estimating floor and building averages. While these results are useful in shoving the total VC1 removed, they are not the best Indicators of average concentrations in the building based on the data at hand. DISCUSSION A. Contrasting Sampling by Different Methods Seta of sampling locations were selected from several different view points and then evaluated as to their effectiveness In predicting floor and building averages. The following approaches were evaluated. 1. The ''best*' set of 6 locations for estimating floor and building averages (selected by the computer through multiple correlation) 2. The fewest locations which could be used to estimate floor and building averages with a standard deviation of less than 10Z for all estimates (9 points for Mass, 10 points for Pearl and Paata). 3. Paired sampling -- 6 sets -- based upon the most likely points as Indicated by the computer in procedure No. 2, above. 4. Readings at the exhaust fans only (6 for Paste and Pearl, 3 for Mass). 5. Readings of the 3 exhaust fans in the Mass building, plus 3 other likely spots In the building. RESULTS: PERCENT VARIATION EXPLAINED Mass Paste & Pearl Method Floor Mesxaaine Prepoly Bldg, 1st Floor 2nd Floor Bld& 1 (Best 6) 91.5 2 ( Best 9 or 10) 97.2 3 (Paired to 6) 97.9 4 ( Fans) 91*2 5 (Fans +) 94-1 91.1 99.0 99.0 72.4 92.1 98.3 98.3 97.8 91.2 91.2 98.3 99.3 99.3 89.9 97.0 92.6 96.7 96.0 74.9 -- 94.6 98.2 98.0 71.5 -- -- 96.7 98.3 97.9 83.5 PREDICTION ERROR - STANDARD DEVIATION BY PERCENT Mass Paste 6 Pearl Method Floor Mezzanine Prepoly Bldg. 1st Floor 2nd Floor Bldg 1 (Best 6) 16.9 2 (Best 9 or 10) 9.4 23.7 7.4 9.7 8.3 9.7 5.3 13.9 9.2 15.3 8.8 9.2 6.8 BFG31743 fcooczsiiS Bendlx Sampling System May 2, 1974 Page 4 PREDICTION ERROR - STANDARD DEVIATION BY PERCENT - CONT'd Mass Paste & Pearl Method Floor Mezzanine Prepoly Bldg* 1st Floor 2nd Floor Bldg. 3 (Paired to 6) 8.0 4 (Fans) 17.2 5 (Fans + 4 13.9 Points in Bldg.) 7.3 45.6 22.5 11.1 23.6 23.6 5.4 21.2 11.3 10.0 27.1 9.1 - 38.8 7.2 21.7 (Tables 1 and 2 contain more complete contrasting of the results) CONCLUSIONS 1. Sampling individual points using only the exhaust fans or all exhaust fans plus a few select locations within the building is not the best way to estimate floor and building average VC1 concentrations. (This has proven true also at Henry and Avon Lake. Long Beach and Louisville were not evaluated.) 2. Paired sampling to 6 probes on one Bendlx is almost as effective a way to determine floor averages as using 2 Bendlx Instruments for sang)ling. The errors of estimating floor and building averages are consistently 1/2 that of esdiaust air sampling -- or even better. 3. An effective combination of paired samples is as follows: Building 1 Points Mass 6, 1 & 4, 2 & 8, 9 & 10p 11 & 12, 13 & 15 Paste and Pearl 1 & 5, 8, 21, 12 & 14, 16 & 20, 25 & 27 B. Process Capability Studies and Upper Control Limits: Can Paired Sampling Detect Excursions? Process capability evaluations were made on the single and paired points a. To determine average VC1 levels at various locations when excur sions were eliminated. b. To determine what upper control limits for VC1 might be practical if excursions were virtually eliminated. SJttCZSfcZ 1. See the diagrams in Figures 1 and 2 for point locations. BFG31744 Bendlx Stapling System May 2, 197A Page 5 c. To select a base leas than 50 ppm VC1 for palrad result* that might be used a an "action" level. E.g. If this base were ex ceeded we would be 992 certain that VC1 was abnormally high and that the source should be tracked down and eliminated. d. To illustrate how the process capability technique might be used on ambient vinyl chloride data to contrast averages and levels of variability at various points. This would help provide .guidance as to what parts of a building need technical or maintenance attention moat. Areas needing attention are obvious at first. As technology improves and demands for low VC1 becomes greater, this type of guid ance may become increasingly important to the plants. The 99Z (or higher) control limit of paired samples which applied to all sets of paired data was tested on all paired data results to determine whether all excursions of single points greater than 50 ppm would be de tected by paired data results. The date were plotted so that the reader may observe the success of this approach. CONCLUSIONS 1. The following table Hate the averages, process capabilities, and upper control limits of points for die two buildings. The contrasts In aver ages and'uniformity are obvious. Process Capability and Upper Control Limits for Single and Palrad Points HASS BUILDING PASTE & PEARL BUILDING Point Avg. SpCL 99Z Lim Point Avg. SPCL 991 Lim (1+4)/2 1 A 18.3 18.0 19.1 7.10 8.78 7.69 34.9 38.5 37.0 (1+5)/2 1 5 10.2 8.6 11.6 3.79 2.61 5.12 19.0 14.7 23.5 6 11.1 5.70 24.4 8 8.7 3.54 16.9 <2+8)/2 2 8 10.A 10.5 10.3 4.35 A. 26 5.11 20.5 20.4 22.2 (i2+i4)n 12 14 8.2 2.64 8.6 3.05 7.6 2.25 14.3 15.7 12.8 (9+10)/2 9 10 15.2 13.6 15.6 7.65 7.56 6.95 33.0 31.2 31.8 (16+20)n 16 20 15.1 14.2 13.6 6.85 5.83 4.93 31.0 27.7 25.0 (ii+i2)n 11 12 (13+15)/2 13 15 1A.0 13.1 1A.7 17.5 16.2 19.9 7.8A 7.13 7.62 6.88 7.53 7.69 32.2 29.6 32.4 33.5 33.7 37.8 21 (25+27)/2 25 27 17.4 12.7 13.7 9.7 10.17 7.76 7.78 4.46 41.0 30.7 31.8 20.1 BFG31745 24523006 Bendlx Ssapling System May 2, 1974 Paga 6 2. Thirty-five ppm VC1 is an appropriate action point for Pedricktcmm, at which level plant personnel can be at least 992 certain that there la an unusual source of VC1 monomer. (Point 21 in the Paste-Pearl Building la the only point violating this criteria. Its data suggests a flurry of problem over a short period of time which caused its high action limit.) It is probable that the action point for the plant is now lower, based on recent plant improvements. 3. If an action point of 35 ppm is used for paired sampling, essentially all excursions at corresponding single points will be detected. In this study all 79 excursions in the data would have been identified for action. AMF-JCS/cls Atts. A. M. Fairlie 24523007 BFG31746 A P PEND I X A. Simulation of "Paired" Stapling Paired sampling was simulated by averaging the results of each pair of points selected for "blending." Thus, if points 1 and 5 were to be blended, the paired result (point 1 ppm + point 5 ppm)/2. This was done by the computer for each paired set of points and the results analyzed by multiple correlation with floor and building averages. Points were selected for pairing in the following manner. The best set of points for estimating floor and building averages were chosen through multiple correlation. This was restricted to less than 12 points, but demanding that the standard error be less than 10X of the true averages. Working from these points, and keeping on the sane floor, pairs were selected for blending, based on those which were highly correlated with each other. Thus, the least possible informa tion would be lost by blending sample points. Further, when an Excur sion occurred, both points would tend to drift high at the same time. As might be anticipated, the pairs selected in this manner tended to be either adjacent, or corresponding to the normal flow of air through the building. (See Figures 1 and 2) BFG31747 s o o 7 .s w : SECOND FLOOR FIRST FLOOR Beat Set of Six Saplg. Points Optinua Model Sanpllng Points by Pairs as Rqd. Beat Set of Six Saplg. Points Optinua Model Sailing Points by Pairs aa Rqd TABLE 1 PASTE AND PEARL BUILDING 24523009 Bendlx Probes Rqd. 3 5 3 Points Saapled 1,8 12 1.5 8,12 14 1& 5 8 12 & 14 X Variation Explained 92.6 96.7 96.0 Avg. PPM VC1 9.7 9.7 9.7 Standard Error 13.9X 9.2Z 10.OX 95X Limits 281 18Z + 20X Tinea Would Mias Eat. 50 PPM 0 0 0 3 16,21 94.6 14.0 15.3X 311 0 27 5 16,20 98.2 14.0 8.8X 18X 0 21,25 27 3 16 & 20 98.0 14.0 9.IX + 18X 0 21 25 A 27 BFG31748 BUILDING 2nd 1st FI, FANS Only CO co 22 *5f S ST *0 & S 000 O0 J? H0 Q. 0ft BUILDING 2rr CO M 00mB I-** B 0 CO I CAO 00 *o0 6 U> O* Wt MJ Ui w O' *H0 O0O' 0A09 & u0.rM Kcn OH-' Nh- H* a* * oo o* M>IOM<H> Np> w* MNHHUi sj H O' Ki NM Mts MM xl Os 00 CO *0 0O 1 T I0-- 0ft a. w" xj * so xj SO 0U0 O0s 3M ?* 3S o.g 1 83.5 11.7 SO o 21.7% u 00 xMj o 1+ Ln & xl PH oM\ rMo <00 3 s CO ** nsM O' 8 04 ft SO MK> *0 O 0. a. 1+ 1 + r 1 + fL s0 H HC-* 8 it m mh t*l 0O ft 0 H* H a h* yO; (B9 X *0 P *0 0 X0 BFG31749 01 OJ o o Sanpling P oints by P a irs as Rqd. PREPOLY MEZZANINE 3 ft CO CD HM* A H- ft 8 CO -8 M ft 00 "o0 O' CO a *9 I ft a 090 *0 O ss rr a. a CO DO rr *H- Art 1 CO on aA .00 o a Hi H3 O P> 3 rt n FIRST FLOOR 5* " CO I*9 &p 39 OQ o*r) H* JP 3 ft Ou A ? on cc H- A i 8 yft H* H ft CO B oHe* f_t *9 o 0 1 13 & 15 97.8 19.7 K> 00 K> W (O iff" a. G L4 N/i u> so vo 00 00 ww NO Q* c* hoo H SO Hro* NO NO NO NO NO oC/5 3 N> ON tr> tb 00 O' O' M 00 O' 0 oo x* 3*3 *4 w . [9 f<4NO NO NO *W I-* K>NO L/l Iff .VO *4 *4 x* o> NO NO NO <o NO -*4 N 1+ 1+ SO NO Ln Ln HN Ul N ON H 0* o a B M oo 1+ 1+ M O' h LO H O' 0n H u> w U> <n00> o o *""* *o X co W rt 00 o N VO 3 ft NO o a M na 2. Sl+ 00 U> W ftS ko ft >4 NO A NH 0* M9 rr oeMac. HH*. ssS *0 3* BFG31750 to Cl t: Ci o 1 1 .1Z + 22.2X Likely Points MEZZANINE FIRST *0 5 to 3 o. w "9 to o 9H* tP>-r sAM* 3 5 to 3 *9 a H09f0t* to *4 to ff sH< 3 090 3 o9 M* 29 rt cl 9 BUILDING ? ft H- 8 B ? o. to 3 A X to r0r to A 09 o -h 9 O Ui u *9 *1 O C8 6* A AS SO O' 0 a. K * a. S' N H O' O' M H OMW so so sO M * -* * rs> MW HH- 'O ro.Os ps 9* 9* 9* H }|I' lM/i HH* **s_O__OiH HH flO 9* I-* M M 00 ff to O U1 N O o* H H H 4S UI U O 4 0D to 9 0 A0 a. wM sO so sO sO so 00 fa U) w CO h. * 9 ft Aa. Moa> NO NO to ut Oo > cw PLcn U1 U) Os O' OS 3 N> M tit Wl ON H >4 1+ 1+ A* so h* Ml HN O* *4 M TT3J a O* r s -a a s SO M NH 1+ 1+ M UJ * S A* N O* Os *- o o B to n rt >4 ui i*i H 00 to H 3Oft 03Q. *t CL. 1 + 1+ c|+ S H* O' PL5 rt >4 3 Os M c/ a 94 PJ *0ft ocMC H4, S"X*89 500 S BFG31751 (y 4* cn r; CJ o BUILDING PREPOLY FANS Only FANS and 3 Likely Points FANS Only FANS and 3 Likely Points TABLE 2 CONT MASS BUILDING 24523013 Bendlx Probes Rqd. 3 3 3 6 Points Sanpied 1 2 16 1 2 16 \ m 1,2 6 9 11 16 Z Variation Explained 93.2 91.2 89.9 97.0 AVg. PPM VC1 19.7 19.7 15.6 15.6 Standard Error 23.6X 23.6Z 21.2Z 11.3X 95Z Units + 47.2Z Tines Would Miss Est. 50 PPM 0 + 47.2 0 + 42.4Z + 22.6Z 1 by 5 ppm 0 BFG31752 Fig 1-\IU &p,rs /n ftio/kiy Altiosffienz Pedt'tzt<+o^n /?75 QW, M ASS QUlLfifNG- 5-12. IrT O G Z S l OFFICE LEt/eu. BFG31753 F/6.irl/Cil \J&for HfiUH/?7S /jo U/orkttyj f>?r *e- 0*\l,A. PeJtric.VLjt&ry rPOLV BO/U)/N6-- 5`/3 . fAfl /ST FU>0fl^ & H A V S T F&*1$ _ 2452301 East *s Vijf __ Z ft-oQK BFG31754 Fig. %.-Flow Chart for Paired Sampling k X K m r S3 ]Si 9T0ezSfc BFG31755 .. :#* 2 U 1' > o - ,, nM N T Cl rI n* 7 ni Cx>til >To3 n m C Cl Xn g- r. h u> NCl 'n Z n BFG31756 /#s*r 8 frmr x N**rs X 44 iur#M6 2Q u& U z ni D ILTZG E N (JR APm P A P fi.' EUGENE O IETZO TN CD. BFG31757 PotVT 9 Points 9 }/o BFG31758 T NO, 3 4 1 -> D D IE TZG E N GRAPH PAPER lO X 1 0 PER IN C H EUG ENE D IE TZG C N CO. MADE IN U. S, A. rootr if frmr t2 'X /W9 /i sat aatm** NO. 341-10 O IE T 2 G E N G R A P H P A P E R EUG ENE O IETZG EN CO. BFG31760