1、INSTITUTE OF ENVIRONMENTAL SCIENCES AND TECHNOLOGY Contamination Control Division Technical Guide 104 IEST-G-C104 Sequential-Sampling Plan for Use in Clasification of the Particulate Cleanlines of Air in Cleanrooms and Clean Zones INSTITUTE OF ENVIRONMENTAL SCIENCES AND TECHNOLOGY Arlington Place On
2、e 2340 S. Arlington Heights Road, Suite 10 Arlington Heights, IL 6005-4516 Phone: (847) 981-010 Fax: (847) 981-4130 E-mail: iestiest.org Web: ww.iest.org 2 Copyrighted material INSTITUTE OF ENVIRONMENTAL SCIENCES AND TECHNOLOGY IEST-G-C104 This Guide is published by the INSTITUTE OF ENVIRONMENTAL SC
3、IENCES AND TECHNOLOGY to advance contamination con-trol technology and the technical and enginering sciences. Its use is entirely voluntary, and determination of its aplicability and suitability for any particular use is solely the responsibility of the user. Copyright 199 by the INSTITUTE OF ENVIRO
4、NMENTAL SCIENCES AND TECHNOLOGY ISBN 978-1-87862-7-9 No part of this publication may be copied or reproduced in any form, or by any means, including electronic, mechanical, photocopying, or otherwise, without the prior writen permision of the Institute of Environmental Sciences and Technol-ogy, 2340
5、 S. Arlington Heights Road, Arlington Heights, IL 6005-4516. INSTITUTE OF ENVIRONMENTAL SCIENCES AND TECHNOLOGY Arlington Place One 2340 S. Arlington Heights Road, Suite 10 Arlington Heights, IL 6005-4516 Phone: (847) 981-010 Fax: (847) 981-4130 E-mail: iestiest.org Web: ww.iest.org IEST-G-CC1004 IN
6、STITUTE OF ENVIRONMENTAL SCIENCES AND TECHNOLOGY Copyrighted material 3DeSequential-Sampling Plan for Use inClassification of the Particulate Cleanlinessof Air in Cleanrooms and Clean ZonesteIEST-G-CC1004CONTENTSSECTION1 Application. 52 Definitions 63 Background 64 Operating characteristic curve. 65
7、 Objectives . 76 Sequential-sampling plan. 87 References. 11APPENDIXAPPENDIX A; Multiple Evaluations . 12APPENDIX B; Cost Savings Example 13FIGUREFigure 1. Operating characteristic (OC) curve 7Figure 2. Plot of sequential-sampling domains 8Figure 3. Sequential-sampling count histories 9Figure 4. Ave
8、rage sample duration vs. true mean count. 10Figure 5. Confidence intervals based on sample counts . 10Figure A1. Fractions accepted for samples of 21 counts . 124 Copyrighted material IEST-G-CC1004 INSTITUTE OF ENVIRONMENTAL SCIENCES AND TECHNOLOGYIEST-G-CC1004 INSTITUTE OF ENVIRONMENTAL SCIENCES AN
9、D TECHNOLOGY Copyrighted material 5INSTITUTE OF ENVIRONMENTAL SCIENCES AND TECHNOLOGYContamination Control DivisionGuide 1004Sequential-Sampling Plan for Use inClassification of the Particulate Cleanlinessof Air in Cleanrooms and Clean ZonesIEST-G-CC1004This guide is based on a paper by Cooper and M
10、ilholland(reference 7a).1 ApplicationThis guide provides background and support forthe use of sequential-sampling plans for the classi-fication of air cleanliness in cleanrooms and cleanzones. Such plans are generally most appropriatefor environments where air cleanliness is expectedto qualify as IS
11、O Class 4 or cleaner.The sequential-sampling plan described hereinhas been developed to match the operating charac-teristic (OC) curve of the traditional single-stage-sampling plan. As a result, the probability of pass-ing or failing a given classification is about the samefor both plans.When compar
12、ed to sampling times required bytraditional single-stage-sampling methods, how-ever, sequential sampling is found to be capable ofsaving about 80 percent of the sampling time whenthe air sampled is very clean, 35 percent when theair just meets the class limit, and more than 80percent when the concen
13、tration of particles in theair exceeds 2.5 times the class limit.Sequential sampling is especially applicable asan alternative to a single-stage-sampling scheme incases where the upper confidence limit (UCL) onthe grand mean of the particle concentration is notrequired or when, with some further dev
14、elopment,it can be used to provide an equivalent estimate ofthe UCL.NOTE: The statistical validity of data from long-term sampling of very clean air at a givenlocation is based upon the assumption that theparticle concentration will remain relativelyconstant or will vary only in a random mannerdurin
15、g the sampling period. Statistical validitymay be degraded, however, if unique phenom-ena are present that can cause significantvariations in concentration.6 Copyrighted material IEST-G-CC1004 INSTITUTE OF ENVIRONMENTAL SCIENCES AND TECHNOLOGY2 Definitions2.1 classificationa) A relative measure of a
16、ir cleanliness, expressedin terms of a standard system of class limits such asthat defined by ISO 14644-1.b) The process of determining the class limit con-centration of particles.2.2 class limitThe maximum allowable concentration of particles(in particlesper cubic meter of air), of a specifiedparti
17、cle size and larger, as determined by the equa-tion upon which the classification system is based.2.3 pass/fail decisionThe result of the analysis of sampling data andcomparison with specified class limits, wherebythe determination of class verification is made.2.4 sequential-sampling planAn alterna
18、tive method of gathering particle con-centration data by multistage sampling which,under appropriate circumstances, can enable moreefficient and cost-effective determination of pass/fail decisions regarding classification.2.5 single-stage-sampling planThe sampling protocol that involves collectingsa
19、mples of air whose volume is at least equivalentto the volume that would be required if the particleconcentration were at the class limit being verified.This sampling protocol is also known as the fixed-sampling plan.3 BackgroundStandard practice for the classification of air incleanrooms, such as t
20、he procedure described in ISO14644-1 (reference 7b), calls for the acquisition ofone or more samples of air at each of a specifiednumber of locations throughout the space beingclassified. The minimum volume of each of thesesamples of air is defined as the volume that wouldtheoretically contain 20 pa
21、rticles, if the particleconcentration in the air being sampled were at thedesignated class limit.In order to meet the specified classification, theaverage particle concentration at each samplinglocation must not exceed the specified class-limitconcentration. In addition, an upper confidencelimit (UC
22、L) calculated from all of the location aver-ages must not exceed the specified class-limit con-centration. If both of these requirements are met,the air cleanliness in the space being sampled issaid to meet a certain class limit, as defined anddetermined in accordance with ISO 14644-1. Typi-cally, e
23、nvironments used for microelectronicsmanufacture are cleaner than ISO Class 5, for ex-ample.The advantages of sequential sampling can beshown by comparing it with the more customaryprocedure of single-stage sampling. The protocolfor single-stage sampling involves the collection ofcomplete, discrete
24、samples of fixed extent. Sequen-tial sampling, as introduced by Wald (reference 7c),is based on the comparison of the running (cumu-lative) count total against a limit that is a functionof the amount (extent) of sampling completed.Wald noted that sequential sampling generallyentails less sampling (a
25、nd, consequently, less time)than any single-stage-sampling plan having thesame probabilities of false acceptance and falserejection.In classification testing of a cleanroom that isexpected to meet ISO Class 5 (per ISO 14644-1),reduced sampling time may not be significant, as aparticle counter operat
26、ing at a volume flow rate of28.3 L/min (1 ft3/min) would register 20 particlesin the size range 0.5 m and larger in only 0.2 min,if the air being sampled were at the class limit.Sampling in a much cleaner environment, how-ever, can extend the time required to register thenecessary 20 counts to impra
27、ctical extremes. Thisextension of time is the result of the combinedeffects of fewer available particles present in the airof lower particle concentration, and the lowersample flow rate that is typical of particle countersor condensation nucleus counters that are capableof detecting particles small
28、enough to be (statisti-cally) present in sufficient quantity. For example,classification of the air in a cleanroom that is ex-pected to meet ISO Class 1, using a particle count-ing system capable of counting particles 0.1 m andlarger at a volume flow rate of 2.83 L/min (0.1 ft3/min) would require a
29、sampling time of 706 min ateach location.4 Operating characteristic curveFigure 1 shows the operating characteristic (OC)curve, the curve that results from counting par-ticles in air according to the single-stage-samplingtechnique at a single location in accordance withIEST-G-CC1004 INSTITUTE OF ENV
30、IRONMENTAL SCIENCES AND TECHNOLOGY Copyrighted material 7ISO 14644-1. It represents the probability of passing(count 21) versus the true (unknown) mean count,based on Poisson counting statistics. The magni-tude of the sample (which has a volume equal to thesample flow rate times the duration of samp
31、ling) issuch that a mean count of 20 is expected when theparticle concentration in the air being sampled is atthe specified class limit.If the measured count does not exceed 20, thesample is within limits and thus passes. If themeasured count is 21 or more, the sample fails,although the option is av
32、ailable to perform addi-tional sampling and compare the average of all ofthe samples at a given location to the criterion of 20counts.The curve in Figure 1 shows the probability, forISO Class 5 conditions, that a sample will contain 20or fewer particles if the true (population) mean is asindicated a
33、nd the sample is taken from a constantand uniform concentration, which gives a Poissondistribution. The probabilities can either be readilycalculated or obtained from standard mathematicstables (reference 7d). The OC curve shows, amongother things, that:a) In 95 percent (1 ) of the instances in whic
34、h thesample is taken from air with a true mean count of14, the sample will pass (have 20 or fewer particles).b) In 56 percent of the instances in which the sampleis taken from air right at the class limit, the samplewill pass.c) In 5 percent () of the instances in which thesample is taken from air w
35、ith a true mean count of29, the sample will pass.For clean zones with more than about four loca-tions, passing or failing this criterion is a strictertest of air cleanliness than the upper confidencelimit (UCL) test (reference 7e). (See Appendix A.)ISO 14644-1 requires that the UCL test be appliedwh
36、en there are fewer than 10 sampling locations.For 10 or more locations, passing or failing based onthe count average at each location determineswhether classification passes or fails.5 ObjectivesThe objectives of this guide are:a) To establish a sequential-sampling plan that hasan OC curve with the
37、same and as the OC curveof the single-stage-sampling procedure of ISO14644-1; andb) To investigate the savings in both sampling timeand associated costs that might result.Figure 1. Operating characteristic (OC) curve. Probability of passing (count21) versus true (unknown) mean count, for Poisson cou
38、nting statistics. ThisOC curve is plotted along with points obtained from simulations done using thesequential-sampling plan of Figure 2, each point being the average of 5000simulated samples.TRUE MEAN COUNT Single-stage samplingSequential samplingFRACTION ACCEPTED (COUNT 21)8 Copyrighted material I
39、EST-G-CC1004 INSTITUTE OF ENVIRONMENTAL SCIENCES AND TECHNOLOGY6 Sequential-sampling planA graphical representation of the format of the planis shown in Figure 2. The number of counts ob-served (C) is plotted against the number of countsexpected (E) for the situation in which the particleconcentrati
40、on of the air being sampled is at the classlimit. A full single-stage sample corresponds to E =20. In the example given previously, which in-volved sampled air with a concentration of 10particles, 0.1 m and larger, per cubic meter (ISOClass 1), the full single-stage sample E = 20 corre-sponds to a s
41、ampling time of 706 min with a sam-pling airflow rate of 2.83 L/min.The domains in Figure 2 were derived fromequations reported in works by Wald (reference 7c)and Duncan (reference 7f), as explained by Cooper(reference 7g). The upper and lower limit bound-aries of the graph are derived from the equa
42、tions:Upper limits: C = 3.96 + 1.03ELower limits: C = 3.96 + 1.03EThis plan has been truncated by design so thatthe single-stage sample time (E = 20) is its longesttime. A third limit was found necessary, such thatC = 20, in order to match the OC for the single-stage-sampling plan (see Figure 1) wit
43、h the sequentialplan.The use of Figure 2 involves recording the cumu-lative count of observed particles, while at the sametime comparing that count, as sampling continues,with the upper and lower plotted lines on thegraph. This comparison should be made almostcontinuously, with many more than 20 com
44、pari-sons scheduled during the full extent of sampling(E = 20). This process usually entails some form ofcomputerized data analysis.As counting of particles in the sample proceeds,if the cumulative observed count crosses the upperboundary of the CONTINUE COUNTING zone,then sampling is stopped and th
45、e air is judged tohave failed to meet the specified class limit. If, onthe other hand, the cumulative observed countcrosses the lower boundary, sampling is stoppedand the air is judged to have met the specified classlimit. If the cumulative observed counts are 20 orless at the end of the prescribed
46、sampling period,having crossed neither the upper nor the lowerboundary, the air being sampled is also judged tohave passed.In practice, the number of counts cannot becontinuously compared against these criteria. In-stead, comparison is made after small incrementsof time, in this case taken as thousa
47、ndths of the totalsingle-stage sample duration (E = 20), as suggestedby Fujii et al. (reference 7h).Figure 2. Plot of sequential-sampling domains. Observed count (C)versus expected count (E). Full duration of a single-stage sample isE = 20. Regions show where cumulative observed count indicates aire
48、xceeds the class limit (Reject), meets the class limit (Accept), or isindeterminate (Continue counting).RejectAcceptContinuecountingOBSERVED COUNTEXPECTED COUNTSEQUENTIAL-SAMPLING LIMITSIEST-G-CC1004 INSTITUTE OF ENVIRONMENTAL SCIENCES AND TECHNOLOGY Copyrighted material 9ExamplesFigure 3 illustrate
49、s three examples of application ofthe foregoing procedures. One shows the results ofsampling air with a high concentration of particles(denoted by a76, true mean = 29). A count exceed-ing the upper limit is registered at about one-thirdof the time it would have taken a single-stage-sampling scheme to determine failure. Anotherexample shows a medium concentration (denotedby , = 20); in this case, a count exceeding 20causes sampling to stop only 10 percent short of thesingle-stage sample duration. The third examplesh
