A DESTRUCTIVE SAMPLING METHOD DESIGNED FOR HIGH QUALITY PRODUCTION PROCESSES (DSM-HQ) by FRANCISCO DELGADILLO, JR., B.S.E.E., M.B.A. A DISSERTATION IN BUSINESS ADMINISTRATION Submitted to the Graduate Faculty of Texas Tech University in Partial Fulfillment of the Requirements for the Degree of DOCTOR OF PHILOSOPHY Approved Chairperson of the Committee Accepted r "• • *- Dean of the Graduate School December, 2004 ACKNOWLEDGEMENTS This dissertation is dedicated to the memory of my grandmother, Guadalupe Rojo de Montano, affectionately known to all her grandchildren as Gileli, who is gone but never forgotten. I want to extend my deepest appreciation to my committee chairperson. Dr. Ronald Bremer, for his support, guidance, patience and advice that enabled my completion of this research. I also want to express my gratitude to my committee members Dr. James Bums, Dr. John Kobza, and Dr. Paul Randolph for their support, comments and help. Every PhD student should be so lucky to have such great professors, teachers, friends, and mentors. I am extremely thankful to all my family members and friends for their faith in me as well as their encouragement. I especially want to thank my mother, Rosario Montano, for her infinite support and unwavering encouragement, which made possible the completion of all my studies including this research. I particularly want to express my gratitude to my father, Francisco Delgadillo, Sr., for liis words of wisdom and extraordinary support in every imaginable way throughout my educational endeavor. I also want to express my appreciation to my brothers, Javier and Luis Delgadillo, my girlfriend, Olga Posazhennikova, and my very special friends Mike and Bomiie Bowman, for their encouragement, support and exceptional friendship. My special thanlcs go to the people of Capsonic Automotive and Aerospace hic. for their timely help, information, and advice. 11 TABLE OF CONTENTS ACKNOWLEDGEMENTS ii ABSTRACT vii LIST OF TABLES ix LIST OF FIGURES xi CHAPTER L INTRODUCTION 1 1.1 Overview 1 1.2 Problem Statement 5 1.3 Objective of the Research 6 1.4 Document Organization 8 n. LITERATURE REVIEW 9 2.1 Acceptance Sampling 9 2.1.1 Single and Double Sampling 10 2.1.2 Sequential and Multiple Sampling 11 2.1.3 Skip-lot Sampling 18 2.1.4 Chain Sampling 21 2.1.5 MIL-STD-105 23 2.2 Destructive Sampling 27 2.3 High Quality Sampling 28 2.4 Empirical Bayesian Methods 30 111 2.5 Acceptance Sampling vs. Quality Monitoring 33 2.4 Summary 34 m. COST OF QUALITY 36 3.1 Economically-Based Cost Model 36 3.2 Simulation Cost Function 39 3.3 Binary search to determine type of defective 41 3.4 Summary 43 IV. DEVELOPMENT OF A DESTRUCTIVE SAMPLING METHOD FOR HIGH QUALITY PRODUCTION PROCESSES 44 4.1 The Gamma Function and Gamma Distribution 44 4.2 The Poisson Distribution 45 4.3 Empirical Bayesian Analysis for the Poisson Distribution 46 4.4 Destructive Sampling Method designed for High Quality Production Processes 48 4.5 Illustration of the DSM-HQ 52 4.6 Out-of-control rules 55 4.6 Summary 55 V. SIMULATION DESIGN 57 5.1 Stage 1 ofSimulation: Defining DSM-HQ 57 5.1.1 Prior Distribution Parameters 58 5.1.2 Cost Function Parameters 61 5.1.3 Paths to minimum sampling 63 IV 5.1.4 Defective Patterns 66 5.1.5 Measures to be Calculated 68 5.1.6 Summary of Stage 1 Simulation 69 5.2 Stage 2 ofSimulation: Simulation design for the comparison of sampling techniques 69 5.2.1 Sigma Levels 70 5.2.2 Defective Patterns 70 5.2.3 Cost Function Parameters 71 5.2.4 Specification based on results of Stage 1 71 5.2.5 Single Sampling 71 5.2.6 Double Sampling 73 5.2.7 Multiple Sampling 74 5.2.8 Skip-lot Sampling 76 5.2.9 Chain Sampling 77 5.2.10 MIL-STD-105E 78 5.2.11 Measures to be Calculated and Compared 80 5.2.12 Summary 81 VI. SIMULATION RESULTS: ANALYSIS AND COMPARISON OF SAMPLING METHODS 94 6.1 Stage 1 of the Simulation results: Defining DSM-HQ 94 6.1.1 Prior Distribution Parameters 95 6.1.2 Long-term sampling values 104 6.1.3 Paths to minimum sampling 108 6.1.4 LowerSigma-LevelResults Ill 6.1.5 Sampling rates with an out-of-control element added 113 6.2 Stage 2 of the Simulation results: comparison of sampling methods 116 6.2.1 Random event comparison of sampling methods 116 6.2.2 Random event with out-of-control comparison of sampling methods 6.3 Summary 124 VIL CONTRIBUTIONS AND SUMMARY 126 7.1 Contributions, Limitations and Future Research 126 7.2 Summary 128 REFERENCES APPENDIX A. 120 131 SEQUENTIAL ANALYSIS UNDER HIGH QUALITY CONDITIONS 134 B. COMPARISON OF METHODS AT A 5-SIGMA - RE 144 C. COMPARISON OF METHODS AT A 6-SIGMA - RE 153 D. COMPARISON OF METHODS AT A 5-SIGMA - REOC 162 E. COMPARISON OF METHODS AT A 6-SIGMA - REOC 172 F. A S AS PROGRAM FOR THE FINfE-TUNING OF DSMHQ, STAGE 1 182 A SAS PROGRAM FOR COMPARISON OF DSM-HQ WITH EXISTING METHODS, STAGE 2 193 G. VI ABSTRACT In manufacturing and assembly, the sampling of units produced is important since in many situations not all of the units can be tested. Destructive sampling, which commonly occurs in the assembly and manufacturing industry, is a form of sampling where all units produced cannot be tested since the parts are destructively tested. In this situation, sampling techniques are used to determine if an entire lot should be accepted or rejected based on the sampling results. The traditional sampling techniques include single or classical sampling, double sampling, multiple sampling, skip-lot sampling, chain sampling and MIL-STD-105E. However, in the modem era of high quality production, traditional sampling techniques require a high number of imits tested in order to guarantee a high level of quality resulting in very high sampling costs. Therefore, to keep costs down, manufacturers and assemblers have used these techniques with lower sampling numbers in order to monitor quality. A goal of this research is to develop a sophisticated technique that monitors quality and outperforms the existing techniques in situations where quality is high and tests are destructive. The proposed technique. Destructive Sampling Method for High Quality production processes (DSM-HQ), is based on a cost function, which balances the costs of sampling versus the costs of finding a defect on the field. DSM-HQ assumes to have a Poisson process defect pattern and uses an Empirical Bayesian analysis to allow the researcher to include prior knowledge. vn The research simulation and results are separated in two stages. Stage 1 fine tunes DSM-HQ and examines its properties, while Stage 2 compares DSM-HQ to the traditional methods. The simulation resultsfi-omStage 2 show that DSM-HQ is superior to the traditional methods in most cases at the 5-sigma level. As the quality increases to 6-sigma, DSM-HQ proves to be significantly superior to all traditional methods in every cost case considered and in both random events combined with out-of-control events case and the random-event-only case. Although DSM-HQ sampling method has some limitations, which will be explored in fiiture research, and the case examined here is limited in scope, which will be expanded in future research, the results and comparisons to traditional methods are very encouraging. vui LIST OF TABLES 2.1 Values for AOQL,/ /, under Procedures Al 21 2.2 MIL-STD-105E Sample Size Code Letters 26 5.1 Sigma Levels of Quality in terms of percent defective 59 5.2 Different prior starting points for the evaluation of the true sigma quality level 60 5.3 Different a and y9 values for each of the prior >! starting points 61 5.4 Values considered for each of the cost parameters 62 5.5 MIL-STD-105E Sampling Plans for Sample Size Code Letter C 83 5.6 MIL-STD-105E Sampling Plans for Sample Size Code Letter D 84 5.7 MIL-STD-105E Sampling Plans for Sample Size Code Letter F 85 5.8 MIL-STD-105E Sampling Plans for Sample Size Code Letter G 86 5.9 MIL-STD-105E Sampling Plans for Sample Size Code Letter J 87 5.10 MIL-STD-105E Sampling Plans for Sample Size Code Letter K 88 5.11 MIL-STD-105E Sampling Plans for Sample Size Code Letter N 89 5.12 MIL-STD-105E Sampling Plans for Sample Size Code Letter R 90 5.13 Average Outgoing Quality Limit Factors (Single Sampling)* 91 5.14a Sampling Plansfi-om3 to 6 sigma level of production quality 92 5.14b Sampling Plansfi"om3 to 6 sigma level of production quality 93 6.1 Effects of changing each of the cost variables 103 6.2 Long-term sampling values for a 3-sigma process 105 6.3 Long-term sampling values for a 4-sigma process 106 ix 6.4 Long-term sampling values for a 5-sigma process 107 6.5 Long-term sampling values for a 6-sigma process 108 6.6 Sampling rates for a 5-sigma process with OC and RE combined 114 6.7 Sampling rates for a 6-sigma process with OC and RE combined 115 6.8 Methods and Sampling plans used for comparison 117 LIST OF FIGURES 2.1 Block diagram of basic sequential sampling process 17 2.2 Switching rules for the MIL-STD-105D (AdaptedfiromGrant and Leavenworth, Statistical Quality Control, Sixth Edition pp 462, 1988) 25 4.1 Positive higher jc intercept 51 4.2 Negative higher JC intercept 51 4.3 DSM-HQ reduces sampling ratefi-om7 to 1 per hour (0 defectives are found) 53 4.4 DSM-HQ adjusts for a defective at day 38 54 5.1 Reduction of sample size and sampling interval for path type 1 64 5.2 Reduction of sample size and sampling interval for path type 2 65 5.3 Conceptual view of reduction of sample size and sampling interval for path type 3 Actual view of reduction of sample size and sampling interval for path type 3 65 66 5.5 Random defective 67 5.6 Process going out-of-control 67 6.1 Intercepts from a 4-sigma process with same X but different a/p combinations 96 5.4 6.2 Two 3-sigma prior values adjusting at different rates to a 5-sigma process 97 6.3 Six-sigma process with 3 and 6-sigma priors 98 6.4 Effects of underestimating and overestimating prior values XI 100 6.5 Effects ofreducingc/'whilekeepingyc and vc constant 101 6.6 Effects of increasing vc while keeping^c and (/constant 102 6.7 Effects of increasing/c while keeping vc and c/constant 103 6.8 Path type 1 to minimxmi sampling using for/c = 5, vc = 2 and cf= 10,000 109 Path type 2 to minimimi sampling using for^ = 5, vc = 2 and cf= 10,000 110 Path type 3 to minimum sampling using for^c = 5, vc = 2 and cf= 10,000 Ill Percent defectives detected by method at the 6-sigma level of quality 119 6.12 Percent of Total Cost divided into TCS and TCF at the 6-sigma RE 120 6.13 Total Weekly Cost where DSM-HQ suggests a sampling rate of 5 per lot 121 Total Weekly Cost where DSM-HQ suggests a sampling rate of 1 per week 122 6.15 Overall Total Weekly Cost at 6-sigma REOC 123 6.16 Out-of-control time for each method in hours at the 6-sigma level 124 A.l Sequential Sampling for a Very High Quality Process 143 6.9 6.10 6.11 6.14 xu CHAPTER I INTRODUCTION 1,1 Overview In contemporary production management, techniques for sampling in quahty control are important because sometimes not all of the products can be tested, especially if the test actually destroys the product (such as in a stress test). Techniques have been developed in order to determine if a production or shipment lot of a particular product should be accepted or rejected (not accepted). In some of these situations, a sampling plan is also put in effect. The most basic technique used for sampling is known as "classical sampling" or "single sampling." This technique calls for the decision on accepting or rejecting a lot on the basis of the evidence of one sample from the lot (Grant & Leavenworth, 1988). "Double sampling," involves taking a first sample and making a decision based on the evidence of the first sample. However, if the sample is neither good nor bad enough, then a second sample is taken and is combined with the first and the decision on whether to accept or reject is based on the information of these two samples (Grant & Leavenworth, 1988). An extension of double sampling is "multiple sampling, which involves taking a first sample and making a decision based on the evidence of the first sample. If the sample is neither good nor bad enough, then just like in double sampling, a second sample is taken and combined with the first sample in order to attempt to make a decision. However, if a decision cannot be made after the second sample, then a third sample is taken. The process is repeated until a decision to accept or 1 reject is made or until the process reaches a pre-determined number of samples. Another sampling method, which yields very similar results to multiple sampling (Grant & Leavenworth, 1988), is called "sequential sampling" where the samples of size one and are evaluated sequentially. Using sequential sampling one of three decisions is made at any stage of the sampling problem: (1) accept the lot, (2) reject the lot, or (3) continue the experiment by taking an additional sample (Wald, 1973). Based on the first sample, a decision is made and the process is carried out sequentially until either the first or second decision is made. It has been shown that double sampling offers statistical advantages over single sampling, and sequential sampling is statistically superior to double sampling (Grant & Leavenworth, 1988; Wald, 1973). Two other methods used in destructive testing are chain sampling and skip-lot sampling. The skip-lot sampling plan used for destructive sampling is referred to as SkSp-2 (Grant & Leavenworth, 1988) where a reference sampling plan is used initially, and once the lots have been found free of defective items, the inspection shifts to a proportion/of the lots received (Perry, 1973a; Dodge, 1955a). Chain sampling is used in cases of continuing production of lots where very small sample sizes are selected for each lot because tests are destructive or costly (Grant & Leavenworth, 1988). The original plans came from Dodge (1955b) and are referred to as ChSP-1 (Soundararajan, 1978). It utilizes a single sampling technique and it is based on n being small, and c (acceptance number per lot) being 0. Under the ChSP-1 plan, a sample size of w units for each lot is selected and tested. Then, the acceptance number of defectives per lot is set to zero, or one in the case where the preceding i lots had zero defectives. Finally, one other type of technique used in this type of situation is the one developed by the United States military, which is known as the Military Standard 105 or MIL-STD-105 (Department of Defense [DOD], 1989). Under this technique, certain conditions will allow the sampling to be relaxed from "Normal" to a "Reduced" inspection, while other conditions will force the sampling to go to a "Tightened" inspection (Pabst, 1963). Under tightened inspection, if the process continues to falter, then the whole production process is stopped. On the other hand, if it improves, it can retum to "Normal" inspection. Industry continues to use these types of statistical sampling techniques even in the modem era of high quality production. Because productivity, cost of quality and cost of scrap are now major concerns in manufacturing and assembly, some of these techniques need to be re-evaluated and compared against new sampling techniques, which include cost of inspection and material as part of its focus. Taking into consideration that in today's manufacturing and assembly processes the customer expects "zero-defects," it should be noted that these sampling techniques are not designed to handle such demands without having an exorbitant cost of quality. A goal of this research is to develop a more sophisticated acceptance technique in situations where quality is high and tests are destructive. This technique will include a cost function to take into consideration costs of sampling as well as costs of finding a defective unit outside the supplier facihties. In addition, it will take advantage of the power of Empirical Bayes analysis, which takes into consideration prior information of the production process. Current sampling techniques fail to do this. Furthermore, the defect pattern is assumed to follow a Poisson process. Finally, it will look at advantages and disadvantages of the developed model versus the existing models imder different quality conditions considering random occurrences and out-of-control events. The cost function will balance the cost of sampling and finding a defective unit within the manufacturing or assembly plant versus the cost of finding the defective unit at the client's site. In addition, the cost fimction will take into account the fact that the test is destructive. The Empirical Bayes analysis allows for the researcher to establish the parameters for the prior distribution. In other words, the Bayesian approach allows the researcher to establish a prior opinion of a parameter in a prior distribution, which means that the researcher does not have to start from the beginning each time (Iversen, 1984). In some cases researchers apply Bayesian analysis repeatedly, each time taking the posterior of the last step as a prior for the next one (Becker & Camarinopolous, 1990). Because a major concern is to reduce the number of items in a sample while at the same time maintain the level of quality, Bayesian estimation models enable the researcher to make better decisions when the data are scarce and incomplete (Chulani et al., 1999). The Poisson distribution makes sense in this situation, since it can be assxmied that the defectives arrive randomly during a fixed time period and the Poisson distribution is usefial in approximating binomial distributions with very small success probabilities (DeGroot & Schervish, 2002). The research will focus on developing and fine-tuning DSM-HQ and then on comparing and contrasting it to the traditional single and double acceptance sampling, the multiple acceptance sampling, skip-lot sampling, chain sampling, and finally MIL-STD105. The research will determine the effects of a sampling technique developed for destructive sampling (which commonly occurs in the assembly and manufacturing industry) compared to the traditional techniques used today. The effects will be in the form economic advantages and different rates of defective units per lot (from several defectives per lot to today's high quality processes), which will include not only random defectives but also out-of-control events. In addition, the research will determine the effects of different prior information. 1.2 Problem Statement When an item is tested during a manufacturing or assembly process, sometimes the test can be accomplished without harming the product. However, there are times where the test required will destroy the product. Currently, the quantity of sampling inspection is determined by the established techniques of single sampling, double sampling, multiple sampling, sequential sampling, skip-lot sampling, chain sampling, or MIL-STD-105. Because the chent expects zero defects, the amount of sampling inspection is usually extremely high with these techniques. This causes a high cost of quality, reduces productivity, and increases scrap. In some of the high quality assembly plants, it is estimated that over 50% of its scrap is a result of destructive testing of perfectly good items (E. Castillo and S. Alvarez, personal communication, October 6, 2003). Therefore, there is a need to determine if a sampling technique, which uses prior information and is modeled after the production and defective process, can produce better results than the current techniques in today's high quality processes. 1.3 Objective of the Research In cases where destructive tests are necessary and where clients demand "zero defects," the only way currently to get a high enough confidence that extremely high quality is being delivered to the client is to destroy a very high percentage of the products. The question is whether an accurate model of production and defective processes will yield a better acceptance sampling technique. Some of the research questions are as follows: 1.) What are the characteristics of DSM-HQ? Does it work better for high or low cost products? Is it better suited for products that have a large or small cost penalty for finding a defective on the field? What are the cost-efficient prior parameter values under different quality sigma levels and different cost considerations? Does it require a large amount of prior information or can small values for the prior information produce adequate results? How does it respond to defective items under different quality performance levels and different prior information? 2.) Will an accurate model of the defective process produce a better or worse sampling technique in the case of destructive testing? Under what conditions is DSM-HQ superior to traditional sampling techniques and under what conditions is it inferior? 3.) What are the sampling advantages and disadvantages of DSM-HQ under different types of sigma quality levels? 4.) What are the economic advantages and disadvantages of DSM-HQ versus the traditional sampling techniques when sampling is destructive, under different sigma quality levels and under different cost considerations? 5.) How does each of these techniques handle out-of-control events and how do they handle random defectives under different quality sigma levels? The purpose of this dissertation is to utilize the empirical Bayes analysis to establish an empirical Bayes model using a Poisson process as well as determine the economic effects compared to the sampling techniques of today in the case of destructive testing. The Poisson process is used to model defective items. In addition, the empirical Bayes analysis allows for the prior distribution of the defective items, which provides the researcher with additional information on what to expect (Iversen, 1984). Although the classical and multiple techniques might end up producing better results in certain cases, this research is determined to find what kind of advantages are offered by modeling the defective process and in what type of situations is this process most desirable. In addition, the research is also aimed to find the economic effects that a modeled technique might bring to a very high quality level production process. Finally, the research will also compare modeling the process at different levels of production quality. Although it will concentrate at very high levels of quality, it will look at advantages and disadvantages in other types of situations where the quality of production is not as high. 1.4 Document Organization This dissertation shall be organized as follows. Chapter I introduces the investigation. Chapter II contains the literature review on Single Sampling, Double Sampling, Sequential Sampling and Muhiple Sampling, Skip-Lot Sampling, Chain Sampling and MIL-STD-105. Chapter III discusses the Cost of Quality in destructive sampling. Chapter IV discusses the Theoretical development of the empirical Bayes analysis for the Poisson random variables within the destructive testing framework. The destructive sampling method designed for high quality production processes (DSM-HQ) is proposed. Chapter V contains the simulation design, where all methods are compared at different levels of quality and within different situations. Chapter VI presents the simulation results including extensive numerical tests and analysis of the proposed DSMHQ. Chapter VII summarizes the proposed study, lists the expected contributions and limitations of the research, and gives recommendations for fiiture research. CHAPTER n LITERATURE REVIEW This chapter reviews the major sampling processes, which will be used to compare against the proposed method, DSM-HQ. Additional topics presented in this chapter include destructive sampling, which discusses the properties of these techniques in a destructive testing environment, and high quality sampling where fiirther tendencies of the sampling processes under high quality are reviewed. Other topics included in this chapter are properties of Bayesian methods, and a comparison of quality monitoring with acceptance sampling. 2.1 Acceptance Sampling Inspection of a production process can be done at 100% or by sampling. In the case when testing is destructive, inspection must be done by sampling. Two widely used systems in acceptance sampling can be classified as acceptance sampling by variables and acceptance sampling by attributes. Acceptance sampling by variables is done when the decision depends on criteria of the frequency distribution of the submitted product (Grant & Leavenworth, 1988). Acceptance sampling by attributes is a "go-not go" decision where the product is deemed either good or bad. This research focuses on acceptance sampling by attributes at high quality levels, and this section discusses the major sampling processes for acceptance sampling by attributes, including single sampling, double sampling, multiple sampling, sequential sampling. Skip-lot Sampling (SkSP-2), chain sampling (ChSP-1) and MIL-STD-105. 2.1.1 Single and Double Sampling The classical acceptance sampling, also known as single sampling, is a procedure still taught in oiu- universities today. This procedure is based on a decision on acceptance or rejection of a lot on the basis of evidence provided by one sample (Grant & Leavenworth, 1988). Double sampling, which was first introduced by Dodge and Romig (1929), involves performing the single sampling technique and then making a decision on whether to accept, reject or take another sample. The lot may be accepted if the sample is good enough or rejected if the sample is bad enough. Double sampling allows for a gray area where the researcher might not be completely sure if the sample is good enough or bad enough and therefore a second sample is taken. If a second sample needs to be taken, the decision is then based on the two samples combined (Govindarajulu, 1981). The first sample in double sampling for a pre-determined lot size and a predetermined acceptable quality level will always be smaller than the one sample in single sampling (Grant & Leavenworth, 1988). However, the combined sample of double sampling (under the same lot size and same acceptable quality level) will always be larger. The relative number of items inspected depends on the quality of the items. If the first sample is generally good enough so that the lot is accepted or bad enough so that the lot is rejected, then very few second samples need to be taken and the number of inspection samples will generally be smaller under double sampling. The more often the 10 results fall in the middle area the number of inspection samples will be larger under double sampling. Dodge and Romig (1944), who prepared the now famous DodgeRomig Sampling Inspection Tables at Bell Telephone System, claim that the savings in inspection due to double sampling is usually over 10% and may be as much as 50%. One characteristic of all double sampling plans is that the acceptance number for the two samples combined (the maximum nxunber of defectives that will permit the acceptance of the lot on the basis of the two samples) is always greater than or equal to one (Grant & Leavenworth, 1988). This is important since double sampling does not allow a lot to ever be rejected as a result of having only one defective. This makes sense since the second sample would not be necessary if no defectives were found and would only be necessary if one or more defectives were found in the first sample. If one defective is found in the first sample, a second sample is taken and if no defectives are«found in this second sample, then the total of both samples combined would be one defective and the lot would not be rejected. 2.1.2 Sequential and Multiple Sampling Whereas double sampling allows for only two samples, Walter Bartky (1943) improvised the method to test the mean of multiple samples of a binomial distribution. This was developed in order to require, on average, a smaller number of observations than single or double sampling and was the forerunner of sequential analysis (Wald, 1973). Friedman and Wallis worked on some modifications of sequential test procedures, which encouraged Wald to develop the sequential probability ratio test. This test was so 11 usefiil in the development of military and naval equipment that the United States government classified it as restricted information under the Espionage Act (Wald, 1973). Sequential sampling was created in order to improve on the number of observations needed over single sampling. This technique is different from single sampling (where the number of observations is predetermined) in that the number of observations required by the sequential test depends on the outcome of the observations, and is therefore a random variable (Wald, 1973). Using the sequential method a rule is given and one of three decisions is made at any point during testing: the lot is accepted, the lot is rejected, or another observation taken. If the first or second decision is made, then the experiment is over and the process is terminated. If the third decision is made a second observation is collected. Based on the first two observations a decision is made to accept the lot, reject the lot or take another observation. If one of the first two decisions is made, then the process is terminated. Otherwise, a third observation is taken. The process is repeated sequentially until a decision is made. It was demonstrated by Govindarajulu (1981) that the probability of sampling forever without reaching a terminal decision under the sequential probability ration test is zero. Although sequential sampling can be used in a variety of problems, the emphasis will be placed on acceptance sampling inspection of a lot where each unit is classified into one of two categories: defective or not defective. The random variable X will take a value 1 if the item is observed to be defective and the value of X will be 0 if a nondefective item is observed. The unknown probability,/?, is the probability the random variable X is 1 and the probability that the random variable X is 0, is (l-p). 12 The researcher selects a value/?' such that he/she would like to accept a lot whenever/? </?' and would reject the lot whenever/? > /?'. The hypotheses consist of accepting or rejecting these two decisions based on a random sample (Wald, 1973). The quality of the lot is on the margin whenever/? =/?'. In this situation the researcher is indifferent on which decision it is made (accept or reject). If/? >/?' there is a preference to reject and the preference increases as the value of/? increases. If/? < /?' there is a preference to accept and the preference increases as the value of/? decreases. As/? gets closer to/?'the preference of rejection or acceptance is only slight and the error is not of practical consequence. Wald (1973) suggests that two possible values/?o and pj should be specified by the researcher, where/?o is below/?' and/?/ is above/?'. In this situation, if /? lies between/?o and/?y, there is no great consequence if there is an acceptance or rejection error. However if/? >/?/ and the lot is accepted the error is regarded as an error of practical consequence. Similarly, if/? <po and the lot is rejected the error again is regarded as an error of practical consequence. The probability of rejecting the lot whenever/? <po should not be greater than a stated value of or, and the probability of accepting the lot whenever/? >/?/ should not be greater than a stated value of/?. The researcher has to determine the values of the four constants a, fi, po, and /?;, which determines the tolerated risks of making the wrong decision. The researcher then tests the null hypothesis/? =/?o, against the alternative hypothesis,/? =/?/. Let^- be the indicator random variable of whether the i'^ unit is defective. Let dm be the number of defectives in the first m units inspected and/? denote 13 the proportion of defectives in the lot. The probability of obtaining a sample equal to the observed probability is Under the null hypothesis,/? =/?o, the probability becomes Po„ = Po'" i^-Por"" • Under the altemative hypothesis,/? =/?;, the probability becomes The following steps are defined by Wald (1973) as the sequential probability ratio test for testing the null vs. the altemative hypothesis. Two positive constants B and A (B < A) are determined. At the m^^ trial, the probability ratio —1^ is computed. The Pom (w+l)* trial is collected if B<^^<A. Pom The process is terminated and the null hypothesis is rejected if P\m Qm The process is terminated and the null hypothesis is accepted if Pom Define a sample of type 0 to be one such that 5<^^^^=^<^,/ = l,(/w-l) and Poim-i) ^^<B. Pom 14 Let a sample of type 1 be such that B<£l--<A,i = l(m-l) and Pom ^^>A. Pom A sample of type 0 resuhs in the acceptance of Ho and a sample of type 1 results in the rejection of Ho (acceptance ofH\), To obtain a lower limit for B the following criterion is used. For any given sample (jcy,,.. x^) of type 0, the probability of getting this sample under H\ is at most B times as large as the probability of obtaining such a sample when HQ is true. Therefore, the probability of accepting Ho is at most B times as large when H\ is true. The probability of accepting HQ is I - a when HQ is true and ^ when H\ is true, thus the inequality >0 < (1 - a)B holds, which can be written as 5 > . v . Therefore, the upper limit for Bis To obtain an upper limit for A the following criterion is used. For any given sample (xy,... x„) of type 1 the probability of getting this sample under HQ is at most A times as large as the probability of obtaining such a sample when H\ is true. Therefore, the probability of accepting H\ is at most A times as large when HQ is true. The probability of accepting H\ (rejection ofHo) is a when HQ is tme and 1 - /? when H\ is true, thus the inequality I- /3> Aa holds, which can be written as A< a Therefore, the upper limit for A is a 15 Using Wald*s (1973) sequential probability ratio test at each stage of inspection it follows that if B<^^<A Pom inspection of the m'^ unit for each positive integer value of/w, the inequality >g ^Plm , 1 - y g (l-«) ^ Pom holds. More specifically j3 ^p^-n-p.r"(l-a) /;/" (1-/70)"-''" j-fi a Taking the log of the ratio results in Pan, 1-Po Po Therefore, the lot should be rejected and inspection terminated if iog^>iogi::^ Pon. a and the lot should be accepted and inspection terminated if log^^<log ^ Pom l - a Inspection should continue when log-^<log^^<log^^ l-a Pom a Figure 2.1 describes the basic process. 16 Start Observe 1 Unit Po^ii-PoT'^ a-«) Accept (1-^) Po^iX-Po) ^ Take Additional Observation Poiy-Po) Reject Figure 2.1: Block diagram of basic sequential sampling process Wald (1973) claimed that the sequential probability ratio test is exactly an optimum test, but he never succeeded in proving this result. He did demonstrate that sequential testing resulted in a savings of about 50% over the single-sampling procedure (which was the most powerfiil test at that time). Multiple sampling also improvises on the double sampling method. Whereas double sampHng only allows for two samples, multiple sampling allows for more samples to be taken. In this method, a first sample is taken and a decision based on the evidence of the first sample. If the sample is neither good nor bad enough, then just like in double sampling, a second sample is taken and combined with the first sample in order to attempt to make a decision. However, if a decision cannot be made after the second sample, then a third sample is taken and all three samples are combined in order to make a decision. The process is repeated imtil a decision to accept or reject is made or until 17 the process reaches a pre-determined number of samples. Under multiple sampling, this pre-determined maximum number of samples recommended is usually seven (DOD, 1989). Multiple sampling is used when a decision is possible after each sample taken. This means that the decision is not made item-by-item like sequential sampling, unless the sample size is equal to one. Furthermore, sequential sampling does not have a predetermined number of total units to be inspected (Wald, 1973). On the other hand, multiple sampling has a limit to the amount of total units that can be inspected since it has a pre-determined maximum number of samples allowed. Sequential sampling and multiple sampling do share some similarities. Both are extensions of the double sampling technique and both methods allow for the possibility to sample more items until a definite decision is made. Also, the results yielded by both methods so similar that many writers refer to the two methods interchangeably (Grant & Leavenworth, 1988). 2.1.3 Skip-lot Sampling Skip-lot sampling (SkSP-1) was developed by Harold Dodge (1955a) for sampling chemical and physical processes in order to bring about substantial savings on inspection of products, which normally conform to specification. This particular sampling plan is usefiil when the lots are small or where inspection is slow and costly (Grant & Leavenworth, 1988). SkSP-1 depends heavily upon the assumption of homogeneity among lots and a good quaUty history. Under this plan, there is 100% 18 inspection of the first / consecutive lots. If no defectives are found in any of the first i lots the inspection is performed on a proportion,/, of lots received. In Skip-lot Sampling a provision is made for skipping inspection on afi-actionof the submitted lots if the submitted product is deemed to be high quality as demonstrated by the history of the submitted product (Perry, 1973b). The values oft and/are selected fi-om some sampling system such as CSP-1 (Dodge, 1943), CSP-2 or CSP-m (Dodge & Torrey, 1951) and are called skipping parameters (Perry, 1973b). Because SkSP-1 rehes on 100% inspection at the outset, it is not practical for destructive testing. Using SkSP-2, developed by Dodge, a reference-sampling plan is used and each lot is inspected using the sampling plan (Grant & Leavenworth, 1988). The rest of the procedure is the same under either SkSP-1 or SkSP-2. The referencesampling plan can be single sampling, double sampHng, or some other sampling plan (Grant & Leavenworth, 1988). Once the reference sampling plan is chosen, some simple rules for switching between what is known as "normal inspection" and "skipping inspection*' are followed. The rules for switching between these two types of inspection are (Perry, 1973b): 1. Select a reference-sampling plan. 2. Start with normal inspection. 3. When i consecutive lots are accepted on normal inspection, switch to skipping inspection where/(a fraction of the lots) are inspected. 4. While the procedure is in skipping inspection, switch to normal inspection only after a lot is rejected. 19 5. Correct or replace all defective units found after screening each rejected lot. The reference plan and the skipping parameters (/'and i) are needed to completely specify the SkSP-2 inspection technique. The fraction of the lot inspected,/ has to be between 0 and 1 (0 < / < 1). If/is equal to one, then SkSP-2 becomes identical to the reference-sampling plan. The number of consecutive lots accepted until the inspection is switched from normal to skipping, z, must be a positive integer. With a Skip-Lot Sampling Plan, an average outgoing quality limit (AOQL) can be set. This means that there is confidence that in the long run, no more than the AOQL percentage of the accepted lots will be nonconforming (Dodge, 1955a). For example, if 2% AOQL is chosen then in the long run no more than two percent of the accepted lots will be nonconforming. This is what Dodge calls a corrective plan, which provides an upper bound to the average percentage of accepted lots that will not conform. Dodge discusses two procedures for Skip-Lot Sampling. Procedure Al is used when each nonconforming lot is corrected or replaced by a conforming one. The other procedure is A2 where each nonconforming lot is rejected and not replaced by a conforming lot. Table 2.1, derived from the Continuous Sampling Plan (CSP-1) graph provided by Dodge (1955a), assists the user in determining values for each of the key variables. For other choices of/and z. Figure 1 of Dodge (1955a) gives curves for Skip-Lot Sampling corresponding to procedure AL Perry (1973a) concludes that Skip-Lot Sampling is a good and usefiil acceptance sampling procedure, which might qualify as a 20 standard system of reduced inspection. Furthermore, when the quality of the product is good, the SkSP-2 technique has a desirable property of reducing the amount of inspection (Perry, 1973a). Table 2.1: Values for AOQL,/ /, under Procedures Al AOQL 1 for Procedure Al / 1% 1/2 27 1% 1/4 60 1.5% 1/2 20 1.8% 1/2 15 2% 1/2 14 3% 1/2 9 4% 1/4 14 5% 1/2 5 7.5% 1/5 9 8% 1/2 3 10% 1/4 5 11% 1/2 2 2.1.4 Chain Sampling Dodge (1955b) developed the Chain Sampling Plan or ChSP-1 for the specific instances when destructive or costly tests are performed. In these situations the sample sizes need to be kept at a minimum for each lot. Dodge proposes that ChSP-1 is a 21 desirable plan if the lots have essentially the same quality level, the lot is one of a series in a continuing supply, and there is no reason to believe that one particular lot is poorer than the immediately preceding ones. If all these conditions hold the samples for each lot are used cumulatively for acceptance purposes. Dodge (1955b) refers to this as "links in a chain" since the individual lots and individual samples are associated. The procedure for this sampling plan indicates that for each lot, a sample of n units is selected and tested. The acceptance number of defectives is zero (c = 0) except if no defectives are found in the preceding i units. Therefore, this technique makes use of cumulative results for several samples (Dodge, 1955b). This technique allows for a small percentage of imperfections in the production process. Dodge (1955b) mentions that for this technique a small percent defective is reasonable. Consider the sampling plan for « = 5, c = 0. If all tests pass, then the lot is accepted. However, if one of the tests fails, then the decision of accept or reject the lot depends on the history of the preceding lots. If any of the immediately preceding lots were found to be nonconforming, then the technique would suggest that the current lot is also nonconforming. However, if all of the immediately preceding lots have been found to be conforming, then the failure of the sampling test on that particular unit is deemed to be a small marginal failure, which is considered reasonable xmder most production processes. According to Dodge (1955b) this technique has the following requirements in order to allow for the occasional marginal failure: 1. The lot should be one of a series of lots in a production process. 22 2. Every lot in the production process should have essentially the same quality. 3. The consumer has no reason to believe that the lot sampled is of poorer quality than the immediately preceding lots. 4. The consumer must feel confident that the supplier will not take advantage of a good record and deliver a bad quality lot on purpose. The procedure for ChSP-1 starts by determining number of units, n, to be sampled and test each unit for conformance. The acceptance number is set to c = 0, and c = 1 in the case where no defectives were found in the immediate / preceding samples ofn. For a particular lot, the decision to accept comes if no defectives are found (c = 0) and the decision to reject comes if two or more defectives are found (c > 2). However, if only one defective is found, then the lot is accepted or rejected based on the history of the previous lot, and the choice of z (the number of preceding samples or links in the chain). For example, if / = 5, and the lot has one defective, then if there were no defectives found in the immediately preceding 5 lots, the lot is accepted. However, if there were one or more defective items in the immediately preceding 5 lots, then the lot is rejected. This implies that the lot can still be accepted with one defective if the last defective was found far enough back in history. Dodge (1955b) concluded that values of / = 3 to 5, were found to be most desirable in practical applications. 2.1.5 MIL-STD-105 MIL-STD-105D Sampling was first issued in 1950 as MIL-STD-105 A. MILSTD-105D (Pabst, 1963) discusses single and double sampling. The latest, MIL-STD- 23 105E was refined in 1989 (DOD, 1989). This technique uses tables as well as normal, tightened and reduced inspection, depending on the history of the previous lots. Under the Military Standard 105 also known as the American-British-Canadian Standard 105 in the international community, the procedure is as follows. The sampling starts at "Normal" inspection. When sampling is under normal inspection, four conditions need to be met in order for the inspection to be relaxed to "Reduced" inspection. 1. 10 consecutive lots (/ =10) must be accepted and pass the normal inspection process. 2. The number of defectives found in those 10 lots must be less than the number specified by Table VIII in the MIL-STD-105E report. 3. Production must be at a steady rate. 4. A responsible authority must approve the "Reduced" inspection rate. For the sampling process to go from reduced inspection to normal inspection, the following four conditions need to occur 1. While the sampling process is in reduced inspection, a lot is rejected 2. A lot is accepted but the number of defectives falls in between the acceptance number and the rejection number. 3. Production is irregular or some other conditions warrant more detailed sampling For the sampling process to go from normal inspection to tightened inspection two out of five consecutive lots are rejected on original inspection. Normal inspection can again be resumed when five consecutive lots are accepted while under tightened 24 inspection. Finally, inspection should be discontinued if 10 consecutive lots remain under tightened sampling and the production process should be reviewed. Figure 2.2 illustrates the MIL-STD-105 technique. / = 10 Defects < Limit (Table W) Steady Production Approved by authority 2 out of 5 consecutive lots rejected Lot rejected Ac<defects<Rc Irregular Production Conditions warrant 5 consecutive lots accepted 10 consecutive lots remain tightened f Stop j Figure 2.2: Switching rules for the MIL-STD-105D (Adapted from Grant and Leavenworth, Statistical Quahty Control, Sixth Edition pp 462, 1988). MIL-STD-105 has several inspection levels. Inspection levels determine the relationship between the sample size and the lot size. The client usually sets the requirement for the level of inspection. In the MIL-STD-105E tables, there are three levels: • Level I - Reduced Inspection • Level II - Normal Inspection 25 • Level III - Tightened Inspection Although Inspection Level II is normally used. Levels I or III can be used in special cases where less or more discrimination is needed. These three levels are coded by letters, which guide the user through the MIL-STD-105E tables. Depending on the lot size and on the level of inspection a code letter is assigned for a particular acceptancesampling plan. Furthermore, MIL-STD-105E includes additional special levels S-1, S-2, S-3 and S-4 for the cases where small sample sizes are necessary (although large sampling risks must be tolerated). Table 2.2 illustrates a portion of the sample size code letters for different lot sizes at Levels of inspection S-1,1, II and III. Table 2.2: MIL-STD-105E Sample Size Code Letters Lot Size Level S-1 Level I Level II Level III 91 to 150 B D F G 151 to 280 B E G H 281 to 500 B F H J 501 to 1200 C G J K 1201 to 3200 C H K L 3201 to 10000 C J L M 10001 to 35000 C K M N 35001 to 150000 D L N P 26 The sample size code letter tables used for the sampling plans in this research are found at the end of Chapter V. These tables are utilized to arrive at sampling plans for single, double, multiple and MIL-STD-105E sampling. 2.2 Destructive Sampling There are instances where 100% inspection is not reasonable and where small sample sizes in the sampling process is necessary such as in the case of destructive sampling. In these situations larger sampling risks must be tolerated. As the sample sizes get smaller, the sampling risks get larger. MIL-STD-105E offers four additional levels of inspection given for situations where small sample sizes must be used. Table 2.2 gives the sample size code letters for special level S-1, which is the sampling plan with the smallest sample sizes offered by MIL-STD-105E. Although the sampling risks in S-1 are larger than in Level II (normal inspection), the sample size under this special plan is more practical and reasonable in the cases where small sample sizes are a must. For example, if there is a lot of 800 units and the Acceptable-Quality Level (AQL) under single sampling is 2.5 percent, then under Normal Inspection (Level II), the minimum sample size required would be 80 units (Tables 2.2 and 5.9). This imphes that the number of sampled units would be 10% of the total lot. To destroy this number of units would be unreasonable. Under Special Level I (S-1) only 5 units would be sampled (Tables 2.2 and 5.5). Skip-lot Sampling (SkSP-2) is another tool used to reduce the number of samples taken since a fraction of the submitted lots is sampled once the conditions of high quality 27 have been met. Because SkSP-2 needs a referent sampling technique, the sampling risks associated with the referent sampling plan carryover. In the case of destructive sampling, it makes sense for SkSP-2 to use a referent sampling plans associated with special level S-1 of MIL-STD-105E. Chain Sampling, like SkSP-2, also reduces the number of samples taken and is therefore well suited for destructive or costly tests. Dodge (1955b) mentions that when tests are destmctive more often than not a sampling plan using a small sample size is chosen. It is worthwhile to note that sampling techniques which reduce the number of samples due to such factors as destructive sampling, emphasize criteria such as lots being from a series of continuing supply and having essentially the same quaUty (Dodge, 1955a, 1955b) in order to overcome the occasional defective or to dampen the effect of large sampUng risks. 2.3 High Qualitv Sampling Although ChSP-1 is ideally suited for destructive sampling, it allows for a small percentage of imperfections in the production process. ChSP-1 allows for the random defective unit to be ignored if the previous /-lots have been accepted. Similarly, double, multiple and sequential sampling never reject a lot which only has one defective. Although Dodge (1955b) suggests that ChSp-1 has better characteristics than single sampling where the acceptance number is zero and rejection number is one, accepting a lot which has an occasional marginal failure is not well suited for high quality sampling. SkSP-2 also falls in this category if the referent sampling plan chosen is double or multiple sampling. In the case of single sampling or SkSP-2 (when single sampling is 28 chosen as a referent sampling process) a smaller size sampling plan such as the special level S-1 of MIL-STD-105E can be utilized. However, as the quality level increases to 5sigma and beyond, even these small sample size plans can increase to unreasonable rates. In the case of sequential sampling Appendix A illustrates mathematically how this sampling process reacts to very high quality levels. Note that at the 5 and 6-sigma levels of quality the numbers are incredibly high. For double sampling, Table 5.12 illustrates that a level close to a 5-sigma quality level (0.023%), the cumulative sample sizes are 1250 for the first sample and 2500 for the second sample. In automotive manufacturing and assembly plants where 6-sigma (Table 5.1) quality is expected by the clients, in the case of destructive sampling the suppliers are being asked to test the products using single sampling in accordance to the MIL-STD105E special quality level S-1, which is 5 units sampled per lot of 3000 units (E. Castillo and S. Alvarez, personal commimication, October 6, 2003). Actually, for lots between 501 and 35,000 units, sample size code letter C is used (Table 2.2). The lowest sample size for the lowest AQL using single sampling is 5 units at an AQL of 2.5% with an acceptance number of 0 and a rejection number of 1 (Table 5.5). At an AQL of 2.5% percent defective this implies that the single sampling process guarantees approximately a 3.5-sigma level of quality (Table 5.1). This implies that the suppUers are testing at much lower quality levels than their true quality production process. It is not uncommon for suppliers performing at very high quality levels to go months and sometimes over a year without encountering a single defective during destructive tests (E, Castillo and S. Alvarez, personal communication, October 6, 2003). 29 2.4 Empirical Bayesian Methods Case studies indicate that the Empirical Bayesian methods* performance can be much better than that of conventional, non-Bayesian methods (Maritz, 1970). One of the major differences that the Empirical Bayesian approach brings to the area of sampling is the incorporation of prior knowledge (Iversen, 1984), or a prior distribution on the unknown parameter. The prior distribution, which is specified by the researcher, should be as informative as possible and reflect specific knowledge about the population. The more informative the prior distribution, the better the parameter estimates tend to be. Iversen (1984) argues that non-informative prior distributions are much more usefiil than total ignorance, but is the lowest level of prior opinion about the parameter being studied. One example of a non-informative prior distribution is the uniform distribution (when used as a prior), since all values in the parameters are equally likely of occurring over a relevant range of values within the distribution. Iversen (1984) points out that the use of non-informative priors can lead to results that numerically correspond to those obtained from the use of classical statistics. There are several known situations where the results using classical statistics are equal to those using a non-informative prior. If a researcher is working with a prior distribution, which gives him/her more information than a noninformative prior, the results can be better than achieved using classical statistics (assuming that the information in the prior is correct). If a researcher has prior information and there is a procedure that will result in more accurate results when using this prior information, he/she should use this procedure. 30 One of the strengths of Empirical Bayesian Analysis is that it allows the use of the information from earlier research studies or tests in the analysis of new estimates. Researchers should seek prior knowledge whenever it*s available and use it as part of their research. Researchers also take a Bayesian approach when they take the posterior of a previous study and use it as a prior for the new study (Becker & Camarinopolous, 1990). In their work Becker and Camarinopolous introduced a model, which estimates the probability that a program might not contain any errors after some debugging. They suggest that if programs are simple and well developed, the chance that the last error found on the program truly is the last error. If the programs are simple and well developed the traditional software reliability models, which always claim that there is always one more bug to fix, will be v^ong. They allow the possibihty that the last bug truly is the last one. This is accomplished by applying Bayes' law repeatedly, each time taking the posterior of the last step as a prior of the next one. This multi-stage Bayesian approach allows the probability mass to get to a zero failure rate. An informative prior is not guaranteed to give better results than a noninformative prior. An important aspect of Empirical Bayesian Analysis is to determine the effects of a prior. Since priors provide the foundation on which Bayesian analysis rests (Iversen, 1984), it is not always the case that they determine the exact shape of the posterior distribution. If the researcher is dealing with a large sample the posterior distribution tends to be dominated by the data and not the prior distribution. Iversen (1984) calls this the principle of stable estimation, which states that even a somewhat 31 informative prior distribution has little or no effect on the posterior distribution if the researcher is dealing with large samples. In this type of situation the informative priors vs. non-informative priors would not make a big difference in the final results. It can be concluded that in a situation with a very large sample size, an expert determining the constants that determine the distribution of the priors would lead to approximately the same result than those coming from an uninformed person. It is worthwhile to note that if only limited information is available from the data different prior distributions will lead to different posterior distributions (Florens, et al., 1990). The reason the posterior distribution will depend more on the prior than on the sample data is because a small sample contains only limited information and the posterior distribution is not overwhelmed by the data in this case. With small sample sizes, the contribution of the prior information becomes more important. It follows then, that there is a difference between an expert and the novice making the decision on the constants, since the posterior distributions depend more heavily on the prior. In this case, the posterior distributions might not be exactly the same, since there might not be enough data to converge the results to a single posterior distribution. In the situation where the researcher is dealing with small samples, Iversen (1984) points out that it is very important to have expert determination of the constants, which determine the prior distribution. These prior distribution parameters become the analysts' opinion of a population parameter even before any new data is collected. 32 2.5 Acceptance Sampling vs. Quahty Monitoring Both acceptance sampling and quality monitoring attempt to compare the results of inspection to an objective criterion. However, they differ in the ways that they attempt to deliver a quality product. Acceptance sampling attempts to guarantee good quality by statistically determining whether the lots akeady produced should be accepted or rejected. Quality monitoring keeps the production process under surveillance in order to take corrective action. As the quality of a product increases in the case of destructive sampling, traditional acceptance sampling plans become unreasonable because they attempt to guarantee quality by statistical means. On the other hand, as the quality of a product increases, Bayesian methods adjust its parameters to the quality level and sample according to the historical quality level. While the traditional sampling methods become inflated and outrageous as quality increases, the Bayesian sampling method goes from an acceptance sampling technique to decreasing it's samphng numbers and becoming a quality monitoring technique. Control charts are a form of quality monitoring. While the/? chart and the np chart deal with monitoring the fraction rejected and the number of nonconforming items respectively, the c and u charts monitor the number of nonconformities and nonconformities per unit respectively (Grant & Leavenworth, 1988). If looking at the fraction rejected as nonconforming items to specifications {p chart), the upper control limit is equal to 33 UCL^=p + 3:P^^-P^ n, and the lower control limit is equal to iCi,= p - 3 J £ f i ^ where p is the probability of a reject and n is the number of units inspected in subgroup z (Nahmias, 1993). As the quality in the process increases/? decreases and the UCL approaches 0. This means that with one defective item, the lot would be rejected. For example, if the quality is at 3-sigma or 6.68% defective and n is 20 (somewhat small since tests are destructive), then the UCL = 0.234%. This means that 4.686 or less than 5 defective units per sample would still be considered under control. However, if the quality level is at 5-sigma or 0.00023 and « = 20, then the UCL = 0.01047%. This means that 0.2094 or less than one unit would have to be defective in order for the process to be under control. Therefore, as the quality level increases, the/? control chart becomes similar to a single sampling process with an acceptance number of 0 and a rejection number of 1. 2.4 Summary Acceptance sampling methods such as single, double, multiple and sequential sampling have been well studied and developed to provide a level of quality desired by the researcher. These methods are designed to guarantee a level of quality based on the sampling results. If the test required is destructive, then the supplier is forced to devise 34 plans to skip sampling lots (SkSP-2), form chains of lots (ChSP-1), or take sampling risks (MIL-STD-105E Level III and Level S-1) in order to keep the number of units tested at a minimum. If these lots are not skipped, chains are not formed or sampling risks are not taken, the destructive sampling experiment can be quite costly for the supplier. If the quahty is increased, the traditional methods, including methods devised for destructive sampling, require a large number of units to be sampled in order to guarantee the high quality. The problem is that the number of units sampled needs to be small for destructive sampling and large to guarantee high quality. In these situations, customers are asking suppliers to destructively test using lower quality sampling specifications even though the production process is performing at a higher quality level. Bayesian sampling methods are based on prior information or prior knowledge and are designed to sample at the level of quality deteraiined by prior information. This prior information can be in the form of previous sampling results or expert knowledge. As quahty increases the Bayesian method adjusts sampling to that particular quahty level and becomes a monitoring technique, while traditional methods require unreasonable sample sizes. Although all of the sampling techniques will be compared at different quality levels, of particular interest will be the comparison at very high quality between the traditional techniques at their highest quality sampling level available versus DSM-HQ. 35 CHAPTER m COST OF QUALITY This chapter examines the costs of finding a defective unit within the manufacturing organization as well as at the client site. An efficient model to determine sampling rates is developed based on these costs. In turn, the model is used to examine the best time to stop the sampling process and either reject or accept the lot. Therefore, the sampling number per lot will be economically-based in this research (Case & Keats, 1982). In other words, this research combines an economically based model into the sampling technique in order to arrive at the decision to accept or reject the lot. The chapter also gives details into how the economically-based model will be applied into the simulation in order to accurately deteraiine sampling costs and costs incurred as a result of defective items finding their way into the customer facilities. Finally, the binary search used to determine whether the process is a random occurrence or an out-of-control event is also discussed. In the case of an out-of-control event, an example on how the binary search determines the instance where the process went out-of-control and on how the costs are determined is given. 3.1 Economically-Based Cost Model In the economically-based model, let Yt be the random variable representing the number of defectives that occur in the unit time interval starting at t[t, t+l), and let At be the value of the parameters at time t. 36 The cost of finding a defective unit in the field, c/, is the cost that the customer must incur in order to fix the problem once the product has been shipped. In this situation, the supplier may have to send a team from the manufacturing plant, or sometimes if the chent is a long distance away, a third party is hired to solve the problem. In other instances the supplier may have to send a brand-new lot of product and bring back the lot that contains the defective item. In any case, this cost is quite high, especially if the client considers changing to a competitor because of continuous and repeating incidents of defective items. If the sampling experiment is stopped at time r, the expected cost over the time interval [/, ^ + 1) is equal to cfE(Y). Let/c be the fixed cost incurred to run the experiment and let vc be the variable cost of testing each unit. The variable cost, vc, is made up of the cost to test each additional item, the cost of scrap for each component destroyed in a destructive test, and the indirect results of having a higher cost of quality since scrap has increased. The fixed cost,/c, is associated with the minimum cost of components that have to be tested and/or destroyed in a destructive test. Whenever a defective unit is found, supplier goes through some additional costs to determine when and where the defect occurred. When a defective unit is found two instances are considered in this research: the process generates a random defective unit or the process is out-of-control. Whenever a defective unit is detected by the sampling technique, a process takes place to determine if the defective unit is a random occurrence, or a process that has gone out-of-control. In the case that the unit detected is a random occurrence, units produced before and after the defective unit will conform to 37 specifications. However, these units will be destroyed by the destructive test. In the case that the nonconforming unit is one of a process that has gone out-of-control, a process also takes place to determine where exactly the process went out-of-control. This process is much more expensive than the random defective unit, since all units will be defective from the time that the process went out-of-control. The supplier incurs a cost to find where the process went out-of-control, a cost to fix the problem in the production lines, and a cost to repair all defective parts in the out-of-control process that were not destructively tested. Let ci be the cost of finding a defective unit in-house through the sampling process. Therefore, the long-run average cost of finding a defective unit through the sampling process is ciE(Y). Let r be the probability of finding a random defective unit and o be the probability of finding an out-of-control defective unit during the sampling process. Also, let cr be the cost of finding a random defective unit during the sampling process and co be the cost of finding an out-of-control defective unit during the sampling process. The cost of finding a defective unit in-house during the sampling process is ci -rcr + oco The cost to test a particular lot (in the case of destructive testing) is equal to vcn +/c + ciE(Yy Typically, the cost in the field, cf is much more than the cost the supplier incurs within its own facilities, vc-n -^fc. + cvE(Y). In this situation, it is more economical to stop testing if cf'E{Y,)<vC'n + fc + ci'E{Y,), 38 If a decision on whether to accept or reject has not been made, it is economical to take an additional observation as long as c / - ^ ( y j > v c w + /c + d-£(y,). This model, which is similar to the one used by Dalai and Mallows (1988) as well as Randolph and Sahinoglu (1995), gives the optimal stopping point for testing, or sampling in this case. In this situation, the break-even point occurs at c/.j5:(yj = vc.n + /c+c/-£(y;), which gives cf-ci 3.2 Simulation Cost Function The simulation will determine the sampling costs and the costs incurred from defectives detected outside the supplier facilities for each of the samphng plans under different quahty levels and different defective patterns. The total cost of sampling is made up of the cost of all sampling experiments and the cost of all investigating experiments which determine whether a defective found is a random occurrence or a part of an out-of-control event. The cost of all sampling experiments is made up of variable costs, vc and fixed costs,yc. The total variable cost in sampling experiments will be made up of all the units that were tested during the sampling experiments multiplied by the variable cost (S«vc). 39 The total fixed cost will be the addition of all the fixed costs per sampling experiment (S/c). If a defective unit is found in-house, a separate experiment is required to determine whether the defective unit is a random occurrence or a defective in a series of an out-of-control event. Each one of these experiments has a fixed cost and variable costs associated with it. The cost of finding a defective unit in-house, ci, will be the costs incurred to determine if the experiment is a random occurrence, cr, or if it is an out-ofcontrol event, co. Then, the total costs of finding a defective unit in-house is Zc/ = Scr -iEco. The cost of each random occurrence, cr, will tend have a low cost and will be based on the search routine (explained in the next section). Only a few units before and after the random defective will be tested in order to ensure that it is not an out-of-control event. This experiment will have its own fixed cost,/c, and a variable cost, vc. Therefore, cr =fc + n/vc, where nr is the number of units destroyed to confirm that the process is a random occurrence. The cost of an out-of-control event, co, will tend to have a larger cost involving the search cost and the rework cost for the parts which are known to be defective but were not destructively tested. For simplicity, the parts that are out-ofcontrol and have not been tested will incur a rework cost of 50% of the variable cost. Therefore, co =fc + n^-vc + m(vc/2), where w^ is the number of units that had to be destructively tested in order to determine where the out-of-control event occurred, m is the number of out-of-control units which were not destructively tested and vc/2 is the rework cost. Therefore, the total cost of sampling, TCS, is equal to 40 7=0 ;=0 7-0 y-O 7=0 ^ 7=0 where k is the number of sampling experiments, / is the number of experiments where a random occurrence was found, and q is the number of experiments with an out-of-control event. The costs incurred from defectives detected outside the supplier facilities will simply be the cost of finding a defective at a chent site, cf, times the number of defectives which escaped the sampling process. This assumes that all defectives that get to the client are found. 3.3 Binary search to determine type of defective When a defective unit is encountered, all sampling processes in this research will go through a search routine to determine if the defective detected through destructive testing is a random occurrence or an out-of-control event. The simulation will use a binary search starting at the previous unit in the production process. The binary search will be conducted backwards. If there are no defectives at the previous four binary observations (1,2,4,8) or the next four binary observations, then the system will assume that it is a random defective. This means that Wr (from section 3.2) is equal to 8. However, if there is a problem in the previous observations, then the system will assume that there is an out-of-control event, and it will search in a binary method until it finds the point where the process went out-of-control. All cost in this search will be considered. A 41 search into the next units produced after the defective is encountered will also be conducted in case that the defective is the start of an out-of-control event. For example, consider an out-of-control event where the process went out-of-control 1257 units before the sampling process detects a defective. At this moment, a separate experiment with a fixed cost,yc and variable cost, vc, takes place to determine if the process is a random occurrence or an out-of-control event. The binary search begins by going 1, 2, 4, and 8 units into the past. Since there is an assumption in this research that a process that goes out-of-control cannot get back into control, all 4 units will be defective in this situation. The search continues to test 16, 32, 64,128, 256, 512, 1024, and 2048 units into the past. At unit, 2048, the item passes the destructive test and the binary search works downward to unit 1536 = 1024+(2048-1024)/2. Since this part also passes the sampling test, the search continues downward to unit 1280 = 1024+(1536-1024)/2, and then to unit 1152 = 1024H-(1280-1024)/2. This unit fails the test so the search moves up to unit 1216 = 1152+(1280-1152)/2 and then to unit 1248 = 1216H-(1280-1216)/2 and then to unit 1264 = 1248+(1280-1248)/2. At this point the item passes the sampling test and it moves down to unit 1256 = 1248+(1264-1248)/2, then to unit 1260 = 1256+(1264-1256), then 1258 = 1256+(1260-1256)/2, then 1257 = 1256+(1258-1256)/2. At this point the process determines that the out-of-control event happened at 1257 units before the detected defective unit. In total 22 items are destructively tested to arrive at the beginning of the out-of-control event. The number of units available for rework, m, is equal to 1257 - 14 (out-of-control units destructively tested) + number of units of production after defective 42 unit of production process is encountered. The cost of finding a defective unit in-house, ci, is fc + 22(vc) + m(vc/2), 3.4 Summary The economically based cost model presented in this chapter will be used to determine not only the sampling number but also the sampling rate for the DSM-HQ. Also, it is the basis for determining costs incurred by the other sampling processes. The simulation cost function section presents an accurate description on how the program will determine all costs associated with sampling and with defectives that found their way into the clients' facilities. The binary search provides a simplified approach to determine if the defective found is a random occurrence or an out-of-control event. In the case of an out-of-control event, the binary search determines where the event began, and the cost fimction determines the associated costs. 43 CHAPTER IV DEVELOPMENT OF A DESTRUCTIVE SAMPLING METHOD FOR HIGH QUALITY PRODUCTION PROCESSES This chapter reviews the topics that are used to support the model. Then it applies and adapts work by Berger (1985) in the area of Bayesian analysis to the problem of acceptance sampling. Because Bayesian analysis and the Poisson distribution are used in this research to represent the defective process, the prior distribution, marginal probability fimction, joint distribution fimction as well as posterior distributions are determined. The result is then combined with the cost fimction from Chapter III to develop the DSM-HQ model. Finally, an illustration of how DSM-HQ is given and outof-control rules are presented. 4.1 The Gamma Function and Gamma Distribution The gamma fimction, r(a), has a positive parameter a and is defined by the integral (DeGroot & Schervish, 2002) r ( a ) = Jjc"-'e-'dtc. For every positive integer n, r(n) = ( « - ! ) ! The continuous random variable X follows the gamma distribution with parameters a and p, where a > 0 and p > 0. It has a p.d.lf{x\a,^), 44 which is specified as T{a) Since the integral of this p.d.f is equal to 1 it follows that r(«) 4.2 The Poisson Distribution The Poisson distribution is commonly used to model the occurrences of random arrivals during an experiment. In addition, the Poisson distribution can be used to approximate the binomial distribution when the success (or failure) probability is very small (DeGroot & Schervish, 2002). Let X be a discrete random variable, which takes on nonnegative integer values. X has a Poisson distribution with mean X (A,>0) if the probability fimction of X is e nx\A)=\ A for x = 0,1,2,.... x\ 0 otherwise. For x=0, fix\n,p) = (l-py=\\- n The conjugate prior density fimction for the Poisson probability distribution is the Gamma distribution, which is given by Y{pc) 45 4.3 Empirical Bayesian Analysis for the Poisson Distribution Given a prior density fimction z(9) and the joint density is f(x\6) the joint probability fimction can be calculated as (Berger, 1985) h(x,d) = n(e)f(x\ey The marginal probability fimction can be calculated by integration m{x)^\n{e)f{x\e)d9. The posterior joint density fimction can be determined (provided that m(x) i^ 0) by m{x) As discussed in section 4.2, the Poisson distribution, our Bayesian joint probability, is given by f{y\hi)-- ; . Let the random variable, Y, be the number of defective items that occur from a lot with n items. The parameter X is used to indicate the average rate of arrival for defective components. To obtain the Bayesian prior density fimction, the prior distribution for the Poisson distribution is needed. As discussed in Section 4.2, the prior distribution commonly used with the Poisson probability fimction is the Gamma distribution. Under this setup, the prior density fimction for X is 7t{X) = T{a) 46 where a and p, both greater than zero, are the parameters of the initial prior density fimction. It should be noted that the posterior density fimction becomes the prior density fimction for the following sample. It follows that the joint distribution fimction of Y and X is the Poisson distribution given by the equation h{y,X)^n{X)f{y\X,n^, which is ,, ,, p^n'X^^'-^e-^^^^''^ r(a):^! It follows that the marginal probabihty fimction of y is given by the equation '"(>')= jh{y,X)dX, which is equivalent to go miy)=l7t(A)f(y\A,n)dA, and therefore ^ ( > ' ) =0 1 — ^ na)y\ ^i^xzi—^^ Solving for and using properties of the Gamma density fimction noted in Section 4.1 gives 47 f x " - e - ^ ^ = n ^ since f j ^ x ^ - ' e " ' ^ =1 and the marginal density fimction is m(y)~— ^^ ^^^. r(a)y\(P-^nr' Finally, the posterior joint density fimction of A is given by the equation: m{y) fi^n' Y{a + y) Y{a + y) " • Recall the gamma density function is given by f{x\a,P) = T{a) ^x--'e~^. Therefore, the posterior distribution produced a gamma density fimction with parameters a-^ydXid P+n. 4.4 Destructive Sampling Method designed for High Quality Production Processes This section presents a Destructive Sampling Method designed for High Quality Production Processes (DSM-HQ), which follow the Poisson distribution. From Section 4.2, the joint probability fimction is the Poisson distribution given by f{y\^)^ e'^'inXy y\ 48 where >^ = 1,2, 3 , . . . . and A > 0. From Ross (2000) the expected value of the Poisson distribution is E(Y) -X =oJp (Randolph & Sahinoglu, 1995). In this situation, X is the rate at which the defectives occur. For the prior distribution X is the rate at which the defectives have occurred in the past, a is the number of defectives that have occurred in the past, and ^ is the number of observations that have been made. Thus, the expected value of the posterior distribution is P+n where a + j ; is the total number of defective units, andyff + n is the total number of units sampled. The expected value from the Bayesian analysis can now be combined with the cost equation from Section 3.1 and the result is a-\-y P-\-n vc'n-\- fc cf -ci which implies vc n^ cf -ci fc-n cf ~ ci vc 2 cf -ci n + VC' pn fc- P cf -ci cf - ci cf-ci ) \cf-ci ~an^ -\-bn-¥c where vc a= cf-ci , fc + VC'P J fC'P .b = ^ ^ a n d c = •" ^ . - a cf-ci 49 cj-ci The zeros of this quadratic equation are / c + (vc.>ff)^ | ( / c + (vc./?)V _-b±ylb^ -4ac _ la cf-ci + \{ cf-ci ) ':^c cf-ci -4vc f fc-p {(cf-cif a ^ cf-ci ^ It is known that if the discriminant b^'4ac is positive there are two x intercepts, and if it is negative, there are no x intercepts (the parabola does not touch the x axis at all). In this situation a = vc/(cf- ci) >0, so the parabola is concave. If there are two x intercepts, the intercept of interest will be the higher x intercept of the two. The reason is that the higher intercept will yield the number of observations needed to ensure equilibrium between the two expected value formulas. Figure 4.1. illustrates the case where there are two x intercepts, where the highest intercept is the number of observations needed to ensure equilibrium in the economically based model. In this example, that the higher of the two X intercepts occurs at 6.44. This indicates that the starting number of units sampled in the experiment, should be between 6 and 7. 50 Positive X intercept -i;oo Figure 4.1: Positive higher JC intercept Negative X intercept -1200 Figure 4.2: Negative higher A: intercept Figure 4.2 shows that the higher of the two x intercepts is negative. In this case the model indicates that there is enough prior knowledge and that the process is so good that 51 the economical decision is to take no samples. In this example, the higher of the two x intercepts occurs at -1.50. This means that economically speaking, there is no need to take any fiirther samples and it is cost effective for the supplier to release the products to the chent. 4.5 Illustration of the DSM-HQ This section graphically illustrates how DSM-HQ automatically adjusts and samples according to the true quality level of the production process. In addition, it provides an example of how DSM-HQ handles defectives and samples after a defective event (random or out-of-control) has occurred. DSM-HQ automatically adjusts the sample size and sampling rate according to the sampling history and prior information. After all prior values and costs are set, the process indicates a starting sampling rate. The parameter yf is adjusted after each unit in the sample is tested, since value of )ff indicates the number of units that have been tested in the history of the unit's production process. If a defective unit is encountered, then the parameter a is adjusted. The parameter a is the number of defectives that have occurred in the unit's production history. Therefore, over time the sampled defective rate of the unit's history, X = a/p, should approach the true quality process level. Figure 4.3 illustrates how DSM-HQ begins at a starting sampling rate of approximately 7 units per hour and reduces the sampling rate to 1 unit per hour in 100 days. 52 Days vs. Sampling Rate (0 defects) From 7.42 units per hour to 1 unit per hour in 100 Days B.OO 7.00 6.00 5.00 « 4.00 a. E A « 3.00 2.00 1.00 0.00 20 40 60 80 100 120 Days Figure 4.3: DSM-HQ reduces sampling ratefi-om7 to 1 per hour (0 defectives are found) This means that if this plan was put into effect and no defectives are encountered, the supplier can reduce the sampling rate significantly within 3 1/2 months. Of course this would mean that the values for the prior information and the fixed and variable costs given in this plan would be consistent with the suppliers values. The value of ^ inversely affects the rate of decrease. If the prior value ofy? is increased, the rate of decrease of sampling rate vs. days is decreased. This means that the number of days to reach a sampling rate of one per hour with all other values held constant would be increased. Conversely, if the value of ^ is decreased, the rate of decrease of the sampling rate vs. days would be increased. The prior value of a directly affects the sampling rate. Therefore, if all other values are held constant, a lower prior a corresponds to a lower 53 starting sampling rate and a higher prior value of a results in a higher starting sampling rate. If DSM-HQ encounters a defective, the process automatically increases the sampling rate at a value consistent with the prior information. For instance, if the same values of the previous example for prior information and costs are used and a defective occurs on day 38 of the sampling process, then the sampling rate is adjustedfi"om2 to 5 units and the total amount of time to reduce to one unit per hour is increased to 174 days (Figure 4.4). Days vs. Sampling Rate (1 defect) 8.00 200 Figure 4.4: DSM-HQ adjusts for a defective at day 38 In this situation, the number of days increases fi-om 100 to 174 if a defective is encountered because the X level increases with the occurrence of the defective on day 38. 54 4.6 Out-of-control rules Each defective, whether it is a random occurrence or an out-of-control event, is treated as a single event. In both situations the value of a will be increased by only one. In the case of an out-of-control event many defective units are being produced and the cost of a process going out-of-control will be accoxmted for. However, it is not logical to add all the defective units of an out-of-control event into the fimction since it would increase the sampling rate to unreasonable numbers. Furthermore, an assumption is made that an out-of-control event cannot retum to an in-control situation. This means that once a process goes out-of-control it stays out-of-control. 4.6 Summary The economically-based cost fimction for a destructive test, Bayesian analysis, and the Poisson distribution are all combined to produce the final model for DSM-HQ. At this point, the DSM-HQ model can indicate the starting sampling rate according to fixed prior values and costs and dynamically react to results in the sampling process. In this chapter, a graphical illustration shows how DSM-HQ reacts to the sampling process if no defectives are encountered and how it reacts to a single defective. Several issues, which are hard to derive mathematically, remain to be investigated. The final mathematical fimction of DSM-HQ does not indicate at what level of quality the prior X needs to be set for a certain true quality process. Furthermore, the fimction does not suggest the appropriate amount of prior history for^S and a needed 55 for different quality process levels. Finally, the example shows only a reduction in the sample size, while keeping experiments at a rate of one per hour. If the aim is to reduce the rate of the experiment to one per day or one per week, the increase of rate vs. decrease of sample size needs to be investigated. All these issues, which can more easily be evaluated by the use of a simulation than by a mathematical process, are discussed in the following chapter. 56 CHAPTER V SIMULATION DESIGN The simulation design for this research will be separated into two stages. Stage one will consist on fine-tuning DSM-HQ. The paths to minimum sampling will be analyzed and the most appropriate one will be selected. Stage two will compare the existing techniques to the one proposed by this research. All techniques will be tested under different process sigma levelsfirom3 to 6 sigma. Furthermore, the techniques will be tested under random type occurrences and out-of-control events. 5.1 Stage 1 ofSimulation: Defining DSM-HO Although the DSM-HQ mathematical model can indicate the starting sampling level and dynamically react to the samphng process, there are issues pending that are more easily defined by a simulation than by a mathematical derivation. Without a simulation, the model does not indicate whether to first increase the experiment rate and then reduce the sample size or first reduce the sample size and then increase the experiment rate or to reduce sample size and increase experiment rate simultaneously. Therefore, the best path to arrive at the minimum allowable sampling rate will be investigated. Also, the DSM-HQ model does not indicate the most cost-efficient prior values for the parameters of the sampling process. Stage 1 of the simulation will finetune and finalize the DSM-HQ sampling process. It will recommend prior values {X, p, and a) for different quality processes and define what path to take to arrive at a minimum 57 sample size and a maximum time between samples. Included in the fine-tuning of the DSM-HQ is an investigation of a range of costs for the cost parameters, as well as, defining the defective patterns used. 5.1.1 Prior Distribution Parameters To establish the path to minimum sampling that DSM-HQ will take, parameters of the prior distribution must be set. For the prior distribution P represents the effective number of previous observations, X represents the historical defective rate and a represents the number of previous defectives. In practice the values of these parameters will have to be determined by a quahfied professional who is familiar with the process being examined. For the purposes of this research the values will be investigatedfi-oma 1.5-sigma below to 1-sigma level above the simulated sigma-level. Table 5.1, adapted fi-om Pande and Holpp (2002), indicates the process sigma-level and corresponding percent defective. 58 Table 5.1: Sigma Levels of Quality in terms of percent defective Process Sigma Defectives per Million Opportunities Percent Defective 6 3.4 0.00034 5.5 32 0.0032 5 233 0.0233 4.5 1350 0.1350 4 6,210 0.6210 3.5 22,750 2.2750 3 66,807 6.6807 2.5 158,655 15.8655 2 308,537 30.8537 1.5 500,000 50 1 691,642 69.1642 The prior levels, when different than the simulated sigma level, will adjust themselves to approach the simulated sigma level based on the results of the sampling process and simulation observations. For example, if the parameters are set at the 4sigma level and the process simulated is programmed at the 5-sigma level, the actual defective rate would be much lower than the starting defective rate. As the number of observations {P) increase during the simulation, the number of defectives (a) does not occur asfi-equentlyas expected. Therefore, the observed defective rate {X) is adjusted until it eventually approaches the 5-sigma level. The starting values for each tme quality process level are given in Table 5.2. 59 Table 5.2: Different prior starting points for the evaluation of the true sigma quality level Simulated Sigma Level Prior Sigma Level 3-sigma 1.5-sigma, 2-sigma, 2.5-sigma, 3-sigma, 3.5-sigma, 4-sigma 4-sigma 2.5-sigma, 3-sigma, 3.5-sigma, 4-sigma, 4.5-sigma, 5-sigma 5-sigma 3.5-sigma, 4-sigma, 4.5-sigma, 5-sigma, 5.5-sigma, 6-sigma 6-sigma 4.5-sigma, 5-sigma, 5.5-sigma, 6-sigma The simulation will examine different starting X's at different process levels to see the effects of different prior information. The amount of prior information within X will also be investigated. For example, at the 3-sigma level, X = 0.068 = a/p which could mean a = 68 and)9 = 1000, or a = 34 and^8 = 500, or a = 17 and^ = 250, or a = 136 and P = 2000. The effects of each of these values will be investigated at each of the quality levelsfi-om3 to 6-sigma. Of interest will be the amount of time that X takes to approximate the actual quality process and the effects on sampling size at the beginning and after a defective unit is encountered. The values of a and p investigated at each of the sigma levels is illustrated in Table 5.3. The prior infomiation will rangefiromthe lowest number possible for a up to a ^ of 1,000,000 units. 60 Table 5.3: Different a and p values for each of the prior X starting points Prior Sigma Level Starting a/p values 1.5-sigma 1/2,16/32,256/512,4096/8192, 65536/131072,500000/1000000. 2-sigma 1/3,16/52, 256/830, 4096/13227, 65536/212434, 308538/1000000. 2.5-sigma 1/6, 16/101, 256/1613, 4096/25810, 65536/412955, 158655/1000000 3-sigma 1/15, 16/240, 64/958, 256/3832, 4096/61317, 66807/1000000 3.5-sigma 1/44, 8/352, 64/2813, 512/22505, 4096/180044, 22750/1000000 4-sigma 1/161, 8/1288, 64/10306, 512/82448, 4096/659581, 6210/1000000 4.5-sigma 1/741,4/2963, 16/11852,64/47407,256/189630, 1350/1000000 5-sigma 1/4292, 2/8584, 4/17167,16/68670, 64/274678, 233/1000000 5.5-sigma 1/31250, 2/62500, 4/125000, 8/250000, 16/500000, 32/1000000 6-sigma 1/294117, 2/588235, 3/882353, 4/1176471 5.1.2 Cost Function Parameters The costs values can have a wide range in practice. For the purposes of this research, the costs considered will be for a unit with a relatively low fixed cost and low variable costs, and a substantial cost (in proportion to the fixed and variable costs) for finding a defective in the field. The variable cost, vc, of testing (and destroying) each unit will rangefi-oma relatively inexpensive $2.00 imit to a moderately expensive $100 unit. It is not common for units that are more expensive to be destructively tested since the supplier would be faced with a considerable loss. More expensive units usually go 61 through elaborate non-destructive tests and 100% inspection. The fixed cost incurred to perform the test will rangefi-om$5.00 to a more costly experiment of $200. The fixed costs will not be very high since the parts considered are not very expensive and it would be unreasonable to have extremely high fixed costs to mn a destructive experiment on low priced units. The cost of finding a defective unit in the field, cf, will vary fi-om $1,000 to $200,000. This cost tends to be very high since it is not uncommon for entire lots to be returned to the supplier. Sometimes the supplier must fly personnel to the customer's facilities to investigate and witness the situation. In other instances, third party members are called to examine the product and perform their own sampling and evaluation at the cost of the supplier. Finally, if their client cancels the contract it could mean a loss of profit in the hundreds of thousands of dollars. Table 5.4 illustrates the maximum and minimum values as well as other quantities investigated for each of the cost parameters. Table 5.4: Values considered for each of the cost parameters Cost Min Max Fixed Cost (/c) $5 $50 $200 Variable Cost (vc) $2 $20 $100 Field Cost {cf) $1,000 $10,000 $50,000 Once the prior variables and all costs are estabhshed, the model can determine samphng rate. After the starting sampling rate is determined, the paths to minimum sampling of DSM-HQ can be investigated. 62 5.1.3 Paths to minimum sampling The number of observations sampled in a determined amount of time will have to be negotiated with the customer. Mathematically, the number can go to zero, which will probably not be acceptable to the customer. In this research, the maximum amount of time between sampling will be set to one week, while the minimum number of observations tested will be set to one. In practice, these numbers can be adjusted based on the agreement between customer and supplier. The simulation will show at what point it becomes ineffective to continue increasing the time between sampling. Presently, customers in the automotive industry are asking manufacturers and assemblers to perform destructive testing on many of the components (E. Castillo and S. Alvarez, personal communication, October 6, 2003). The number of destructive tests will be designated nmax* which is the maximum number of units observed in destructive testing. On the other hand, nmim is the minimum number of units observed during destructive testing. Let tmtm be the minimum amount of time between tests, and tmax, be the maximum allowable time between tests. The objective is to getfi"omn^ax and t^in to nmin and tmax- This would imply that if there are no defectives eventually, with a good history, the process would movefi-oma frequent/maximum number of units destroyed to an infi-equent/minimum number of units destroyed. Minimum sampling is achieved at Hmin and tmax and is defined as the minimum number of observations sampled in the maximum allowable time between samples. 63 The first result of interestfi^omthe simulation is to determine the number of days it takes DSM-HW to bring the number of needed samples to a minimum and the largest amount of time between samples if no defectives are found. To accomplish this prior information, such as a history of previous observations, and an approximate error rate is needed. Additionally, fixed costs, variable costs and costs of detecting a defective item in the field will be needed. The two dimensional pathfi-omseveral observations per hour to one observation per week must be determined. The different paths to be investigated are discussed below. The paths will be discussed in the idealized situation of no defectives occurring. Path type 1 reduces the number of observations of seven per hour, to one per hour, and then continues to reducefi"omone per hour to one every week. Path type 1 is illustrated in Figure 5.1. n nmin > ^max ^ Figure 5.1: Reduction of sample size and sampling interval for path type 1. Path type 2 increases the time periods between samplingfi-omtmm to tmax. while holding the number of samples destroyed constant, and then once tmax is reached n is reduced fi-om nmax to nmin- Path type 2 is illustrated in Figure 5.2. 64 n n. n. min Figure 5.2: Reduction of sample size and sampling interval for path type 2. Path type 3 gradually reduces n and gradually increases t simultaneously. The path type 3 results in a more direct linefi-omthe coordinates {tmtnf ^max) to (tmax. ^min)- The path type 3 is illustrated in Figure 5.3. n nmax > *'max ' Figure 5.3: Conceptual view of reduction of sample size and sampling interval for path type 3 The path will not be continuous, since the number of observations is discrete. Furthermore, although the time units are continuous, the units will be treated discretely for convenience. Therefore, path type 3 will look more like Figure 5.4. 65 ^max ^ '^min ^min H ^max ^ Figure 5.4: Actual view of reduction of sample size and sampling interval for path type 3. The simulation will be to show which of these three methods is most practical and effective in terms of getting to the objective of nmin and tmax as well as allowing a minimum amount of risk in terms of defective units going to the customer undetected under different types of quality production. In addition, an out-of-control event at different stages of production will help determine which path best serves the supplier and the client. The first simulation will establish the research's final model. The simulation will give guidelines for an upper limit to tmax- As the times between samples increase, the costs of an out-of-control event will dramatically increase. The simulation will help determine a point where the maximum time between samples should take place. 5.1.4 Defective Pattems The simulation will consider two different defective pattems. The first defective pattem will be that of a process in-control with the defective rate being determined at a predetermined sigma level. The second defective pattem will have an out-of-control 66 event added to the first defective pattem. The timing of the out-of-control event will be randomly chosen. Figures 5.5 and 5.6 depict the two defective pattems. Status 1 non-conforming conforming -mmK Figure 5.5: Random defective Status 1 non-conforming conforming -- Figure 5.6: Process going out-of-control In the out-of-control event the set of defectives will be treated as a single instance that caused the defective. If all the defectives in an out-of-control event are included in a, then the process will take an unreasonable amount of time to be reduced to minimum sampling (or nmin and tmax) and an inappropriate amount of sampling would be conducted. Also, if the out-of-control process is detected and corrected before it leaves the supplier facilities, then the outgoing quality can be maintained at a very high level. However, the costs of determining where the problem occurred and the cost of replacing all the defective parts will be accounted for. 67 5.1.5 Measures to be Calculated The structure of factors will consist of 4 sigma levels tested (3 to 6 sigma), with starting sigma levels 1.5 sigma levels below to 1 sigma-level above, with 6 sigma being the exception and only being testedfi-om4.5-sigma to 6 sigma each having 6 separate a/p except for 6-sigma which has 4 oJps, It will also consist of 3/c, 3 vc, 3 cf, 2 defective pattems and 3 paths to minimum sampling. The 3 and 4-sigma levels tested have 6 different starting points with 6 different a/p. The 5-sigma level has also 6 starting points but of those only five have 6 different aJp and one has 4 different a/p. The 6-sigma level has 4 starting points and only the first 3 have 6 different a/p, while the last one has 4 different aJp. This makes it [(2)(6)(6)] + [(6)(5)+4] + [(6)(3)+4] = 128 combinations of sigma levels and prior values. These 128 sigma-level combinations will be tested at Zfc, 3 vc, 1> cf,l defective pattems and 3 paths to minimum sampling, making it a total of (128)(3)(3)(3)(2)(3) = 20,736 combinations. The simulation will be programmed using SAS, which will facilitate the generation of defectives, the series, and the out-of-control events using its random number generator. The out-of-control event will occur according to the defective rate programmed into the simulation. If the out-of-control event at the 5-sigma defective level is investigated, the out-of-control will occur on average every 233 times per one million observations. 68 5.1.6 Summary of Stage 1 Simulation During Stage 1 of the simulation the DSM-HQ will be completely defined. Not only will the model be able to indicate the most cost-efficient sample size and dynamically adjust if any defectives occur, but also the ideal path to minimum sampling will be resolved and the starting sigma-level with appropriate prior values will be defined. 5.2 Stage 2 ofSimulation: Simulation design for the comparison of sampling techniques To facilitate the comparisons in the second stage of the simulation lots of 1000 units per hour will be used. The length of the series will go 10 years into the fiiture. This amount of time will magnify the costs and should provide a clear-cut distinction between sampling methods in terms of cost and efficiency. Sequential sampling will not be used for comparison since the multiple sampling plans yield very close results (Grant & Leavenworth, 1988). Current techniques such as classical/single sampling, double sampling, multiple sampling, chain sampling, skip-lot sampling and MIL-STD-105E, will be compared against DSM-HQ. The tables for the latest MIL-STD-105 version E (DOD, 1989) will be used for single sampling, double sampling and multiple sampling plans as well as reduced, normal and tightened inspection of MIL-STD-105E. Because of the nature of destractive sampling relatively small sample sizes are necessary and large sampling risks must be tolerated. Therefore, special level S-1 of MIL-STD-105E will be used for single, double and multiple sampling. Three levels of inspection (Level I, H, III) will be used for 69 the MIL-STD-105E reduced, normal and tightened inspection. Advantages and disadvantages for each of the processes will be noted and discussed. The comparisons will be made between the 3 and 6 sigma levels of quality. Table 5.1 defines in terms of percent defective these sigma levels of quality. The following sections discuss details of the traditional sampling techniques under the different quality sigma levels. Also, the cost function variables, sigma levels used and specifications of DSM-HQ are also discussed. Finally, the measures to be compared are considered and a summary is presented. 5.2.1 Sigma Levels The comparison of all techniques, including DSM-HQ, will be conducted at the 3 to 6-sigma levels of quality. Although the main focus will be at the 5 and 6-sigma levels of quality, the comparison will be done at lower levels to determine at what point DSMHQ starts developing an advantage (if any) over the other sampling processes. Also of interest will be the advantages and disadvantages that the traditional methods will have between one another under different defective pattems and different cost values. 5.2.2 Defective Pattems The defective pattems will be similar to the ones examined in Stage 1. The first defective pattem considered will be that of a process that produces random defectives at each of the sigma quality levels. This pattem will be in-control and the defective rate will correspond to each of the quality sigma levels. The second defective pattem will 70 consider an out-of-control event added to the random defective pattern discussed above. The timing of the out-of-control event will be randomly chosen with the same likelihood for a process to go out-of-control or have a random defective. 5.2.3 Cost Function Parameters The same cost fimction parameters as in Stage 1 will be used. The range of each parameter will also be the same as in Stage 1. This means that each unit will have a relatively inexpensive fixed cost and variable cost to a moderately expensive fixed cost and variable cost. The cost of finding a defective in the field will again have a wide range from $1,000 to $50,000. Table 5.4 gives the range of cost values for each of the parameters. 5.2.4 Specification based on results of Stage 1 From the results of the simulation in Stage 1, DSM-HQ will have a defined starting sigma-level for each tme quality process level. It will also have a cost-efficient starting prior history a/p combination for each of the quahty process levels. Furthermore, the best path to minimum sampling will be known. Therefore, DSM-HQ will be finalized and completely ready to be tested against the traditional sampling plans. 5.2.5 Single Sampling In the case of single or classical sampling, MIL-STD-105E is used to determine the appropriate sampling plans. Because of the nature of destmctive sampling small 71 sample sizes must be used and therefore special level S-1 of the MIL-STD-105E is utilized. A section of the MIL-STD-105E, which shows the sample size code letters, is shown in Table 2.2. Since this research will consider sample sizes of 1000, the corresponding sample size code letter is letter C for Level S-1. Table 5.5 shows a portion of the sampling plans for MIL-STD-105E sample code letter C. At the 3-sigma level performance, or 6.6 percent defective. Table 5.5 suggests that the single sampling plan should refer to sample size code letter D. Table 5.6 shows a section of the sampling plans for sample code letter D. For an AQL of 6.5, the single sampling plan has an acceptance number is 1 and the rejection number is 2 with a sample size of 8. The 4-sigma level of performance indicates that the defective percentage is around 0.62% defective (Table 5.1). Table 5.5, which gives the sampling plans for sample size code letter C, indicates that for AQL levels less than 2.5% defective the user should use the next subsequent sample size code letter for which acceptance and rejection numbers are available. In this case, it is the sampling plans for sample size code letter F. Table 5.7 gives the sampling plans for sample size code letter F, which gives an AQL of 0.65%. This level is approximately equal to the 0.62% of the 4-sigma level. For the single sampling plan Ac is equal to 0 while Re is equal to 1 with a sample size of 20. At the 5-sigma level of performance, the defective percentage level is 0.0233% (Table 5.1). For single or classical sampling, double sampling and multiple sampling, the AQL given in Table 5.5 indicates that the next subsequent sample size code letter for which acceptance and rejection numbers are available should be used. In this situation, it is the sampling plans for sample size code letter N (Table 5.11). The sampling size for a 72 single sampling plan has a sampling size of 500 with mAc^O and Re=l. It is obvious that these numbers are unreasonable for destmctive testing. At special level S-1, the single sampling plan with the lowest sample size/AQL combination available for a 1000 unit lot is given by sample size code letter C. This plan is at an AQL of 2.5% defective (approximately 3.5-sigma level of quality) and a sample size of 5 units with an ^ c = 0 and /?e = 1. This plan and the specifications for the plan used at the 4-sigma level will be used at the 4, 5 and 6-sigma levels. 5.2.6 Double Sampling For double sampling the special level S-1 of the MIL-STD-105E will also be used. Although there be sampling risks associated with using this special level, the smaller sample numbers makes it reasonable for destmctive sampling to take these risks. Looking at table 2.2, the sample size code letter associated with a 1000 unit lot at level S1 is sample size code letter C. Table 5.1 indicates that the 3-sigma quality level has a 6.6% defective rate. Table 5.5 indicates that for a sample size code letter C, a double sampling plan and an AQL of 6.5 percent defective the user should refer to sample size code letter D. Table 5.5 indicates that for these same values the number of samples to be tested should be 5 items per sample. The corresponding acceptance number for the lot is zero defectives, while the rejection number for the lot is two defectives. If there is only one defective, then 5 more items are sampled and the lot is accepted if there is one defective in the 10 items sampled or is rejected if there are two or more defective items. 73 Table 5.1 indicates that at the 4-sigma level of quality the defective percentage is at 0.62%. The sample size code letter C table instmcts the user to use the next subsequent sample size code letter which acceptance and rejection numbers are available for quality levels under 2.5%. In this situation, it is sample size code letter J (Table 5.9). Table 5.9 indicates that sample sizes of 50 should be taken for double sampling. At an AQL of 0.65%, Ac is equal to 0 for the first sample and 1 for the second sample. Re is equal to 2 for the first and second samples. At the 5-sigma level of quality the defective percentage is at 0.0233% (Table 5.1), The double sampling methods recommend using a single sampling plan or using code letter R. Using code letter R (Table 5.12) the sampling plan for double sampling at an AQL of 0.025% are two samples of 1250. These numbers are unreasonable since the lot sizes are 1000 units. In the case of destmctive sampling all of the units would have to be destroyed. For a lot of 1000 units and special sampling level S-1, the lowest sample size/AQL combination available is found in Table 5.7 (sample size code letter F). Table 5.7 gives a double sampling plan of 13 units per sample size and Ac-0,1 and Re = 2, 2. This plan, along with the specifications for the 4-sigma level plan will be used when testing at the 4, 5 and 6-sigma levels. 5.2.7 Multiple Sampling The multiple sampling plan also uses special level S-1 of the MIL-STD-105E, which indicates that sample size code letter C is applicable. Table 5.5 indicates that 74 Sample size code letter D should be used for cases where 6.5% defective (3-sigma) is apphcable. Table 5.6 (Code Letter D) indicates that for the multiple sampling plan a sample size of two is required. The lot cannot be accepted until the third sample is taken indicating that at least 6 units must be tested if the lot is to be accepted. The rejection number for the first sample is 2, which means that both the units sampled must be defective for the lot to be rejected. The maximum number of units that will be sampled under this plan is 14 imits, while the minimum 2. Multiple sampling for the 4-sigma level (0.62%) requires the user to start at sample size code letter C and move to sample size code letter J (as in double sampling). Table 5.9 indicates that samples of 20 are tested at a time with a maximum of 140 units tested. The lot can not be accepted until the third sample which means that a minimum of 60 units need to be tested if the lot is to be accepted. Acceptance numbers rangefi-om0 to 2, while the rejection numbers are rangefi-om2 to 3 for the first through seventh samples. At the 5-sigma level (0.0233% defective) MIL-STD-105E the first available sample size code letter for which acceptance and rejection numbers are available is sample size code letter R (Table 5.12). The sampling plan for multiple sampling and an AQL of 0.025% correspond to samples of 500 units and the lot cannot be accepted until the third sample is taken. As in double sampling, all of the parts would have to be destmctively tested in order to know if the lot should have been accepted. Table 5.7 (sample size code letter F) gives the plan with the lowest sample size/AQL combination for a lot of 1000 units under special level S-1. Table 5.7 shows 75 that at an AQL = 2.5% the multiple sampling plan has sample sizes of 5 with acceptance numbers ranging from 0 to 2 and rejection numbers rangingfi-om2 to 3. This plan, together with the plan for the 4-sigma level will be used at the 4, 5 and 6-sigma levels. 5.2.8 Skip-lot Sampling The Skip-lot Sampling plan will use the single sampling plan as a referent plan. This implies that at the 3-sigma level of quahty sample size code letter D will be used (see Section 5.2.5). The sample size of 8 is used at the 3-sigma or 6.5% percent defective level of AQL. Table 5.13 contains the Average Outgoing Quality Limit (AOQL) values needed to arrive at the SkSP-2 sampling plan. For sample size code letter D, a sample size of 8 and an AQL of 6.5, the corresponding AOQL is 11%. From Table 2.1 a plan for the number of successive lots found to be conforming, /, and the fraction of lots to be inspected during sampling,/, can be chosen. At an AOQL of 11%,/= 1/2 and / = 2. Therefore, if two successive lots are found to be conforming, then 1/2 of the lots should be inspected during sampling. At the 4-sigma level the single sampling plan uses sample size code letter F (Section 5.2.5). This means that an AQL of 0.65%, sample size code letter F, and a sample size of20, the corresponding AOQL is 1.8% (Table 5.13). From Table 2.1/= 1/2 and / = 15 where the AOQL = 1.8%. This means that if 15 successive lots are found to be conforming, then 1/2 of the lots should be inspected during sampling. The other high quality single sampling referent plan has an AQL of 2.5% and an AOQL of 7.4% (Table 5.13). Table 2.1 indicates/= 1/5 and z = 9 at an AOQL of 7.5%. 76 Therefore, if 9 consecutive lots have no defectives, 1/5 of the lots should be inspected during sampling. For the 4, 5 and 6-sigma level of quality this SkSP-2 sampling plan and the 4-sigma level sampling plan will be used. Because the SkSP-2 uses a referent sampling plan such as single, double or multiple sampling, at 5 and 6-sigma levels of quality 50% to 100% of a lot would have to be destmctively tested. Therefore, it does not matter how many lots are skipped, since it is not reasonable to destructively test half or an entire lot. 5.2.9 Chain Sampling The sampling plan for the Chain Sampling Plan (ChSP-1) at the 3-sigma level will use an AQL of 6.5% and an AOQL of 11% as in SkSP-2 (Section 5.2.8). Therefore the number of preceding samples, i, for a plan where the AQL is 6.5% and the AOQL is 11% is 1, while then umber of samples, n, is set to 5 (Soundararajan, 1978). The acceptance number is set at 0 (or 1 if there are no defectives in the preceding i = 1 lot) while the rejection number is set to 2 or more, (or 1 if there is one or more defectives in the preceding / = 1 lot). The ChSP-1 at the 4-sigma level uses an AQL of 0.65% while the corresponding AOQL is 1.8% (Section 5.2.8). Therefore, this plan will use a sample size of n = 19 and the number of preceding samples, i, will be set to 4 (Soundararajan, 1978). The acceptance number is set at 0 (or 1 if there are no defectives in the preceding / = 4 lots) while the rejection number is set to 2 or more, (or 1 if there are one or more defectives in the preceding / = 4 lots). 77 The ChSP-1 at the 5-sigma level uses an AQL of 0.023%.while the corresponding AOQL is 0.074% (Table 5.13). This imphes that this plan requires a sample size of more than n = 504 and the number of preceding samples, / = 1 (Soundararajan, 1978). To be consistent with single, double multiple sampling and the SkSP-2 plan, the 3.5-sigma plan is also considered. At an AQL = 2.5% and n~5, the AOQL = 7.4% (Table 5.13). Therefore, / is equal to 4 (Soundararajan, 1978). This means that the plan will use a sample size of n = 5 and the number of links in the chain is set to 4 with an acceptance number of 0 (or 1 if there are no defectives in the preceding 4 lots) and a rejection number of 2 (or one if there are one or more defectives in the preceding 4 lots). This plan and the specifications for the 4-sigma level of quality will be used at the 5 and 6-sigma levels. 5.2.10 MIL-STD-105E For "normal" inspection the MIL-STD-105E at a lot size of 1000 units the recommended sample size code letter is J (Table 2.2). Using double sampling as a referent plan, at an AQL of 6.5% sample sizes of 50 units are required (Table 5.9). The acceptance number, Ac, will be equal to 5 for the first sample and 12 for the second sample. The rejection number. Re, will be equal to 9 for the first sample and 13 for the second sample. If ten consecutive lots conform to specifications then the process will move to "reduced" inspection. For a lot size of 1000 the corresponding sample size code letter for a reduced plan is G (Table 2.2). Under reduced inspection for an AQL of 6.5% defective Ac = 2 and 6, while Re = 5 and 7 for the first and second samples respectively 78 with sample sizes of 20 for each of the two samples (Table 5.8). If a lot is rejected under reduced inspection or if the number of defectives fall between 3 and 4 (v4c<defectives<i?e), or conditions warrant, then the process will move again to normal inspection. Under normal inspection, if two out of five consecutive lots are rejected, then the process moves to "Tightened" inspection. For 1000 units, the sample size code letter for Level III or tightened inspection is K (Table 2.2). Under tightened inspection and for an AQL of 6.5 Ac is equal to 7 and 18 and Re is equal to 11 and 19 for the first and second samples respectively (Table 5.10). If 10 consecutive lots remain under tightened inspection, then the process is stopped. If 5 consecutive lots are accepted, then the process moves to normal inspection. At the 4-sigma level of quality and an AQL level of 0.65%, Ac is equal to 0 and 1 for the first and second samples respectively while Re is equal to 2 (Table 5.9). The sample size undei this plan is 50 units per sample. Level I (reduced inspection) has a corresponding sample size letter G. At an AQL level of 0.65%, sample code letter F should be used (Table 5.8). The sampling plans for sample size code letter F (Table 5.7) indicates that for double sampling at an AQL level of 0.65%, single sampling should be used. This means that a sample size of 20 is taken and Ac = 0 while Re=l, Level III (tightened inspection) has a corresponding sample size letter K. This implies that a sample size of 80 units is taken under a double sampling plan (Table 5.10). Ac = 0 and 3, while Re = 3 and 4 for the fnst 80 and second 80 units respectively. At an AQL level of 0.025% (5-sigma) Table 5.9 recommends to go the most subsequent sample size code letter for which acceptance and rejection numbers are 79 available. In this situation it is the sampling plan for sample size code letter R. As discussed above, the sample sizes for double sampling are 1250. Under reduced inspection (Level I) the sample size code letter used is G (Table 5.8), which suggests finding the subsequent sample size code letter for available acceptance and rejection numbers. Again, the subsequent sample size code letter for which acceptance and rejection numbers are available is sample size code letter R (for double sampling). For Level III (tightened inspection) the sample size code letter is K. Table 5.10 suggests to find the next subsequent sample size code letter for which acceptance and rejection numbers are available, which again is sample size code letter R. For all 3 levels, sample sizes of 1250 need to be tested. There is no need to test at the 3.5 sigma level of quality at higher levels, since at an AQL = 2.5% the sample sizes are still 50 and 100 as in the 3 and 4-sigma levels (Table 5.7). The only things that vary are the acceptance and rejection numbers. Therefore, at the 4, 5 and 6-sigma levels, only the sampling plan for 4-sigma will be used. 5.2.11 Measures to be Calculated and Compared One of the important measures calculated and compared is the economic impact that each one of these techniques has on the suppher at each level of quality. Samphng costs, the total costs of finding defectives in the field and the total costs will be examined. The sampling costs will be made up of the number of units tested and the cost to test each unit (Evc-n) plus the total fixed costs (S/c) plus the total cost of finding a defective unit in-house and determining if the process is out-of-control event or a random occurrence 80 (Zc/). The total cost associated with finding defectives in the field will be the number of defectives that got through to the client, times the cost of finding a defective in the field {cf). The total cost will be the sampling costs plus the total costs of finding a defective in the field. The number of defectives detected and not detected by each of the sampling processes at each of the levels of quality will be compared. Finally, the Type I and Type II errors for each of the processes under each of the quality levels will be compared. There are 7 sampling processes (single, double, multiple sampling, SkSP-2, ChSP-1, MIL-STD-105E, and DSM-HQ). Of the 7, the first 5 are being tested at the 3 to 6-sigma level using 3.5 and 4-sigma level specifications to test at the 4 to 6 sigma, which results in [(5)(l+24-2+2)] = 35 tests. MIL-STD will use 3-sigma specs to test at the 3sigma level and 4-sigma specs to test at the 4 to 6-sigma levels for a total of 4 tests. DSM-HQ will test at each level with its appropriate specifications for a total of 4 tests. There are again 3fc,Z vc, 3 c/'and two defective pattems making it a total of (35+4+4)(3)(3)(3)(2) = 2,322 combinations to be investigated. 5,2.12 Summary It can be seen above that none of established methods are designed for very high quality processes. Even ChSP-1, which claims to be specially designed for situations where small samples are needed due to costly or destmctive testing, falls apart at very high quality levels. The reason is that while these sampling methods are designed to 81 guarantee a quality level based on samphng, DSM-HQ is designed to sample at the quality level of the production process. It does not make sense to follow the sampling techniques specifications at the 5 and 6-sigma levels for the estabhshed methods. This research will consider the sampling technique specifications at the 3.5-sigma and 4-sigma levels even when the production quality is at the 5-sigma and 6-sigma levels of quality. The reasoning behind this is that 4-sigma is the highest level of quality where the sampling can be tested at high, yet somewhat reasonable sampling numbers while 3.5-sigma offers a somewhat high quality with low sampling numbers. Therefore, the 3-sigma quality process will be tested at the 3-sigma level specifications, while the 4, 5, and 6-sigma quality processes will be tested and evaluated at the 3.5-sigma and 4-sigma sampling specifications. At the 5 and 6-sigma levels DSM-HQ will probably have a starting point below the actual quality process level. Intuitively it may be safe to say that Stage 1 will recommend values below the tme process quality level and let DSM-HQ adjust to perform the proper sampling rate. The X value will determined at Stage 1 of the simulation with appropriate prior values for a and p. If the process is tmly world class, X will eventually reach the appropriate sigma level and the sampling process will adjust itself accordingly to test at that level of performance. Table 5.14a and Table 5.14b summarize the acceptance and rejection numbers as well as the sample size for each of the sampling plans under each of the sigma level of production. 82 Table 5.5: MIL-STD-105E Sampling Plans for Sample Size Code Letter C AQL (percent defective) Sampling Plan Sample Size <2.5 2.5 4.0 6.5 10 Ac Re Ac Re Ac Re Ac Re Ac Re (Cumulative) Single 5 T 0 1 Use Use 1 2 Double 3 T • Code Code 0 2 Letter Letter 1 2 B D T 6 Multiple T • T = Use next subsequent sample size code letter for which acceptance and rejection numbers are available. Ac = Acceptance number. Re = Rejection number. • = Use single sampling plan above (or alternatively use code letter F) t = Use double sampling plan above (or alternatively use code letter D) 83 Table 5.6: MIL-STD-105E Sampling Plans for Sample Size Code Letter D AQL (percent defective) Sampling Sample Size <1.5 1.5 2.5 4.0 6.5 Ac Re Ac Re Ac Re Ac Re Ac Re T 0 1 T • (Cumulative) Plan Single 8 Double 5 1 2 0 2 1 2 10 Multiple 2 Use Use * 2 4 Code Code * 2 6 Letter Letter 0 2 C E 0 3 8 T • 10 1 3 12 1 3 14 2 3 _ __ T = Use next subsequent sample size code letter for which acceptance and rejection numbers are available. Ac = Acceptance niunber. Re = Rejection number. • = Use single sampling plan above (or alternatively use code letter F) 84 Table 5.7: MIL-STD-105E Sampling Plans for Sample Size Code Letter F AQL (percent defective) Sampling Plan Sample Size <0.65 0.65 1.0 2.5 6.5 Ac Re Ac Re Ac Re Ac Re Ac Re T 0 1 1 2 3 4 0 2 1 4 T • 1 2 4 5 (Cumulative) Single 20 Double 13 26 Multiple 5 Use * 2 * 3 10 Code * 2 0 3 15 Letter 0 2 1 4 E 0 3 2 5 25 1 3 3 6 30 1 3 4 6 35 2 3 6 7 20 T • • = Use next subsequent sample size code letter for which acceptance and rejection numbers are available. Ac = Acceptance number. Re = Rejection number. • = Use single sampling plan above (or alternatively use code J) 85 Table 5.8: MIL-STD-105E Sampling Plans for Sample Size Code Letter G AQL (percent defective) Sampling Plan Sample Size <.40 0.40 0.65 2.5 6.5 Ac Re Ac Re Ac Re Ac Re Ac Re • 0 1 2 3 5 6 0 3 2 5 • • 3 4 6 7 (Cumulative) Single 32 Double 20 40 Multiple 8 Use * 2 * 4 16 Code 0 3 1 5 24 Letter 0 3 2 6 F 1 4 3 7 40 2 4 5 8 48 3 5 7 9 56 4 5 9 10 32 T • T = Use next subsequent sample size code letter for which acceptance and rejection numbers are available. Ac = Acceptance number. Re = Rejection number. • = Use single sampling plan above (or alternatively use code letter F) 86 Table 5.9: MIL-STD-105E Sampling Plans for Sample Size Code Letter J AQL (percent defective) Sampling Plan Sample Size <.15 .15 0.65 2.5 6.5 Ac Re Ac Re Ac Re Ac Re Ac Re T 0 1 1 2 5 6 10 11 0 2 2 5 5 9 • • 100 1 2 6 7 12 13 20 * 2 * 4 0 5 40 * 2 1 5 3 8 60 0 2 2 6 6 10 0 3 3 7 8 13 100 1 3 5 8 11 15 120 1 3 7 9 14 17 140 2 3 9 10 18 19 (Cumulative) Single 80 Double 50 Multiple 80 T • T = Use next subsequent sample size code letter for which acceptance and rejection numbers are available. Ac = Acceptance number. Re = Rejection number. • = Use single sampling plan above (or alternatively use code letter F) 87 Table 5.10: MIL-STD-105E Sampling Plans for Sample Size Code Letter K AQL (percent defective) Sampling Plan Sample Size <.10 0.10 0.25 0.65 6.5 Ac Re Ac Re Ac Re Ac Re Ac Re T 0 1 2 3 14 15 0 3 7 11 3 4 18 19 (Cumulative) Single 125 Double 80 T • 160 Multiple 32 Use * 2 1 7 64 Code 0 3 4 10 96 Letter 0 3 8 13 L 1 4 12 17 160 2 4 17 20 192 3 5 21 23 224 4 5 25 26 120 T • T = Use next subsequent sample size code letter for which acceptance and rejection numbers are available. Ac = Acceptance number. Re = Rejection munber. • = Use single sampling plan above (or alternatively use code letter H) 88 Table 5.11: MIL-STD-105E Sampling Plans for Sample Size Code Letter N AQL (percent defective) Sampling Plan Sample Size < 0.025 0.025 0.065 0.25 0.65 Ac Re Ac Re Ac Re Ac Re Ac Re T 0 1 3 4 7 8 1 4 3 7 • • 4 5 8 9 (Cumulative) Single 500 Double 315 630 Multiple 125 Use * 3 0 4 250 Code 0 3 1 6 375 Letter 1 4 3 8 P 2 5 5 10 625 3 6 7 11 750 4 6 10 12 875 6 7 13 14 500 T • T = Use next subsequent sample size code letter for which acceptance and rejection numbers are available. Ac = Acceptance number. Re = Rejection number. • = Use single sampling plan above (or alternatively use code R) 89 Table 5.12: MIL-STD-105E Sampling Plans for Sample Size Code Letter R AQL (percent defective) Sampling Plan Sample Size 0.010 0.025 0.065 0.25 0.65 Ac Re Ac Re Ac Re Ac Re Ac Re (Cumulative) Single 2000 Use 1 2 3 4 10 11 21 22 Double 1250 Code 0 2 1 4 5 9 11 16 2500 Letter 1 2 4 5 12 13 26 27 500 Q * 2 3 0 5 2 9 1000 * 2 0 3 3 8 7 14 1500 0 2 1 4 6 10 13 19 2000 0 3 2 5 8 13 19 25 2500 1 3 3 6 11 15 25 29 3000 1 3 4 6 14 17 31 33 3500 2 3 6 7 18 19 37 38 Multiple Ac = Acceptance number. Re = Rejection number. • = Use single sampling plan above 90 Table 5.14a: Sampling Plansfrom3 to 6 sigma level of production quality Sigma Level 3 4 through 6 4 through 6 3 SampHng Plan Single Single Single Double 4 through 6 Double 4 through 6 Double 3 Multiple 4 through 6 Multiple 4 through 6 Multiple Cumulative Sample Size 8 5 20 5 10 13 26 50 100 2 4 6 8 10 12 14 5 10 15 20 25 30 35 20 40 60 80 100 120 140 92 Ac Re 1 2 0 1 0 1 0 2 1 2 0 2 1 2 0 2 1 2 * 2 * 2 0 2 0 3 1 3 1 3 2 3 * 2 * 2 0 2 0 3 1 3 1 3 2 3 * 2 * 2 0 2 0 3 1 3 1 3 2 3 Table 5.14b: Sampling Plans from 3 to 6 sigma level of production quality Sigma Level 3 4 through 6 4 through 6 3 4 through 6 4 through 6 3 Sampling Plan SkSP-2 f=\l2 i=2 SkSP-2 / = l / 5 1 = 9 SkSP-2 /•=l/2 / = 1 5 ChSP-1 / = 1 ChSP-1 1 = 4 ChSP-1 1 = 4 MIL-STD-105E Level I (double sampling) Level n (double sampling) Level in (double sampling) 4 through 6 MIL-STD-105E Level I (single sampling) Level II (double sampling) 4 5 6 Ac 1 0 0 0,1 0,1 0,1 Re 2 1 1 1,2 1,2 1,2 20 40 50 100 80 160 2 6 5 12 7 18 5 7 9 13 11 19 20 50 100 80 160 automatically adjusted automatically adjusted automatically adjusted automatically adjusted Level ni (double sampling) 3 Cumulative Sample Size 8 5 20 5 5 19 DSM-HQ starting X from Stage 1 DSM-HQ starting X from Stage 1 DSM-HQ starting X from Stage 1 DSM-HQ starting X from Stage 1 93 0 1 0 2 1 2 0 3 3 4 automatically adjusted automatically adjusted automatically adjusted automatically adjusted CHAPTER VI SIMULATION RESULTS: ANALYSIS AND COMPARISON OF SAMPLING METHODS The simulation results, like the simulation design, are separated in two stages. Stage one fine-tunes the DSM-HQ, The paths to minimum sampling are analyzed and the most appropriate is selected. In addition, the appropriate prior parameters are selected. Furthermore, DSM-HQ is fine-tuned under random defective units type occurrences and random events combined with out-of-control events. Using the randomevent-only situation properties of DSM-HQ are determined. During the random event/out-of-control combination, the long-range sampling rate is determined for each cost combination. In Stage two the techniques are compared using random-type occurrences as well as out-of-control events added to the random occurrences. The recommended rates under each cost combination are used in Stage two in order to compare the DSM-HQ to the other techniques. Total weekly cost is the main variable to determine which technique is superior, although other variables are examined and compared, 6.1 Stage 1 of the Simulation results: Defining DSM-HO The DSM-HQ mathematical model indicates the starting sampling level and dynamically reacts to the sampling process. However, it does not indicate whether the model should first reduce the sample size and then reduce the sampling rate, reduce the 94 sample size while reducing the sampling rate simultaneously, or reducing the sampling rate first and then reduce the sample size. In addition, the mathematical model does not indicate what the starting prior parameter values should be neither for cost-efficiency nor practical purposes. This section investigates these issues as well as other properties of the DSM-HQ. 6.1.1 Prior Distribution Parameters Although the DSM-HQ mathematical model can adjust its sampling parameters to approach the actual production sigma-level as sampling information gets accumulated, it does not indicate the appropriate prior levels of sampling history to begin the sampling process. The number of previous observations, p, and the number of previous defects, a, are examined and analj'zed. The simulation results show that the lowest total cost of sampling is obtained if the prior parameters match the process sigma quality level. Therefore, if the historical or prior defective rate, X, matches the process sigma rate, the lowest total cost of sampling will result. Furthermore, if X matches the process sigma level, the long-term sampling rate will be the same whether a and p have high or low values. For example, if the process sigma level is 4-sigma (0.621 % defective),/c = 50, vc = 2, c/= 10000 and the prior parameters are either a = 1 and p = 161 or a = 6210 and p = 1,000,000 the long-run sampling rate will approximate the same intercept value of 5.87 (or 6 units sampled per lot). In Figure 6.1 it can be seen that the combination with less history (a = 1 and p = 161) hovers around the intercept value. After a few weeks the value starts settling as 95 more information is gathered and the value approximates the intercept value. On the other hand, the value with a higher level of history (a = 6210 and p = 1000000) always stays very close to the intercept value of 5,87. Therefore, if the first few days is simulated thousands of times the prior parameters with most history will have a slightly lower total cost than the prior parameters with less history if the prior X is matched to the process sigma level. However, after a few weeks the weekly-total cost in both instances become the same as information gets accumulated and the sampling rate approximates the intercept value in both cases. 4-sigma process with different alpha/beta combinations ,N CO o Q, E (0 Q. 0> 2 B c 14 12 10 8 6 4 2 0 -2 d alpha = 1. beta= 161 alpha = 6210. beta = 1000000 Aj/^g'^fvJ^' -66- "20- -80 Days Figure 6.1: Intercepts fi'om a 4-sigma process with same X but different a/p combinations If the value of the sigma quality level is not matched to the prior sigma level and one has to choose among several a/p combinations of the same X value, the one with the lowest history, P, will always yield better results. For example, if the process sigma level is 5-sigma and a 3-sigma X prior value is chosen, the a/p values range from 1/15 to 96 66807/1000000 (see Table 5.3). The smallest p in this range is equal to 15 and can more quickly adjust to the true sigma process level. Figure 6.2 shows the adjustment of two different prior 3-sigma values to a 5-sigma process. Both prior values have a X, = 0.667, which is approximately equal to a 3-sigma quality process. However, the process with the lower a/p combination, which is a = 1 and P = 15, adjust within 3 days to a 5-sigma process, while the higher a/p combinafion of a = 2 and p = 30, takes 18 days to adjust to a 5-sigma process. If a is increased to 16 and P to 240 it would take about 152 days for DSM-HQ to adjust to a 5-sigma process. 3-sigma prior values to 5-sigma process alpha = 1, beta = 15 10 alpha = 2, beta = 30 15 20 Days Figure 6.2: Two 3-sigma prior values adjusting at different rates to a 5-sigma process Therefore, the higher the prior history, p, the longer it takes DSM-HQ to adjust to the true process sigma level if the prior information is not matched to the true process sigma level. Figure 6.3 shows the average weekly cost over a 20-year period of a 6- 97 sigma process level with different prior values. The first two prior values of 1/15, 66807/1000000 are 3-sigma while the others are 6-sigma prior combinations of X,. 6-sigma process with 3 and 6-sigma priors / \ /\ 1 nnn \J lUUU / I" 800 0 / DUU \ y 5• ^ AC\C\ *fUU w o 1- \ \ onn 1 0- 1 1 s<^ •^'^ * SJ 4 ^^ ^^ • SS ^ * 1 ^ 4 A^<^ •f ^ Starting Lambda Figure 6.3: Six-sigma process with 3 and 6-sigma priors The 1/15 combination produces over 20 years a weekly total sampling cost of $373.20, while the 66807/1000000 has a weekly total sampling cost of $1134.00. The matched X values of 6-sigma all produce a weekly total sampling cost of $334.70. Figure 6.3 illustrates that if the researcher is not going to precisely match the prior sigma level to the actual production quality sigma level, given a starting sigma level (in this case 3sigma) he/she should choose a X with a lowest a and P to give the system the opportunity to quickly adjust to the correct quality level. If the value of the sigma quality level has not been pre-determined and the researcher is unsure of the true sigma process level, then it is better to underestimate than to overestimate the quality level. This will allow the process to adjust more quickly to 98 the true sigma level. For example, if the researcher believes that the true process sigma level is between 4.5 and 5.5 sigma it is better to underestimate quality and choose 4.5sigma with as little prior history as possible. For prior values that can result in a 5.5 and 4.5-sigma the smallest a and p values are a = 1 and p = 31250 for 5.5-sigma and a=l and p=741 for 4.5-sigma (see Table 5.3). The lower of the two a and p values are a = 1 and p = 741. This combination of prior levels will give the best long-run weekly costs when X is not matched to the true process level. Especially at higher sigma quality levels, overestimating quality causes the system to sample at a very low rate making the increase in a less likely and harder for the system to adjust to the true quality level. Figure 6.4 shows that at a true quahty process of 5-sigma and/c = 5, vc = 2 and cf = 50000 the underestimated quality level at 4.5-sigma (a/p combination of 1/741) adjusts much quicker than at 5.5-sigma (a/p combination of 1/31250) prior values. The 4.5-sigma prior value begins approximating the 5-sigma value after 75 days while the 5.5-sigma value is still overestimating quality after 1000 days. Therefore, if the researcher is unsure of the true process level it is better to underestimate quality than to overestimate quality. 99 Underestimating vs.Overestimating prior values alpha = 1. beta =741 alpha = 1, beta = 31250! 5 sigma 1200 Days Figure 6.4: Effects of underestimating and overestimating prior values A property of the DSM-HQ worth discussing is the effect of increasing cf As seen in Figure 6.4 even the underestimated prior value of 4.5-sigma took a long time to approximate the true 5-sigma process quality level. This is due to a higher c/value, which causes the sampling rate to be reduced at a much slower rate. Figure 6.5 shows the effects of decreasing c/while keeping/c and vc constant. 100 6-sigma process with 3-sigma priors under different cf values 200 B 2 150 o> S cf= 1000 cf= 10000 100 cf= 50000 Q. ^ to 50 0.1 100 1000 Figure 6,5: Effects of reducing cf while keeping^c and vc constant In this illustration vc = 2 and^c = 5, while varying cf The chart with a logarithmic A:-axis scale shows that as c/'increases, the starting sample size also increases and the sampling rate decreases at a much slower rate. The total days it takes to reach a one per week sampling rate is 466 days for cf= 50,000, 93 days for cf= 10,000, and 3 days for cf= 1,000. If on the other handle and cfaro held constant while vc is increased the effect is different from the above example. Figure 6.6 shows a graph were cf= 10,000 and/c = 5, while varying vc. As vc is increased the starting sampling rate is decreased (opposite reaction to the previous example), but the amount of time to get to a one per week sampling rate is increased. When vc = 2, 20, and 100, the corresponding initial sampling rates are 62,15 and 5 units per lot respectively and the days needed to get to a one per week sampling rate are 98, 614, and 1134 days respectively. 101 6-sigma process with 3-sigma priors under different vc 70 « 60 vc = 2 vc = 20 vc= 100 o 40 g- 30 ^n 20 (0 10 0.1 10 100 1000 10000 Days Figure 6.6: Effects of increasing vc while keeping/c and c/'constant Finally, changing^c while keeping vc and c/'constant has different effects than the previous two examples. Figure 6.7 shows that holding vc = 2 and cf= 10,000 and increasing/c from 5 to 200 reduces the starting sampling rate (similar to the vc example and opposite to the fc example) from 62 units per lot to 24 units per lot. Furthermore, increasing/c reduces the amount of time to reach one lot per week (opposite effect to both the vc and the/c examples) from 98 days to less than one day. 102 6-sigma process with 3-sigma priors under different fc 70 « 60 N 50 0 40 fc = 5 fc = 50 o- 30 1 20 " 10 fc = 200 0.1 10 mm 100 Days Figure 6.7: Effects of increasing/c while keeping vc and c/'constant Table 6.1: Effects of changing each of the cost variables Increasing c/"while maintaining^c Starting sample size Days to one per week Increased Increased and vc constant Increasing vc while maintaining/c Decreased Increased and c/'constant Increasingyb while maintaining vc Decreased Decreased and c/'constant Table 6.1 summarizes the effects of changing each of the cost variables. This implies that a low cf combined with a high vc and high/c will have a very low starting sample size. Also, a high cf combined with a low vc and low fc will have a very high 103 starting sample size. More importantly, a high c/combined with a high vc and a low/c will result in a very slowly decreasing function. In this situation it in particular it is important to closely match the prior X value to the true process sigma level. For example, if a = 1 and P = 15 (3-sigma priors) and the true process sigma is 6-sigma where cf= 50,000, vc = 100 and/c = 5, the starting sample rate is 5 units per lot, but to arrive at a one unit per week sampUng rate it would take over 5,800 days. Conversely, if a low cf is combined with a low vc and a high/c the result is a very quickly decreasing function. For example, if c/= 1,000, vc = 2 and/c = 50, the starting sampling rate is 3 units per lot but it takes less then one day to reduce sampling to one sample per week. 6.1.2 Long-term sampling values As discussed above, the lowest total cost will result when prior values with more history are matched to the actual process sigma value. In this situation the sampling number will be very close to the intercept value throughout the process (assuming that the process is constant and the actual sigma level does not change over time). The following tables illustrate the long-term sampling values for 3, 4, 5 and 6-sigma suggested by DSMHQ. 104 Table 6.2: Long-term sampling values for a 3-sigma process cf= 1000 c/= 10,000 c/= 50,000 / c = 5, vc = 2 10/lot 10/lot 10/lot fc = 5, vc = 20 3/lot 10/lot 10/lot /c = 5,vc=100 1 per 2 lots 7/lot 10/lot fc = 50, vc = 2 7/lot 10/lot 10/lot fc = 50, vc = 20 1 per 2 lots 10/lot 10/lot /c = 50,vc=100 1 per week 6/lot 10/lot fc = 200, vc = 2 1 per week 10/lot 10/lot fc = 200, vc = 20 1 per week 10/lot 10/lot /c = 200,vc=100 1 per week 5/lot 10/lot Table 6.2 shows that at a 3-sigma process the long-term sampling rates vary from 1 sample per week all the way to 10 samples per lot. As c/increases, while keeping/c and vc constant, the number of samples per lot increase. This makes sense since it is worthwhile to increase the number of samples as the penahy for finding defectives in the field gets more costly. Also it can be noted that as vc increases, while keeping c/and/c constant, the number of defects tested is reduced. Similarly, as/c increases, while keeping c/and vc constant, the number of defects tested is also reduced. Since both vc and/c are in-house costs, as these in-house costs increase, it becomes more cost effective to reduce the number of units tested (if c/is held constant). 105 Table 6.3: Long-term sampling values for a 4-sigma process cf= 1000 c/= 10,000 cf= 50,000 / c = 5, vc = 2 1/lot 10/lot 10/lot fc = 5, vc = 20 1 per 2 days 3/lot 10/lot /c = 5,vc=100 1 per week 1/lot 4/lot fc = 50, vc = 2 1 per week 6/lot 10/lot fc = 50, vc = 20 1 per week 1/lot 10/lot /c = 50,vc=100 1 per week 1/lot 3/lot fc = 200, vc = 2 1 per week 1 per week 10/lot fc = 200, vc = 20 1 per week 1 per week 6/lot /c = 200,vc=100 1 per week 1 per week 2/lot Table 6.3 illustrates that for a 4-sigma quality process and a low c/'the long-term sampling process is for the most part one sample per week. This is because with a higher quality level than 3-sigma, it is more cost effective to let the low cf cosXs accumulate than to spend money sampling (except for very low/: and vc). Most of the sampUng rates at the 4-sigma level are lower than the ones at the 3-sigma level. There are still some sampling rates at the 4-sigma level that are at 10 units per lot in the cases where the inhouse costs {fc and vc) are low and c/is very high. At the 5-sigma quality level the long-term sampling rates are for the most part reduced to one sample per week. Table 6.4 shows that only in three instances this is not 106 the case. At the lowest/c and highest cf the long-term suggested sampling rates vary from 4 samples per lot to 1 per day. Table 6.4: Long-term sampling values for a 5-sigma process c/= 1000 c/= 10,000 cf= 50,000 /c = 5, vc = 2 1 per week 1 per week 4/lot fc = 5, vc = 20 1 per week 1 per week 1 per 2 lots / c = 5, vc=100 1 per week 1 per week 1 per day fc = 50, vc = 2 1 per week 1 per week 1 per week fc = 50, vc = 20 1 per week 1 per week 1 per week /c = 50,vc=100 1 per week 1 per week 1 per week fc = 200, vc = 2 1 per week 1 per week 1 per week fc = 200, vc = 20 1 per week 1 per week 1 per week / c = 200,vc=100 1 per week 1 per week 1 per week Finally, at the 6-sigma level, in the long-term suggested sampling rates all costs combinations can be reduced to destructively sample one unit per week. Table 6.5 illustrates that even at the lowest/c and vc in combination with the highest c/the samphng rate will eventually go to one sampled unit per week. 107 Table 6.5: Long-term sampling values for a 6-sigma process c/= 1000 c/= 10,000 c/= 50,000 /c = 5, vc = 2 1 per week 1 per week 1 per week fc = 5, vc = 20 1 per week 1 per week 1 per week /c = 5,vc=100 1 per week 1 per week 1 per week fc = 50, vc = 2 1 per week 1 per week 1 per week /c = 50,vc = 20 1 per week 1 per week 1 per week /c = 50,vc=100 1 per week 1 per week 1 per week fc = 200, vc = 2 1 per week 1 per week 1 per week fc = 200, vc = 20 1 per week 1 per week 1 per week /c = 200,vc=100 1 per week 1 per week 1 per week 6.1.3 Paths to minimum sampling The 3 paths to minimimi sampling examined are discussed in Section 5.1.3. To determine the path to minimum sampling, the starting parameter values (a and p) should underestimate the actual quality sigma level. This means that if the actual quality process level is at 6-sigma then the prior X level should be at either 5.5-sigma or lower. If the prior X level is matched to the actual process level, then the starting sampling rate is at or approximately close to the cost-efficient sampling rate. To determine the path to minimum sampling the prior X rate is set to 3-sigma, while the while the actual process quality level is set to 6-sigma. This allows the 108 sampling rate to steadily decrease and eventually arrive to the cost-efficient sampling rate while demonstrating the most effective path. The simulation results show that path type 1 is the most efficient path to minimum sampling. Although all 3 paths lead to the eventual long-term cost-efficient sampling rate, path type 1 does it in a more efficient manner. To illustrate this conclusion, let's take for example/c = 5, vc = 2 and c/= 10,000. The starting sampling rate is at 62 units per lot. In our case our maximum allowable sampling rate is 10 units per lot, but 62 is allowed in this situation to illustrate the findings. If the sampUng rate is reduced from 62 per lot to 1 per lot, and then from one per lot to 1 per week, the number of days needed to achieve minimum sampling is 93 days and 1962 observations (see Figure 6.8). Path Type 1: Reducing sample size then increasing sampling rate 70 0) 60 N 50 0) a> 4U Q. 30 E 20 CO 10 93 days 1962 observations Path Type 1 I 0 IIIMHI«*>**>I» > » > > > , 10 20 30 40 50 Lots (for sampling rate) Figure 6.8: Path type 1 to minimum sampling using for/c = 5, vc = 2 and cf= 10,000 109 On the other hand, if the sampling rate is first reduced from 62 per lot to 62 per week and then reduced to 1 per week, the number of days needed to achieve minimum sampling is 354 days with 1962 observations (see Figure 6.9). Path Type 2: Increasing sampling rate then reducing sample size 70 1 a, 60- • • i i • i i 4 4 # 4 4 A ^—«—« 4 A # ^ / /^ i > ;: ;; fl, 40 - "5. 30 ^ 20 «^ 100(D 1 1 1 10 20 30 i / 1/ 1^ / 354 days 1962 observations —•— Path Type 2 T 40 50 Lots (for sampling rate) Figure 6.9: Path type 2 to minimum sampling using for^c = 5, vc = 2 and cf= 10,000 The number of observations remains the same, while the number of days needed to achieve minimum sampling is much longer with path type 2 than with path type 1. Using path type 3 yields a result in-between path type 1 and path type 2 (see Figure 6.10). 110 Path Type 3: Increasing sampling rate and reducing sample size Lots (for sampling rate) Figure 6.10: Path type 3 to minimum sampling using for fc = 5, vc = 2 and cf= 10,000 hi this situation, the number of observations needed is still 1962. However, the number of days comes out to 184 days, which is in between the 93 days and 354 days needed for path type 1 and path type 2 respectively. hi all cases path types 1, 2 and 3 need the same number of observations to arrive at minimum sampling. Also, in every case path type 1 produces the most efficient path since it yields the same results in a minimum amount of time when underestimating the quality sigma level. 6.1.4 Lower Sigma-Level Results The simulation results show that at lower sigma levels the average total sampling cost per week is astronomical. The reason is that the total weekly cost attributed to finding a defect in the field (c/) is extremely high at the lower sigma values. For 111 example, at a 3-sigma quality level, 6.68% defective are getting through to the customer. This means that in the study's situation at 40,000 units per week production (1000 units per lot, 8 lots per day and 5 lots per week) on average 2,672 units per week are finding their way to the final customer. Even at the lowest cfper unit considered in this study of c/= $1000, the average weekly cost attributed to c/would be $2,672,000 which would put any manufacturing/assembly company out of business. With/c=5, Vc=2, and cy=1000 the average weekly sampling cost is $2,696,768 for a 3-sigma quality production process. For a 4-sigma quality process the sampling costs are still very high at the lowest cf In this situation, the number of weekly defects finding their way to the customer is on average 248 (40,000 at 0.6210% defective) and therefore the total cost attributed to cfper week at cf= 1000 on average is $248,000. With/c=5, Vc=2, and c/=1000 the average total sampling cost for a 4-sigma quality production process is $248,691. These costs are even higher when considering higher cffc and vc values. This indicates that modem manufacturing/assembly companies, which commonly and contractually incur a stiff penalty from their customers when defective is foxmd at the cUent's site caimot survive for very long under these lower quality standards. Since these numbers are ridiculously high (and are even higher under stiffer cf costs) only comparisons at the high quality levels of 5 and 6 sigma are considered. From this point on the 3-sigma and 4-sigma values are only used when taking a conservative prior value X estimates (when underestimating quality). 112 6.1.5 Sampling rates with an out-of-control element added When out-of-control events (OC) are added to random defectives the costefficient sampling rates change in almost every cost situation. Because it is assumed that all out-of-control events are detected and the out-of-control situation does not last forever, DSM-HQ's X rate changes to reflect a lower sigma level than the actual process sigma value, hi the random-event-only (RE) case, the number of defectives encountered in samphng approximates the number of defectives produced since in the long-run there is approximately the same chance of detecting a defective than there is of producing a defective. This changes when an out-of-control element is added to the process since all out-of-control events are detected. This means that after information is gathered DSMHQ will reflect a lower quality process than the quality of the actual process resulting in most cases in a higher sampling rate. Table 6.5 shows the DSM-HQ suggested sampling rates for a 5-sigma process with random events combined with out-of-control events assuming the chance of getting a random event and an out-of-control event are the same. Table 6.6 shows that adding out-of-control situations to the random events increase the sampling rate. Comparing Table 6.6 to the random-event-only table (Table 6.4) it can be seen that for cf= 1000, the numbers have changed from a long-term rate of one sample taken per week to five samples taken per lot. For cf= 10,000 only where vc = 2 has the rate remained at one sample tested per week. Everywhere else it has increased to either 5 or 6 samples per lot. Where cf= 50,000, fc = 5, and vc = 2 is the only instance where the sampling rate has decreased. The reason is that with the randomevent-only case the costs outside the company are balanced with the costs incurred inside 113 the company. As the out-of-control events are added, only where vc = 2 is the re-work cost low enough to allow the system to go to a one per week sampling rate. Table 6.6: Sampling rates for a 5-sigma process with OC and RE combined. cf= 1000 cf= 10,000 cf= 50,000 fc = 5,vc = 2 5/lot 1/week 1/week / c = 5,vc = 20 5/lot 5/lot 5/lot /c = 5,vc=100 5/lot 6/lot 6/lot fc = 50, vc = 2 5/lot 1/week 1/week fc = 50, vc = 20 5/lot 5/lot 3/lot fc = 50, vc = 100 5/lot 5/lot 6/lot fc = 200, vc = 2 5/lot 1/week 1/week fc = 200, vc = 20 5/lot 6/lot 3/lot fc = 200, vc = 100 5/lot 6/lot 6/lot Table 6.7 illustrates the DSM-HQ suggested sampling rates for a 6-sigma process with out-of-control events and random events having the same chance of occurrence. When adding an out-of-control element to the random-event-only case at the 6-sigma quality level all of the sampling rates for every cost combination are increased (see Tables 6.5 and 6.7). 114 Table 6.7: Sampling rates for a 6-sigma process with OC and RE combined. c/= 1000 c/= 10,000 c/= 50,000 fc = 5,vc = 2 1/lot 1 per 2 lots 1 per 2 lots fc = 5, vc = 20 1/lot 1/lot 1/lot fc = 5, vc = 100 1/lot 1/lot 1/lot fc = 50, vc = 2 1/lot 1 per 2 lots 1 per 2 lots /c = 50,vc = 20 1/lot 1/lot 1/lot /c = 50,vc-100 1/lot 1/lot 1/lot fc = 200, vc = 2 1/lot 1/lot 1/lot fc = 200, vc = 20 1/lot 1/lot 1/lot /c = 200,vc=100 1/lot 1/lot 1/lot This particular situation is of interest since all manufacturing/assembly companies attempt to operate at the 6-sigma quality level and have production processes with out-ofcontrol elements in addition to the random events of defectives. However, the percentage of out-of-control to random events varies from company to company and from product to product. For simplicity this study considers a situation where there is the same likelihood that an out-of-control event and a random event will occur. Furthermore, for all of the cost values and for this particular lot size the suggested sampling rate comes out to either one unit or two units sampled per lot. However, if the lot sizes or/c, vc, c/were to change, the suggested sampling rates would also change. 115 6.2 Stage 2 of the Simulation results: comparison of sampling methods All sampling techniques are compared and contrasted. The advantages and disadvantages of each technique under differentyc, vc, and cfare discussed. The comparison of sampling methods is separated into random events only (RE) and random events with out-out-of control situations combined (REOC). 6.2.1 Random event comparison of sampling methods DSM-HQ is compared to single/classical sampling, double sampling, multiple sampling. Skip-lot sampling. Chain sampling and MIL-STD-105E. The comparison is made with different costs (fc, vc, cf) and under 5 and 6-sigma quality production levels. The principal measuring variable which decides which is the best sampling technique is weekly total cost. However, other variables are also observed. These include: • Total weekly cost (TC/week) • Number of total defectives created by the simulation (defectives) • Number of defectives detected by each of the sampling techniques (detected) • Percentage of total cost attributed to c/to total cost (TCF%) • Percentage of total sampling cost to total cost (TSC%) • Total samples taken during the simulation (totalsamp) • Average weekly sample per method (samp/week) Appendix B and Appendix C illustrate the output for 5-sigma and 6-sigma respectively. Table 6.8 illustrates the methods and its sampling plans used in Stage 2. 116 Table 6.8: Methods and Sampling plans used for comparison. Method Sampling Plan DSMHl DSM-HQ underestimating the true quahty process and beginning at the 3sigma level DSM-HQ beginning the sampling at the cost-efficient sample size and prior level Single sampling with a sample size of 5 and^c = 0 and i?e = 1. Single sampling with a sample size of 20 and ylc = 0 and i?c = 1. Double sampling with cumulative sample sizes of 13,26 and Ac = 0,1 and Re = 2,2 Double sampling with cumulative sample sizes of 50,100 and Ac = 0,1 and Re = 2,2 Multiple-sampling with cumulative sample sizes of 5,10,15,20,25,30,35 and ^c = *,*,0,0,1,1,2 and Re = 2,2,2,3,3,3,3 Multiple-sampling with cumulative sample sizes of 20,40,80,100,120,140, and^c = *,*,0,0,1,1,2 mdRe = 2,2,2,3,3,3,3 SkSP-2 with/= 1/5 and i = 9 and a sample size of 5 with ^c=0 and Re = 1 SkSP-2 with/= 1/2 and z = 15 and a sample size of 20 with ^c=0 and Re = 1. ChSP-1 with i = 4 and a sample size of 5 with ^ c = 0,1 and Re=l,2 ChSP-1 with i = 4 and a sample size of 19 with^c = 0,1 and Re=l,2 MIL-STD-105E with Level I (single sampling) sample size of 20, Ac = 0, Re = 1, Level II (double sampling) cumulative sample size of 50,100 and^c = 0,1, Re = 2,2, and Level III (double sampling) cumulative sample size of 80,160 and Ac = 0,3, Re = 3,4. DSMH2 SINGl SING2 DOUBl D0UB2 MULTl MULT2 SkSPl SkSP2 ChSPl ChSP2 MILST In each case the principal method of interest is DSMH2, which provides the costefficient values for the DSM-HQ. DSMHl is also of particular interest since it is an underestimation of the initial process level eventually adjusting to cost-efficient results. In the 5-sigma quality simulation illustrated in Appendix B, DSMH2 has the lowest total weekly cost in 17 out of the 27 cases, while DSMHl is lowest in 2 cases, MULT2 is lowest in 5 cases, D0UB2 is lowest in one case and SkSPl is lowest in 2 cases. In both cases where DSMHl was lowest (fc = 200, vc = 2 cf= 1000 and/c = 200, vc = 20 and cf 117 = 1000) DSMH2 came in a very close second place indicating that the adjustment of DSMHl from 3-sigma to 5-sigma was done very rapidly and that the differences in cost is negligible. Of the remaining 8 cases where DSMH2 does not have the lowest total weekly cost c/is equal to 10,000 or 50,000. This indicates that either DSM-HQ is not the most cost-efficient method in every case or that more production into the fiiture is needed to determine if DSM-HQ is the best sampling approach under each condition. As the sigma level quality is improved a pattem starts developing indicating the methods that are better suited for high quality testing. Appendix C shows that under 6sigma quahty DSMH2 is the best method in 26 out of 27 cost combinations. The exception comes a t / : = 200, vc = 20 and cf= 1000, where DSMHl is better by less than $2. Again here the difference is negligible since in this situation DSMHl immediately adjust itself to a cost-efficient samphng rate and is essentially sampling at the same rate as DSMH2 and therefore the difference is due to chance. Out of the existing methods SkSPl performs at the highest level occupying the second position 7 out of 27 times and never falling worse than third. One limitation of DSM-HQ is the percent detected during sampling. Overall, DSMH2 came in last place, while SkSPl and DSMHl also performed poorly in this area tied at second to last place. Figure 6.11 illustrates the percentage of defectives detected by method. As expected the methods that have the highest number of samples per lot detected the highest percentage of defectives, while the methods with the lowest number of samples per lot detected the lowest percentage of defectives. 118 Percent defectives detected by method, 6-sigma level (RE only) Percent detected 8.0% 6.2% 5.3% O) a> 4.0% a 0 0. ).5% 0.7% 0.1%0.0% 0.0% T- CM •- CM T- C^ I— Q. CL OQ m X X CO C 0 a ) D D 2 2 _ j _ ^ ^ ; F = ; f = O O C 0 C 0 ^ ^ ^ O O Q - Method Figure 6.11: Percent defectives detected by method at the 6-sigma level of quality Figure 6.11 also shows that at the 6-sigma level-RE, out of the more than 35,000 defectives created by the simulation the method with the highest percent detected, MULT2, only could detect 220 defectives. This also means that the method with the highest sampling rate allowed 93.8% of the defectives to go undetected. The total cost is composed of the total cost of sampling (TCS) and the total cost of field (TCF). The methods that tended to do better at the 6-sigma level with random-event only (RE) situations are the ones with the highest TCF percentages. Figure 6.12 shows that DSMHl, DSMH2 and SkSPl have the highest TCF percentages indicating that these methods allow more defects to get to the final customer than the other method. This is due to the limited number of samples taken per lot. This shows that at very high quality levels in RE situations it is more cost effective to minimize sampling. 119 Percent of Total Cost 1.20 1.00 •£ 0.80 S 0.60 I 0.40 -H 0.20 m TCS% • TCF% ^ CM Q. a. CM T- CM I— T- CM 1- CM T- CM m o Q X X W H h o O Q - Q - Method Figure 6.12: Percent of Total Cost divided into TCS and TCF at the 6-sigma RE 6.2.2 Random event with out-of-control comparison of sampling methods Again DSM-HQ is compared with all discussed sampling techniques. However, an out-of-control element is added to the production process (REOC). The principal variable used for comparison is total weekly cost. The variables used for comparison in random-events only type production are also used when an out-of-control situation is added to the simulation. In addition, the average time to detect an out-of-control event in hours (ochrs) is also used for comparison. Appendix D and Appendix E show the results of the comparisons made at 5-sigma and 6-sigma. The out-of-control added to the random event (REOC) situations is more realistic than the random-event only (RE) situation since a percentage of the time the process produces an out-of-control event compared to a random event. In this situation, 50% of the defectives are random events while the other 50% are out-of-control events. 120 As expected, Appendix D shows that in every case where DSMH2 suggests a sampling rate of 5 samples per lot (see Table 6.6) the total weekly cost is almost identical to SINGl and ChSPl, This is due to the fact that both SINGl and ChSPl also use a 5samples per lot sampling plan. Figure 6.12 shows an example where/c = 50, vc = 20 and cf= 10,000 and the sampling number for DSM-HQ is 5 samples per lot. As expected, Figure 6.12 illustrates that DSMHl, ChSPl and SINGl have very similar total weekly costs since all have a 5-samples per lot sampling plan. Total Weekly Cost for fc = 50, vc = 20, cf = 10000, at 5-sigma REOC #9 1 on nnn nn >• •r— «n nnn nn \J\) ^\J\j\j .\j\j % 60.000.00 - 5 40.000.00 - S 20.000.00 - TCperWeek C N f r - i - ^ i - T - C M C M C g C M l — C M ^ X Q - O X m H c L O D - m c o h Q 5 ( 0 Z 2 ^ = ^ W Z C 0 Z ) : J = d C 0 Sampling Method Figure 6.13: Total Weekly Cost where DSM-HQ suggests a sampling rate of 5 per lot Where DSMH2 can differentiate itself is in instances where the sampling is significantly different than 5 samples per lot. For example, in the cases where vc = 2 and cf>= 10,000 where the sampling is one per week there is a considerable difference in sampling rates. In these situations, DSM-HQ has a potential for a considerable advantage over the other methods since there is such a pronounced difference in sampling rates. 121 Figure 6.13 shows an instance where/; = 50, vc = 2, and cf= 10,000, and where the samphng plans are quite different allowing DSMH2 to differentiate itselffi-omthe other methods and providing much lower weekly costs than the traditional methods. Total Weekly Cost for fc = 50, vc = 2, cf = 10000, at 5-sigma REOC 52,000. 50,000. 48,000. o 46,000. 0) 44,000. 42,000. o 40,000. 38,000, n o o ChS ChS SINi SkS DSM SIN TCperWeek T- CN OQ m Sampling Method Figure 6.14: Total Weekly Cost where DSM-HQ suggests a sampling rate of 1 per week The only case where DSMH2 did not perform near the top is where/: = 5, vc = 2 and cf= 50,000. The reason might be that the simulation needs to be run a few more years into the fiiture either to obtain a better cost-efficient value in Stage 1 or to allow the best method to rise to the top in Stage 2. Using both REOC and a quality production process at the 6-sigma level, DSMH2, DSMHl and SkSPl consistentiy occupied the top 3 positions (with few exceptions). The only times DSMHl outperformed by DSMH2 was in the cases where DSMHl improved to a cost-efficient sampling rate very quickly and the difference between the two was neghgible. Therefore, without considering DSMHl, DSMH2 outperformed each and every one of the traditional sampling methods. Figure 6.14 shows the overall total 122 weekly cost by method at the 6-sigma level of quality considering both out-of-control and random events. Figure 6.14 shows that DSMH2 outperformed the nearest competitor, SkSPl, by 35% and the popular destructive sampling technique SINGl by 76%. 18,375 • TCperWeek • 1 10,609 38,081 ^ 13,864 i 29,962 •i BH^HH 102,949 39,460 •• 1 7,858 1 8,141 86,802 ^mm m 26,747 36,466 IB 13,864 i O o Total Weekly Cost rail Total Weekly Cost by Method, 6-sigma REOC ^ C M T - C M ^ C N t - ; ' - C N T - ( M T - C M D . Q . C D 0 Q I X ( 0 h h C 3 O Q . Q . Method Figure 6.15: Overall Total Weekly Cost at 6-sigma REOC At the 6-sigma level considering REOC the average number of hours that the process is out-of-control before it is detected influences total cost. Figure 6.15 illustrates the average amount of time (in minutes) that each method takes to detect an out-ofcontrol event. 123 c Out-of-control time in minutes per method, 6-•sigma REOC 00 ai H SkSPl in CO H • fflOC minutes pi •• 42.8 DSMH2 m SkSP2 pi p138.0 • 17.0 MULTl DSMH1 • ChSP2 • 20.8 MILST • ChSPl I SING2 • 20.7 CO (d SING1 <q (d i r^ ir> D0UB1 • 17.4 in (d MULT2 0 (o in D0UB2 p Minutes in Method Figure 6.16: Out-of-control time for each method in minutes at the 6-sigma level In this situation, SkSPl and SkSP2 have the highest amount of time since it skips lots with a high quality process. The methods that sample most often, D0UB2 and MULT2 tend to take a lower amount of time to detect an out-of-control situation. DSMH2 takes on average slightly under 43 minutes to detect an out of control situation. This indicates that in the REOC case the most cost-efficient amount of time to detect an out of control is neither with the high sampHng nor with the low-sampling techniques. 6.3 Summary The simulation results show that DSM-HQ is slightly superior under 5-sigma RE. As the quality improves to 6-sigma in the RE case, DSM-HQ significantly outperforms the traditional methods. Because the suggested sampling rates in both of these cases are reduced to either one per week in most cases and in few cases to 4 per lot, 1 per two lots 124 or one per day (Tables 6.4 and 6.5) it is not surprising that the closest competition comes fi-om SkSPl. In the REOC case under a 5-sigma quality level, the nearest competition comes fi"om SINGl and CHSPl which both have a 5 per lot sampling rate. In the many cases where the suggested sampling rate for DSM-HQ is 5 per lot the results are almost identical to these other two traditional methods. In the cases where DSM-HQ has a pronounced difference in sampling rate to these two methods the savings in total costs are very apparent. As the quality improves to 6-sigma in the REOC case, the differences in both sampling rate and total costs are more pronounced. These results are encouraging since DSM-HQ is developed specifically for high-quality processes and under both RE and REOC conditions the results provide a significant cost savings. 125 CHAPTER v n CONTRIBUTIONS AND SUMMARY 7.1 Contributions, Limitations and Future Research Current traditional sampling techniques are not designed to work under very highquality production processes. DSM-HQ is an economically-based sampling technique designed to minimize total cost when the process quality is very high. This technique outperformed the existing traditional techniques in every 6-sigma simulation for both RE and REOC and in most of 5-sigma simulations for both RE and REOC. However, these results only show performance under limited conditions. These include one production line, 1000-unit lots, and a fixed cf This research only considers a single production/assembly line. In many cases there are multiple manufacturing/assembly lines that contribute to the production of a lot. As the number of production lines increase then the suggested sampling rate might also increase. Additionally, the values generated in the cases where the out-of-control element is added are based on lots of 1000 units. Like with all other methods, sampling plans would have to be determined based on different lot sizes. Reducing the lot size might reduce the DSM-HQ suggested sampling rate, while increasing the lot size might increase the recommended sampling rate. Another limitation of DSM-HQ is that the sampling number is highly dependent on/c, vc, and cf This means that either the manufacturer/assembler needs to know what 126 each of the costs are for each one of the products or a more generic method of DSM-HQ would have to be developed. If for example, cf has several values, a weighted average of c/could be considered. Finally, in the REOC case the limitations include an out-of-control process where every unit that follows a defective unit is defective. Although this can actually happen, there are also cases where the process produces units at a lower sigma level. The limitations of this study will be addressed in fiiture research which include: 1. Widening the out-of control scope. In this research when a process goes out-of-control, each and every one of the following parts is considered to be defective. Future research would include an out-of-control process to be at, for example, a 3-sigma level and then observe the effects on the suggested sampling rate and total cost. 2. Changing the lot sizes. The cost and the sampling number of the traditional sampling techniques would change making a difference in total cost comparisons. 3. Considering more than one production line. Currently, only one production line is considered in this research. Future research includes increasing the number of production lines while allowing each line to be a part of the lot formulation. 4. Considering a weighted average of cf In the case where cf has several values for a particular product, a weighted average of cf can be used to arrive to the cost-efficient sampling rate. 127 5. Comparison vs. CUSUM control scheme and other methods. Although this research focuses on comparing DSM-HQ with the most popular and traditional sampling methods, there are other methods, which need to be considered in fiiture research. 6. Different percentage of REOC. This particular research is limited to a 50% random event and 50% out-of-control. If this number varies, then the suggested sampling number might also change. 7. Considering a compound-Poisson to model defects within the defective units. In the case where a unit can have multiple defects, a geometricPoisson distribution can be used to model the production process yielding different cost-efficient sampling numbers. 7.2 Summary DSM-HQ tends to perform better as the quality of the process improves. In the lower sigma levels, DSM-HQ performs very similar to single sampling. At these quality levels, DSM-HQ suggests a very high sampling rate and the upper limit only allows a predefined maximum sampling rate. Therefore, there are no significant differences to the traditional single sampling technique. As the quality improves, there are differences in some of the sampling rates under certain conditions. These differences in sampling rate become apparent at the 5-sigma level and are pronounced at the 6-sigma level of quality. At the 5-sigma level DSM-HQ suggests some similar sampling rates to the traditional techniques. When this is the case, there are no significant differences in total cost. 128 However, in the cases where there are pronounced differences, DSM-HQ provides significant savings. At the 6-sigma level of quality, all sampling rates are very different to the traditional sampling rates. In this high-quality environment, DSM-HQ consistently outperforms the traditional techniques resulting in significant cost savings. However, DSM-HQ does have some limitations. The study and comparison is based on a specified number of units per lot. In addition, unlike the traditional techniques, the sampling rate is based on the cost variables. The out-of-control rule is limited to a process going completely out-of-control with all parts being defective after the first part in the out-of-control process is defective. The REOC case is limited to random events having the same likelihood as out-of-control events. The study is also limited to a single production Une producing all units in the lot. Finally, other techniques, such as CUSUM, need to be compared and contrasted against DSM-HQ. These and other issues will be addressed in fiiture research. Although DSM-HQ has many limitations, the initial results are very encouraging. For manufacturing/assembly companies that have a situation similar to the one in this study, the cost savings can be very significant. Furthermore, if such a company has several different product lines, the cost savings are compounded. Future research in the areas of out-of-control pattems, and single vs. multiple production lines along witii more specific information on particular cases with regards to fc, vc, cf lot-size, and percent of random-events vs. out-of-control events are needed to provide better estimates on sampling rates and costs. The research planned for the fiiture should yield a cost-efficient 129 sampling rate specific to each particular case and, in high-quality environments, provide manufacturing/assembly companies significant cost savings. 130 REFERENCES Bartky, W. (1943). Multiple sampling with constant probability. The Annals of Mathematical Statistics, 14, 363-377. Berger, J. 0. (1985). Statistical decision theory: Foundations, concepts, and methods. {3^^ ed.). New York: Springer-Verlag. Becker, G, & Camarinopolous, L. (1990). A Bayesian estimation method for the failure rate of a possibly correct program. IEEE Transactions on Software Engineering, 16, 1307-1310. Case, K., & Keats, J. B. (1982). On the selection of a prior distribution in Bayesian acceptance samphng. Journal of Quality Control, 14, 10-18. Chulani, S., et al. (1999). Bayesian analysis of empirical software engineering cost models. IEEE Transactions on Software Engineering, 25, 573-583. Campodonico, S. & Singpurwalla, N. (1994). A Bayesian analysis of the logarithmicPoisson execution time model based on expert opinion and failure data. IEEE Transactions on Software Engineering, 20, 677-683 Dalai, S. R., & Mallows, C. L. (1988). When should we stop testing software? Journal of American Statistics Association, 83, 872-897. DeGroot, M. H. & Schervish, M. J. (2002). Probability and statistics (3'"* ed.). Boston, MA: Addison Wesley. Department of Defense (1989). Military standard: Sampling procedures and tables for inspection by attributes, Washington, DC: Department of Defense. Dodge, H. F. (1943). A sampling inspection plan for continuous production. The Annals of Mathematical Statistics, 19, 264-279. Dodge, H. F. (1951). Additional continuous sampling inspection plans. Industrial quality Control 7(5), 7-12 Dodge, H. F. (1955a). Skip-lot sampling plan. Industrial Quality Control, 11(5), 3-5. Dodge, H. F. (1955b). Chain sampling inspection plan. Industrial Quality Control, 11(4), 10-13 131 Dodge, H. F., & Romig, H. G. (1929). A method of sampling inspection. The Bell System Technical Journal, 8, 613-631. Dodge, H. F., & Romig, H. G. (1944). Sampling inspection tables - single and double sampling. New York: John Wiley & Sons, Inc. Dodge, H. F., & Torrey, M. N. (1951). Additional continuous sampling plans. Industrial Quality Control, 7(5) 5-9. Florens, J.P., et al. (1990) Elements of Bayesian statistics. New York: Marcel Dekker, Inc. Govindarajulu, Z. (1981). The sequential statistical analysis of hypothesis testing, point and interval estimation, and decision theory, Columbus, OH: American Sciences Press. Grant, E. L. & Leavenworth R. S. (1988). Statistical quality control (6^^ ed.). New York: McGraw-Hill. Iversen, G. (1984). Bayesian statistical inference, Newbury Park, CA: Sage Publication. Maritz, J. S. (1970). Empirical Bayes methods. London, England: Methuen and Co. Ltd. Nahmias, S. (1993). Production and operations analysis. New York: Richard D. Irwin Inc. Pabst, W.R., Jr. (1963). MIL-STD-105D. Industrial Quality Control, 20(5), 4-9. Pande, P., & Holpp, L. (2002). What is six sigma? New York: McGraw-Hill. Perry, R. L. (1973a). Skip-lot sampling plans. Journal of Quality Technology, 5(3), 123130. Perry, R. L. (1973b). Two-level skip-lot sampling plans - Operating Characteristic Properties. Journal of Quality Technology, 5(4), 160-166. Press, S. J. (1989). Bayesian statistics: Principles, models and applications. New York: Wiley. Randolph, P., & Sahinoglu, M. (1995). A stopping rule for a compound Poisson random variable. Applied Stochastic Models and Data Analysis, 11, 13 5-143. Ross, S (2000). Introduction to probability models {f^ ed.). London, England: Harcourt. 132 Soundararajan, V. (1978). Procedures and tables for construction and selection of chain sampling plans (ChSP-1). Journal of Quality Technology, 10, 56-60 and 99-103. Tennant, G. (2002). Design for six sigma: Launching new products and services without failure. Hampshire, England: Gower. Wald, A. (1973). Sequential analysis. New York: Dover Publications. 133 APPENDIX A SEQUENTIAL ANALYSIS UNDER HIGH QUALITY CONDITIONS This section analyzes the sequential analysis technique and more specifically, Wald's (1973) Sequential Probability Ratio Test (SPRT). From Chapter 2, the SPRT had the following steps. If: P.'"(l-A)""'" ^ P Po'-a-z'o)'""'''" (i-«) then the decision is to accept the lot. Otherwise, if: then the decision is made to reject the lot. Finally, if: {\-a) j9/'"(l-/7o )"-""• fi ^p/"(i-/,,)"-"" a j-p then, another observation is taken and a decision to accept, reject or take an additional observation is made. Let's first analyze the first formula. If we take the log of both sides, the result is: log^-. ^•._. ^log This means that: .„ l o g A , ^ I o g £ z ^ _ , „ l o g ^ , l o g - A ^ Po g-/'o) g-Po) g-«) Therefore: 134 d , l-« (l-p.) Po (1-Po) Because the lot is accepted whenever the right part of the equation is greater than dm, then the right part of the equation be considered the acceptance number or a, ^mTherefore, dm < Om^ and therefore: log^^-;«log(^-^-^ logA_,og ('-".) Po (1-Po) Similarly, for the second formula, if the log of both sides is taken, the result is: m-rf Which means that; <i.ioga,„,„g£z£il_rf.,„g£zAl^,„,izA Po (1-Po) (1-Po) ^ Therefore: ,ogl^-™,ogfl^ rf > — ^ "--P"' Po g-i^o) Because the lot is rejected whenever the right part of the equation is less than dm, then the right part of the equation be considered the rejection number or r^. Therefore, dm>r„. and: 135 log m ^-mlog ^' a g-Po) logA_iog(l-^.) Po g-Po) Finally, taking the log of the third equation yields the following equation: log—-— < log \ —^-^—— < log—— From the previous equations the following inequality can be deduced: , P log_^l \-a . (l-p,) m\og) ^ ' ; g-Po) ^ log^-log^ Po • (1-Po) , \-p . {\-p,) log—^-wlog\ a g-Po) logA-,„gilz£li Po (1-Po) This means that the researcher should keep taking observations as long as the number of defectives lies in between the two equations. Although these equations are not exactly the same as those in Wald (1973), they are mathematically equivalent. Let's analyze now the equation for the acceptance number: log-/-^log(^-^'> l-or g-Po) logA_iog(l-^.) Po g-Po) This implies that m, which is the amount of units inspected is: log-^-a.flogA-log^^ l-a 1^ Po g-Po); m= log g-Pi) g-Po) 136 It can be seen that for the SPRT, the amount of units inspected depend on the numerical values of a„, po, p,, a andy9. Taking the examplefromWald (1973), the values that he gives each one of the variables ispo = .l,p, = 3.a = .02 andyS = .03. If the researcher has no defectives sequentially, then a^^, = 0 and therefore, the minimum number of units inspected would have to be 14 for the lot to be accepted under this situation. The reason is the following: P , 0.03 l o g - ^ ^ = log = -1.5141 l-a 1-0.02 and logillM = log(lzM = _o.i091 g-Po) g-o.i) therefore, i-'^ m= I Po r, \ g-Po)> -1.5141-0 ^^—^-^^ = = 13.8725 « 14 logA-ilog^-log"-"''^ g-Po) This number agrees with the table provided by Wald (1973, pp. 93). This means that the researcher caimot accept the lot under these po, pj, a and ^ values, until the 14'^ unit is observed (if all 14 units are non-defective). This implies that the minimum number of observations that the researcher will test will be 14. Ifpo is allowed to go to 0, then the value of m would be reduced to 9.7746 - 10, reducing the number of observations. This implies that even if ;5 and a axe held constant and ifpo is equal to zero (which reduces the amount of observations taken under SPRT), then: 137 hmm = log- fi l-a - r,„gA_i„ga-.,)l a„ Po g-Po) i,gg-P,) g-Po) = 00 The numbers for m would increase even more if the consumer's risk {fi) and producer's risk (a) were to be reduced. The values given forpo, pi, a and ft in the previous example are now considered poor to average quality (Tennant 2002). Under "six sigma," the defectives per miUion opportunities are 3.4 (Pande and Holpp 2002). This implies that if the manufacturer/assembler wants to operate under "six sigma" it should allow for only one defective per 294,117 conforming items in order to be considered a "world class" manufacturer (Tennant, 2002). This in turn imphes that the values of po, pj, a and ft need to be adjusted for today's high quality standards. The consumer's risk ^ (the probability of accepting product of some undesirable quality) has to be adjusted since the producer does not want to risk shipping anything less than high quality products. In addition, recall that ifp >pj and the lot is accepted, then the error is regarded as an error of practical consequence. Therefore,/?/ needs to be adjusted since the error regarded as "practical consequence" needs to be shifted to a lower value. In the case of world-class manufacturing, let po equal the proportion for six sigma (3.4/1,000,000). Letpj equal 233 defectives per miUion, which is five sigma (Pande and Holpp, 2002). Although 5 sigma is only considered "excellent" quality and is not "world class" (Tennant, 2002), it will be useful for the purposes of the example. If all the other variables are allowed to stay the same andpj is adjusted to a five sigma level (py = 138 (233)7(1,000,000)), and;?^ to a six sigma level po = (3.4)7(1,000,000)) then the minimum amount of observed units before the lot could be accepted under the SPRT is: iog/-ajiog^-iogfl::^' ^ = l-a I Po " ,, , g-Pi) log g-Po) ^—-^^^ = -1.5141048-0 ; «15,182.679 -0.000099726 g-Po) This means that under these conditions, there would have to be 15,183 observations without a defective for the researcher to be able to accept the lot. Therefore, the SPRT is not suitable for today's high quality sampling. Similarly, the number of observations until rejection would also yield a similar outcome. Analyzing the formula: cc g-Po) log^L.logO-^.) m Po (1-Po) Then the number of observations is: log m= r„ a L Po g-Po)J log(^-^-> g-Po) Taking the values in the example oipo = .1. /?/ = .3, a = .02 and /9 = .03, then the results are: l o g i z l = l o g ^ = 1.6857417 a 0,01 and 139 iog£z£i)=,og(l::M = _o.io9i g-Po) g-o.i) and, l o g ^ = l o g — = 0.47712 Po 0.1 Therefore, logi^-Jlog^-logil^l a m= I g-Po)j 1.6857417 - r (0.47712 + 0.1091) Po log g - P i ) -0.1091 g-Po) which means that: /w =-15.4513+ r„ (5.3732) This implies that the researcher has to find a minimum of 4 defectives before it can reject the lot. The reason rm cannot be equal to 3, is because m would equal 0.6683 1. This is impossible since there cannot be 3 defective observations {rm) with only observation (m) taken. It also means that the researcher needs to wait until the 4^*" observation until he7she can decide to reject the lot. Again, these numbers are not suitable for "world class" manufacturing. Here again if the limit is taken on rm, aspo and PJ go to zero, then: log—^-mloga (l~Po) -7. lim r„ = — :;— - °o l o g - L , log Po yy-Po) Additionally, 140 __ ,„glzZ.J,„gA_,ogilz^^ lim m = ^ Pi^Po->0 {^^—:: Po ^ ^ (l~Pl) log (1-Po) i^-Po) ^-^ = 00 Returning to the case of pj being at the five sigma level andpo at the six sigma level and allowing y9 and a to continue to have their original values of 0.03 and 0.02 respectively (although smaller values are needed and would produce larger sample sizes), the results are as follows: l o g ^ ^ = l o g — = 1.6857417 a ^0.02 233 log ( I Z ^ = l o g ^ l M O ^ = log.^99976^ = -0.000099726 g-Po) n 3.4 ^0.9999966 1,000,000 , p, , 0.000233 , o.,^„^^ log^^ = log = 1.835877 Po 0.0000034 Therefore, logllli^-.JlogA-iogC-/'.) m m= a { log Po g-Pi) g-Po); 1.6857417-r„ (1.835877+ 0.000099726) -0.000099726 g-Po) which means that, /w = -16,903.733 + r„ (18,410.211) This means that the researcher has to wait for one defective until he7she can reject the lot. However, it also means that the first defective can come during the first 10,527 observations (using 1.49 for r^ to calculate m), 141 In the first example (adapted from Wald 1973), the researcher needs a minimum of 14 observations to decide to accept the lot. He7she also needs a minimum of 4 observations to decide to reject the lot. These numbers do not seem unreasonable. However, the quality level for these observations is incredibly low. In the second case where the quality level is increased to today's standards, the researcher using SPRT has to rely on incredibly high number of observations. The lot as soon as the first defective observation within the first 10,711 observations since: /w = -16,903.733 + r„(18,410,211) Additionally, if the researcher did not find a defective in the first 10,711 observations, then he7she needs a minimum of 15,183 observations to decide to accept the lot. However, if a defective occurs after observation 10,711, then the researcher must wait until observation 24,388 (if no other defectives occurred) to accept the lot. These results can be represented graphically as follows: 142 4.000 If the defect occurs after observation 10,711 then the researcher must wait until observation 24,388 to accept the lot if no other items were found to be defective 3.500 •3.000 -0.500 •1.000 15,183 oservations must be tested with no defects before the researcher can consider to accept the lot •1.500 number of observations Figure A. 1: Sequential Sampling for a Very High Quality Process The conclusion is that the SPRT is no longer appropriate for today's quality levels. The example presented used five sigma and six sigma levels of manufacturing quality. Companies are now attempting to push the manufacturers to "nine sigma" (E. Castillo and S. Alvarez, personal communication, October 6, 2003), which would make current sampling results even more outrageous. 143 APPENDIX B COMPARISON OF METHODS AT A 5-SIGMA - RE method fc vc cf DSMH2 DSMH1 SkSPl ChSPl SING1 SkSP2 D0UB1 MULT1 ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 2 2 2 2 2 2 2 2 2 2 2 2 2 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1.000 1,000 1,000 1,000 1,000 1,000 DSMH2 DSMH1 SkSPl SING1 ChSPl SkSP2 D0UB1 MULT1 ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 20 20 20 20 20 20 20 20 20 20 20 20 20 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 9,550 9.807 10.254 13,440 13,455 15,393 19,845 21,430 24.595 25.355 26.216 49.532 57.408 97.70 95.30 90.90 68.70 68.80 60.10 46.40 42.90 37.30 36.00 34.80 17.80 15.20 2.30 4.70 9.10 31.30 31.20 39.90 53.60 57.10 62.70 64.00 65.20 82.20 84.80 DSMH2 DSMH1 SkSPl ChSPl SING1 SkSP2 D0UB1 MULT1 ChSP2 SiNG2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 100 100 100 100 100 100 100 100 100 100 100 100 100 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 9,628 9,980 13,174 29,496 29.498 39.179 61.622 69,583 85,417 89.420 93,699 211,451 251,417 96.90 92.90 70.50 31.40 31.50 23.70 15.00 13.20 10.70 10.20 9.70 4.20 3.50 3.10 7.10 29.50 68.60 68.50 76.30 85.00 86.80 89.30 89.80 90.30 95.80 96.50 1 46 93 474 427 729 1,183 1,420 1,884 1,831 1,967 4,970 5,797 samp/ week 1 10,000 19 194,646 37 367,910 200 2.000,000 200 2,000,000 297 2,967,260 522 5,215.028 601 6.014,005 760 7,600,000 800 8,000.000 843 8.431,470 20,231,250 2,023 24,218.880 2,422 93,291 93,454 93,242 92,839 92,992 93,198 93,241 93,328 93,550 93,307 93,329 93,235 93.034 1 26 86 490 488 731 1,206 1.439 1.756 1,927 2,020 4,959 5,902 1 10,000 13 131,720 37 367,580 200 2.000,000 200 2,000,000 297 2,969,760 522 5.215.509 601 6.014,165 760 7.600.000 8.000,000 800 8,433.810 843 20.230,750 2,023 24.222,260 2,422 93.276 92,727 92,970 93,241 93,249 93,562 93,394 93.280 93,198 93,330 93,051 92,958 93,303 3 13 93 479 465 706 1,207 1,413 1,744 1,908 1.977 4.788 5,923 1 10,000 50.259 5 368,080 37 2,000.000 200 2.000.000 200 2.966,420 297 5,215,353 522 6.013.865 601 7.600.000 760 8.000.000 800 8,431,100 843 20,224,000 2,022 24,223,860 2.422 TC/week TCF TCS% defectives % 9.537 97.90 2.10 93,343 9,559 97.40 2.60 93,123 93,354 9,602 97.10 2.90 9,887 93.90 6.10 93,336 9,904 93.90 6.10 93,459 10,026 92.10 7.90 93,034 10.413 88.00 12.00 92,864 10,584 86.70 13.30 93,204 10,846 84.10 15.90 93.117 10,952 83.50 16.50 93,322 10,984 82.80 17.20 92,914 13,066 67.40 32.60 93,099 93,294 13,802 63.40 36.60 144 detected totalsamp method fc vc Cf DSMH2 DSMHl SkSPl ChSPl SING1 SkSP2 DOUBl MULTl ChSP2 S1NG2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 2 2 2 2 2 2 2 2 2 2 2 2 2 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1,000 1,000 1,000 1,000 1,000 1,000 11,345 11,354 11,396 11.687 11.698 11,859 12,274 12.377 12,644 12,731 12,846 14,890 15.645 82.30 82.20 81.60 79.40 79.40 78.00 75.10 74.00 72.00 71.60 71.20 59.20 56.00 17.70 17.80 18.40 20.60 20.60 22.00 24.90 26.00 28.00 28.40 28.80 40.80 44.00 93,382 93,311 93.102 93,233 93,356 93.229 93.381 93.018 92.875 92.975 93.418 93.040 93,505 3 6 86 435 455 695 1.192 1.403 1.793 1.817 1,990 4,932 5,889 1 10,000 1 14,099 37 367.765 200 2,000,000 200 2,000,000 296 2,962,460 522 5,215,197 601 6,013.775 760 7,600,000 800 8.000,000 843 8,432,680 20,230,050 2,023 24,221,740 2.422 DSMH2 DSMHl SkSPl SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 S1NG2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 20 20 20 20 20 20 20 20 20 20 20 20 20 1,000 1,000 1,000 1,000 1.000 1.000 1.000 1.000 1.000 1,000 1,000 1,000 1,000 11.321 11.465 12.061 15,288 15,323 17.166 21.638 23.211 26.389 27,178 27,971 51,383 59,253 82.10 81.30 77.10 60.70 60.80 53.80 42.50 39.50 34.70 33.70 32.50 17.20 14.70 17.90 18.70 22.90 39.30 39.20 46.20 57.50 60.50 65.30 66.30 67.50 82.80 85.30 92,967 93.243 93.128 93.214 93.575 93,007 93,057 93,060 93,382 93,359 92,703 93,475 93,184 5 2 81 446 454 668 1,188 1,485 1,769 1,872 1,921 4,950 5,824 1 10,000 13,024 1 37 367,585 2,000,000 200 2,000,000 200 2,957,240 296 5,215.184 522 6,014.575 601 760 7.600.000 8,000,000 800 8.431.210 843 20.230,800 2.023 24,218,120 2.422 DSMH2 DSMHl SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 100 100 100 100 100 100 100 100 100 100 100 100 100 1,000 1.000 1.000 1,000 1,000 1.000 1.000 1.000 1,000 1,000 1,000 1,000 1,000 11,424 11.427 15.056 31.275 31.297 40.780 63.399 71.393 87.218 91.247 95,532 213.326 253.161 81.60 81.60 62.20 29.60 29.60 22.70 14.50 12.90 10.50 10.00 9.50 4.10 3.40 18.40 18.40 37.80 70.40 70.40 77.30 85.50 87.10 89.50 90.00 90.50 95.90 96.60 93,191 93.219 93.646 92,964 93,177 93,230 93,057 93,268 93,103 93,449 93,037 93.324 92.880 4 5 74 478 460 612 1,205 1,413 1,769 1,875 1,965 4,824 5.790 10,000 1 10,010 1 367,445 37 2.000.000 200 2,000,000 200 2,948,200 295 5,215,418 522 6,013,920 601 7,600,000 760 8,000,000 800 8,433.110 843 20,225,400 2.023 24,218,680 2.422 1,000 1,000 1,000 17,346 17,352 17,458 53.80 46.20 53.80 46.20 53.30 46.70 93.237 93.305 93.096 1 2 91 DSMHl 200 2 DSMH2 200 2 SkSPl 200 2 TC/week TCF TCS% defectives % 145 detected totalsamp 10,000 10,000 367,900 samp/ week 1 1 37 method fc vc cf ChSPl 200 SINGl 200 SkSP2 200 DOUBl 200 MULTl 200 ChSP2 200 SING2 200 MILST 200 D0UB2 200 MULT2 200 2 2 2 2 2 2 2 2 2 2 1.000 1.000 1.000 1.000 1.000 1,000 1,000 1,000 1.000 1.000 DSMHl 200 DSMH2 200 SkSPl 200 SINGl 200 ChSPl 200 SkSP2 200 DOUBl 200 MULTl 200 ChSP2 200 SING2 200 MILST 200 D0UB2 200 MULT2 200 20 20 20 20 20 20 20 20 20 20 20 20 20 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1.000 1,000 1,000 1,000 1.000 17,336 17,366 18,092 21,295 21,328 23,276 27.724 29.261 32,390 33.186 34,059 57.452 65.363 53.60 53.70 51.20 43.50 43.60 39.90 33.30 31.40 28.20 27:50 26.80 15.30 13.40 46.40 46.30 48.80 56.50 56.40 60.10 66.70 68.60 71.80 72.50 73.20 84.70 86.60 DSMH2 200 DSMH1 200 SkSPl 200 ChSPl 200 SINGl 200 SkSP2 200 DOUBl 200 MULTl 200 ChSP2 200 SING2 200 MILST 200 D0UB2 200 MULT2 200 100 100 100 100 100 100 100 100 100 100 100 100 100 1.000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 17.406 17.437 21.111 37,321 37,357 46,850 69,444 77,385 93,285 97,260 101,605 219,437 259.346 53.30 53.40 44.20 24.90 24.90 19.60 13.20 11.80 9.80 9.40 9.00 4.00 3.40 2 2 2 2 2 2 2 10,000 10,000 10,000 10,000 10,000 10.000 10,000 91.951 92.754 92.804 92.980 93.157 93.191 93.201 94.50 95.40 98.00 99.40 98.70 98.10 99.70 MULT2 D0UB2 MILST SINGl DOUBl SING2 SkSPl 5 5 5 5 5 5 5 449 443 685 1,256 1.382 1,705 1,856 1,979 4,988 5,845 samp/ week 200 2,000.000 200 2.000,000 2.960.040 296 522 5,216,055 601 6,013.455 760 7.600.000 800 8.000,000 843 8,430,720 20.233.250 2,023 24.219.760 2,422 92.967 93,260 92.800 93.084 93,432 93.552 93,613 93,136 92,912 92.993 93.172 93.237 93.273 3 3 96 461 482 728 1,242 1,385 1.681 1.846 1,938 5,054 5,961 1 10,000 1 10,000 37 368,145 200 2,000,000 200 2,000,000 297 2,969,480 5,215,938 522 601 6,013,660 7,600.000 760 800 8.000,000 843 8,430,620 20,235,900 2,024 24,222.320 2,422 46.70 46.60 55.80 75.10 75.10 80.40 86.80 88.20 90.20 90.60 91.00 96.00 96.60 92.859 93.166 93.483 93,182 93,556 92,696 93,134 92,795 93,365 93,172 93,537 93,159 93,622 1 1 101 418 464 663 1,265 1,449 1,779 1,931 2,015 4.951 5,856 1 10,000 1 10,000 37 368,345 2,000,000 200 2,000.000 200 2,957.260 296 5.216.107 522 601 6,014.340 7.600,000 760 8,000,000 800 8.431.180 843 20,229.950 2,023 24.219.760 2,422 5.50 4.60 2.00 0.60 1.30 1.90 0.30 92,776 93.333 92.916 92,859 93,151 93,215 93,007 5,877 4,831 2,002 480 1,239 1.827 82 24.220.040 2.422 20.225.250 2.023 8.432,530 843 2,000,000 200 5.215.860 522 8.000.000 800 367.700 37 TC/week TCF TCS% defectives % 17,691 52.40 47.60 93.062 17,693 52.40 47.60 93.084 17,880 51.70 48.30 93.069 18.280 50.30 49.70 93.159 18.475 49.90 50.10 93.615 18,764 49.00 51.00 93,587 18,799 48.60 51.40 93,262 18,879 48.40 51.60 93,293 20,958 41.90 58.10 92.861 21,718 40.20 59.80 93.172 146 detected totalsamp method fc vc cf ChSP2 MULTl SkSP2 DSMH2 ChSPl DSMHl 5 5 5 5 5 5 2 2 2 2 2 2 10.000 10.000 10.000 10.000 10.000 10.000 DSMH2 SkSPl DSMHl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 20 20 20 20 20 20 20 20 20 20 20 20 20 10,000 10,000 10,000 10,000 10,000 10,000 10.000 10.000 10.000 10.000 10,000 10,000 10.000 93,264 94,349 94,451 96,582 96,888 98,702 102,794 103.929 106.832 107.982 108.331 128.633 136.265 99.80 99.00 98.40 95.60 95.70 93.80 89.60 88.20 85.60 85.00 84.20 68.40 64.30 0.20 1.00 1.60 4.40 4.30 6.20 10.40 11.80 14.40 15.00 15.80 31.60 35.70 93,044 93.494 93.055 92.873 93,136 93,268 93,425 93,074 93,176 93,591 93,153 92,811 93,374 1 83 156 496 452 707 1,275 1.384 1.759 1.825 1,898 4,875 5,796 1 10,000 37 367,765 67 674,879 200 2,000,000 200 2.000,000 297 2,966,940 522 5,216,172 601 6,013,615 760 7,600,000 800 8.000,000 843 8,429,750 20,227,500 2,023 24,218,800 2,422 DSMH2 DSMHl SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 S!NG2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 100 100 100 100 100 100 100 100 100 100 100 100 100 10,000 10,000 10,000 10,000 10,000 10.000 10.000 10,000 10,000 10,000 10,000 10,000 10,000 93.851 96.066 96.606 112.929 113.151 122.424 144,144 152,587 167,838 171,209 175,905 291,058 329.812 99.70 96.60 96.00 82.10 82.10 75.60 63.60 60.40 54.60 53.10 51.90 30.40 26.40 0.30 3.40 4.00 17.90 17.90 24.40 36.40 39.60 45.40 46.90 48.10 69.60 73.60 93,552 92,857 92,807 93,155 93,379 93,242 93,012 93,616 93.322 92,894 93,234 93,284 93.037 2 61 83 445 447 736 1,284 1.429 1.756 1.965 1.936 4.927 5.887 1 10,000 306,715 31 367,635 37 2.000.000 200 2,000,000 200 297 2,968,700 5,216,328 522 6,014.175 601 7,600,000 760 8.000,000 800 8.432.800 843 20,230.100 2.023 24.222.280 2.422 MULT2 D0UB2 DSMH2 MULTl ChSP2 MILST SING2 SINGl SkSP2 SkSPl ChSPl 50 50 50 50 50 50 50 50 50 50 50 2 2 2 2 2 2 2 2 2 2 2 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 94.208 94.427 94.670 94.670 94.850 94.968 95.081 95.136 95.277 95.303 95,310 92.70 93.60 97.90 96.60 96.30 96.10 96.20 97.50 97.30 97.80 97.50 7.30 6.40 2.10 3.40 3.70 3.90 3.80 2.50 2.70 2.20 2.50 93,145 93,260 92,666 92,831 93,053 93,250 93,270 93,212 93,364 93,296 93,341 5.820 4.912 3 1.377 1,738 1,985 1,804 484 692 88 439 24,218.900 2.422 20,228,200 2.023 10,000 1 6,013,660 601 7,600,000 760 8,432,270 843 8,000,000 800 2,000,000 200 2,963.060 296 367,810 37 2,000,000 200 TC/week TCF TCS% defectives % 93,295 98.20 1.80 93,360 93,460 98.50 1.50 93,479 93.488 99.20 0.80 93.364 93,521 99.80 0.20 93.320 93,598 99.40 0.60 93.449 93,811 99.40 0.60 93,679 147 detected totalsamp 1,788 1,424 670 2 452 402 7,600,000 6,013.960 2,960,760 10,000 2,000.000 1.666,516 samp/ week 760 601 296 1 200 167 method fc vc DOUBl 50 DSMH1 50 2 2 TC/week TCF TCS% defectives % 10.000 95,460 96.80 3.20 93.596 10.000 95,602 97.80 2.20 93.580 DSMH2 DSMHl SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 20 20 20 20 20 20 20 20 20 20 20 20 20 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10.000 10,000 95,288 95,568 96,026 98,741 98,970 100,288 103,953 106,010 108,658 109,572 109,838 130,736 137,888 97.90 97.40 97.10 93.90 93.90 92.10 88.00 86.70 84.10 83.50 82.80 67.50 63.40 2.10 2.60 2.90 6.10 6.10 7.90 12.00 13.30 15.90 16.50 17.20 32.50 36.60 93.266 93.093 93,361 93.218 93.443 93.033 92.776 93,329 93,185 93.402 92,831 93,150 93,317 3 40 93 488 484 683 1,277 1,369 1,755 1,859 1,887 4,947 5,962 1 10.000 19 193.291 37 367.910 200 2.000.000 200 2,000,000 296 2,959.960 522 5.216,341 601 6.013.530 760 7.600,000 800 8.000.000 843 8,432,280 20,231,700 2,023 24,225,460 2,423 DSMH2 DSMHl SkSPl SINGl ChSPl SkSP2 DOUBl MULT1 ChSP2 SING2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 100 100 100 100 100 100 100 100 100 100 100 100 100 10,000 10.000 10.000 10,000 10,000 10,000 10,000 10,000 10.000 10.000 10.000 10,000 10,000 95,557 97,059 98,828 114,220 114.547 124,221 146,385 154,236 169,852 173,554 177,592 292,889 331,465 97.80 96.30 94.20 80.70 80.80 74.50 63.00 59.70 54.00 52.70 51.30 30.20 26.20 2.20 3.70 5.80 19.30 19.20 25.50 37.00 40.30 46.00 47.30 48.70 69.80 73.80 93,453 93,464 93,210 92,674 92,975 93,262 93,358 93,393 93,563 93,342 93.147 93.274 92,869 1 39 83 481 453 665 1,182 1,357 1,797 1,878 1,981 4,887 5,874 1 10,000 16 162,191 367,630 37 2,000,000 200 200 2,000,000 296 2,958,560 522 5,215,080 601 6,013,385 760 7,600,000 800 8,000,000 843 8,433,230 20,227,700 2.023 24,220,060 2.422 10,000 99,995 87.80 12.20 10,000 100,274 87.00 13.00 10,000 100,915 90.50 9.50 10,000 100,951 90.30 9.70 10,000 100,987 90.80 9.20 10,000 101.017 92.10 7.90 10,000 101,018 92.00 8.00 10,000 101,099 91.70 8.30 10,000 101.178 90.50 9.50 10,000 101,224 91.90 8.10 10,000 101,340 91.00 9.00 10,000 101,352 91.70 8.30 10,000 101,430 91.50 8.50 92.907 93,171 93.139 93.107 93.173 92.997 92.897 93.121 93,406 93,150 93,425 93,367 93,493 5,085 5,884 1,801 1,902 1,439 2 9 451 1.887 84 1,172 444 707 20,235,050 2,024 24,220.780 2,422 7,600,000 760 8,431,670 843 6,014.110 601 10.000 1 44.433 4 2,000,000 200 8,000,000 800 367.760 37 5.215.080 522 2,000,000 200 2,966,880 297 93,443 3 D0UB2 200 MULT2 200 ChSP2 200 MILST 200 MULTl 200 DSMH2 200 DSMH1 200 ChSPl 200 SING2 200 SkSPl 200 DOUBl 200 SINGl 200 SkSP2 200 2 2 2 2 2 2 2 2 2 2 2 2 2 cf DSMH2 200 20 10,000 101,480 92.10 7.90 148 detected totalsamp 1,191 47 5,215,301 202.292 10,000 samp/ week 522 20 1 TC/week TCF •rcs% defectives % 101,748 "31.60 8.40 93,252 101,845 91.30 8.70 93,101 104,697 88.50 11.50 93,136 104,882 88.50 11.50 93,303 106,529 86.90 13.10 93,251 110,503 83.30 16.70 93,276 111,876 82.00 18.00 93,209 114.466 79.70 20.30 92,957 115.137 79.10 20.90 92,840 116.623 78.60 21.40 93.679 93.074 136.754 64.40 35.60 144,188 60.70 39.30 93.312 method fc vc cf DSMHl :200 SkSPl :200 ChSPl '200 SINGl 200 SkSP2 200 DOUBl 200 MULTl 200 ChSP2 200 S1NG2 200 MILST 200 D0UB2 200 MULT2 200 20 20 20 20 20 20 20 20 20 20 20 20 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 DSMH2 200 DSMHl 200 SkSPl 200 SINGl 200 ChSPl 200 SkSP2 200 DOUBl 200 MULTl 200 ChSP2 200 SING2 200 MILST 200 D0UB2 200 MULT2 200 100 100 100 100 100 100 100 100 100 100 100 100 100 10,000 10,000 10,000 10,000 10.000 10,000 10.000 10,000 10,000 10,000 10,000 10.000 10.000 100,936 102,060 104,765 120,776 121.241 129,932 152.288 160,000 175,553 179,547 183,574 299,010 338.011 92.00 91.40 88.80 76.80 76.90 71.00 60.40 57.30 52.10 50.90 49.60 29.60 25.90 8.00 8.60 11.20 23.20 23.10 29.00 39.60 42.70 47.90 49.10 50.40 70.40 74.10 samp/ week 4 37 200 200 296 522 601 760 800 843 2.023 2.421 detected totalsamp 16 75 472 454 705 1,260 1,421 1,760 1,773 1,998 4,938 5,733 44,699 367,390 2.000,000 2,000.000 2,964,440 5,216,068 6.013.920 7.600.000 8.000.000 8,432.850 20.229,850 24,214,200 92.819 93,291 93,083 93.199 93,677 92,950 93,304 93,165 93,177 93.271 93.047 93,296 93,262 3 6 82 471 485 662 1,275 1,408 1,749 1,855 1,924 4,872 5,790 1 10.000 41,265 4 367,705 37 2,000,000 200 2,000,000 200 2,957.040 296 5,216.263 522 6.013.860 601 7.600.000 760 8.000.000 800 8,431,560 843 20,227.400 2,023 24,217,180 2,422 MULT2 D0UB2 MILST ChSP2 SING2 MULTl DOUBl ChSPl DSMHl SINGl SkSP2 DSMH2! SkSPl 5 5 5 5 5 5 5 5 5 5 5 5 5 2 2 2 2 2 2 2 2 2 2 2 2 2 50,000 50,000 50.000 50.000 50,000 50,000 50.000 50,000 50,000 50,000 50,000 50,000 50,000 440.626 444.474 458.229 458.628 460,663 460,665 461.605 462,651 462,696 463,151 464,171 464,839 467,736 98.90 99.00 99.60 99.60 99.60 99.70 99.70 99.90 99.80 99.90 99.80 99.90 99.90 1.10 1.00 0.40 0.40 0.40 0.30 0.30 0.10 0.20 0.10 0.20 0.10 0.10 92,919 93.051 93.212 93.066 93,664 93,257 93,238 92,860 93,176 92,964 93,409 93.229 93,568 5,804 5,007 1.944 1.685 1.892 1,405 1,166 450 825 454 734 362 76 24.217,200 2,422 20,233.400 2,023 8,432,360 843 7,600,000 760 8,000,000 800 6,013,810 601 5,214,898 521 2,000,000 200 3,695,381 370 2,000,000 200 2,969,800 297 1,515,025 152 367,515 37 SkSPl DSMH1 DSMH2! SINGl ChSPl 5 5 5 5 5 20 20 20 20 20 50,000 50,000 50,000 50,000 50,000 465,387 467,472 467.517 467,624 467,629 99.80 99.30 99.90 99.10 99.1C 0.20 0.70 0.10 0.90 0.90 92,965 93,212 93,435 93,125 93,154 75 380 54 441 469 37 367,360 1,554,011 155 205,487 21 2,000,000 200 1 2,000,000 1 200 149 method fc vc cf TC/week SkSP2 DOUBl SING2 MULT1 ChSP2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 20 20 20 20 20 20 20 20 50,000 50,000 50,000 50,000 50.000 50.000 50,000 50,000 468,040 470,665 472.631 472,897 473,805 473,942 480,620 485,865 TCF TCS% defectives % 98.70 1.30 93,088 97.70 2.30 93,165 96.60 3.40 93,160 97.40 2.60 93,610 96.70 3.30 93.410 96.40 3.60 93,333 92,954 91.50 8.50 90.00 10.00 93,342 DSMH2 SkSPl DSMH1 SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 100 100 100 100 100 100 100 100 100 100 100 100 100 50.000 50.000 50.000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 466,964 468,740 471,829 483,768 484,350 494,127 514,454 518,769 532,019 535.955 539,209 644,066 678,422 99.80 99.20 98.40 95.80 95.80 93.90 89.80 88.40 85.70 85.00 84.30 68.50 64.20 0.20 0.80 1.60 4.20 4.20 6.10 10.20 11.60 14.30 15.00 15.70 31.50 35.80 MULT2 D0UB2 SING2 MILST ChSP2 MULTl DOUBl SkSP2 ChSPl DSMHl DSMH2 SINGl SkSPl 50 50 50 50 50 50 50 50 50 50 50 50 50 2 2 2 2 2 2 2 2 2 2 2 2 2 50.000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 444,173 446,621 459,676 460,558 461,461 461,745 462,285 463,909 465,343 466,587 466,807 468,023 468,595 98.50 98.60 99.20 99.20 99.20 99.30 99.30 99.40 99.50 99.50 99.60 99.50 99.60 SkSPl DSMH2 SINGl DSMHl DOUBl SkSP2 ChSPl MILST SING2 50 50 50 50 50 50 50 50 50 20 20 20 20 20 20 20 20 20 50,000 50,000 50.000 50.000 50.000 50,000 50.000 50.000 50,000 468,147 468.510 469.686 470,470 470,912 470,972 471,411 473,322 474,155 99.40 99.60 98.70 99.20 97.40 98.30 98.70 96.00 96.20 707 1,160 1,877 1,479 1,732 1,962 4,971 5,908 samp/ week 296 2,964,300 521 5,214,755 800 8,000,000 601 6,014.525 7,600,000 760 843 8,434,710 20,231.350 2,023 24,222,080 2.422 93,265 93.061 93.055 93.141 93.301 93.560 93.647 93,037 92,958 92,985 92,943 93,219 92,967 13 90 184 431 475 715 1,238 1,361 1.809 1.849 2,021 4,946 5,802 5 50.304 37 367,890 73 726,606 2,000,000 200 2.000.000 200 2.967,240 297 522 5,215,821 601 6,013.375 7.600,000 760 8,000,000 800 8,431,690 843 20,230,000 2.023 24,216,160 2,422 1.50 1.40 0.80 0.80 0.80 0.70 0.70 0.60 0.50 0.50 0.40 0.50 0.40 93,316 93,107 93,090 93,414 93,405 93,052 93,025 92,948 93,041 93,123 92,964 93,590 93,382 5,858 4,999 1,878 2,043 1,820 1.346 1.179 687 454 279 4 467 82 24,219.780 2,422 20.232.550 2,023 8.000,000 800 8.430,910 843 7.600,000 760 6.013.235 601 5.215,080 522 2.959,700 296 2,000,000 200 1,245,980 125 10,000 1 2,000,000 200 367,500 37 0.60 0.40 1.30 0.80 2.60 1.70 1.30 4.00 3.80 93,156 93,298 93,178 93,538 92,916 93,287 93,527 92,848 93,124 78 1 443 235 1.224 680 447 1.962 1.899 150 detected totalsamp 367,605 10,000 2.000.000 973,476 5.215.613 2.959.860 2.000,000 8,430,800 8,000,000 37 1 200 97 522 296 200 843 800 1.423 1.737 5,836 4,879 samp/ week 601 6,014,045 7,600,000 760 24.217,740 2,422 20,227.650 2,023 93,185 93.438 93,134 92,824 92,834 93,281 93,447 93,178 93,258 93,636 93,241 92,852 93,299 1 97 141 473 477 734 1,246 1,408 1,845 1,816 1,941 4,847 5,811 1 10.000 368.230 37 575.099 58 2.000.000 200 2.000.000 200 2.969,460 297 5.215.886 522 6.013,790 601 7,600,000 760 8,000,000 800 8,433,220 843 20,226,650 2,023 24.220,020 2,422 2.90 2.70 2.10 2.00 2.10 1.90 2.00 1.80 1.70 1.80 1.80 1.80 1.70 93,378 93,193 92,738 93,293 93,438 93,172 93,420 93,226 92,697 93,260 92,937 93,537 93,445 5,932 5,038 1,800 1,739 1.852 1,231 1,410 722 85 444 58 454 5 24,222.160 2,422 20,235,100 2,024 8,000,000 800 7,600,000 760 8.431,280 843 5.215,704 522 6.013,970 601 2,965,060 297 367,840 37 2,000,000 200 258,379 26 2.000,000 200 10.000 1 1.70 2.50 1.90 1.90 4.20 2.50 2.90 4.90 3.90 5.00 5.20 9.90 11.50 92,795 93.053 93.349 93.398 92.717 93,379 93,331 93,005 93,424 93,268 93,352 93,384 93,182 1 428 71 97 1,416 448 720 1,838 1,280 1,890 1,939 4,970 5,830 10.000 1 2.000.000 200 258,319 26 368,055 37 6,014,035 601 2,000,000 200 2.968.240 297 7.600.000 760 5.216.393 522 8.000.000 800 8,429,890 843 20,231,650 2,023 24,218,640 2,422 method fc vc cf TC/week MULTl ChSP2 MULT2 D0UB2 50 50 50 50 20 20 20 20 50,000 50,000 50,000 50,000 474,257 474,878 485,201 485,534 TCF TCS% defectives % 93,464 97.00 3.00 93,267 96.40 3.60 92,773 89.60 10.40 93,481 91.20 8.80 DSMH2 SkSPl DSMHl ChSPl SING1 SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 100 100 100 100 100 100 100 100 100 100 100 100 100 50.000 50.000 50,000 50.000 50.000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 468,025 472.412 472.727 483,781 483,811 494,471 515,225 521,056 535.153 541.187 542,925 644,515 681,907 99.60 98.80 98.40 95.40 95.40 93.60 89.50 88.10 85.40 84.80 84.10 68.30 64.10 0.40 1.20 1.60 4.60 4.60 6.40 10.50 11.90 14.60 15.20 15.90 31.70 35.90 MULT2 200 D0UB2 200 SING2 200 ChSP2 200 MILST 200 DOUBl 200 MULTl 200 SkSP2 200 SkSPl 200 ChSPl 200 DSMHl 200 SINGl 200 DSMH2 200 2 2 2 2 2 2 2 2 2 2 2 2 2 50,000 50,000 50,000 50,000 50,000 50.000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 450,218 452,947 464,347 467,346 467,675 468,794 469,302 471,162 471,218 472,509 473,041 473,844 475.222 97.10 97.30 97.90 98.00 97.90 98.10 98.00 98.20 98.30 98.20 98.20 98.20 98.30 DSMH2 200 SINGl 200 DSMH1 200 SkSPl 200 MULTl 200 ChSPl 200 SkSP2 200 ChSP2 200 DOUBl 200 SING2 200 MILST 200 D0UB2 200 MULT2 200 20 20 20 20 20 20 20 20 20 20 20 20 20 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 472.010 475,157 475,281 475,327 476,593 476,688 477,047 479,106 479,209 480,963 481.999 490.692 493,381 98.30 97.50 98.10 98.10 95.80 97.50 97.10 95.10 96.10 95.00 94.80 90.10 88.50 151 detected totalsamp method fc vc cf DSMH2 200 DSMHl 200 SkSPl 200 ChSPl 200 SINGl 200 SkSP2 200 DOUBl 200 MULT1 200 ChSP2 200 SING2 200 MILST 200 D0UB2 200 MULT2 200 100 100 100 100 100 100 100 100 100 100 100 100 100 50,000 50,000 50.000 50.000 50.000 50.000 50.000 50.000 50,000 50,000 50,000 50,000 50,000 TC/week TCF TCS% defectives % 473,540 476,054 477,978 491,919 492.645 498.539 517.925 528.322 541,679 546,019 547,987 651,368 687,810 98.30 97.80 97.50 94.30 94.30 92.50 88.40 87.10 84.50 83.90 83.10 67.70 63.60 1.70 2.20 2.50 5.70 5.70 7.50 11.60 12.90 15.50 16.10 16.90 32.30 36.40 152 93,087 93,120 93,338 93,183 93.413 92,824 92,760 93,414 93,325 93.401 93.048 93.123 93,352 detected 3 43 96 408 494 633 1,225 1,398 1,815 1,823 1,940 4,976 5,911 totalsamp samp/ week 1 10,000 241,407 24 368,000 37 2,000,000 200 2,000,000 200 2,951,260 295 5,215,665 522 6,013,780 601 7.600.000 760 8.000,000 800 8,431,090 843 20,231,400 2,023 24,223,060 2,422 APPENDDC C COMPARISON OF METHODS AT A 6-SIGMA - RE TC/week TCF % TCS defectives detected % 0 3537 42 59.8 338 1 3528 356 38.1 61.9 1 3531 33.4 66.6 406 12 3535 736 18.4 81.6 21 3561 18.5 81.5 736 20 3523 84.7 883 153 43 3548 1375 09.8 90.2 69 3541 1534 08.7 91.3 69 3519 1853 07.2 92.8 76 3526 1933 6.9 93.1 89 3535 1965 6.7 93.3 169 3517 4330 97.0 3.0 247 97.5 3544 5128 2.5 totalsamp samp/ week 1 8 35 200 200 274 520 600 760 800 816 2000 2400 method fc vc cf DSMH2 DSMHl SkSPl SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 S1NG2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 2 2 2 2 2 2 2 2 2 2 2 2 2 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 DSMH2 DSMHl SkSPl SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 20 20 20 20 20 20 20 20 20 20 20 20 20 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1.000 1,000 1,000 1,000 1,000 356 458 1,030 4,336 4,336 5,815 10,735 12,335 15,533 16,333 16,658 40,336 48.334 38.2 29.7 13.2 3.1 3.1 2.3 1.2 1.1 0.9 0.8 0.8 0.3 0.3 61.8 70.3 86.8 96.9 96.9 97.7 98.8 98.9 99.1 99.2 99.2 99.7 99.7 3,534 3,530 3,549 3,534 3,552 3,520 3,536 3,538 3,534 3,529 3,559 3,530 3,519 1 0 1 14 17 22 49 45 81 80 75 188 211 1 26,000 158,000 6 900,030 35 5,200,000 200 5,200,000 200 7.123,800 274 13,520,637 520 15,600.450 600 19,760,000 760 20,800,000 800 21,220,080 816 52,008,950 2,000 62,407.780 2,400 DSMH2 DSMHl SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 100 100 100 100 100 100 100 100 100 100 100 100 100 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1.000 1.000 1,000 436 31.2 621 21.8 3.797 3.5 0.7 20,334 0.7 20,335 0.5 27,737 0.3 52,336 0.2 60,335 0.2 76,335 0.2 80,336 0.2 81,950 200,365 0.1 240,365 0.1 68.8 78.2 96.5 99.3 99.3 99.5 99.7 99.8 99.8 99.8 99.8 99.9 99.9 3,539 3,513 3,500 3,495 3,509 3,579 3,497 3,488 3,545 3,575 3,489 3,553 3,563 0 0 2 14 19 22 47 47 78 68 72 168 216 26.000 1 73,000 3 900,060 35 5.200,000 200 5,200,000 200 7,123,940 274 13,520,598 520 15,600,470 600 19,760,000 760 20,800,000 800 21,220,370 816 52,008,250 2,000 62,408,380 2,400 153 26000 202189 900030 5200000 5200000 7123460 13520546 15600670 19760000 20800000 21220420 52007900 62409280 totalsamp samp/ week method fc vc cf DSMH2 DSMH1 SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 2 2 2 2 2 2 2 2 2 2 2 2 2 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1.000 1,000 1,000 2,140 2,147 2,217 2.537 2.538 2.689 3.176 3,336 3.656 3.735 3.767 6.133 6.931 6.3 6.4 6.1 5.3 5.4 5.0 4.2 4.0 3.7 3.6 3.5 2.1 1.8 93.7 93.6 93.9 94.7 94.6 95.0 95.8 96.0 96.3 96.4 96.5 97.9 98.2 3,525 3.571 3,539 3.520 3.564 3.532 3.532 3.529 3.556 3.532 3.515 3.548 3.538 0 0 1 16 17 17 46 53 84 72 78 176 210 1 26,000 30,000 1 900,030 35 5,200,000 200 5,200,000 200 7.123.000 274 520 13,520.598 15,600,530 600 19,760.000 760 20,800,000 800 21,220.000 816 52.008.300 2,000 62.408,180 2,400 DSMH2 DSMHl SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 20 20 20 20 20 20 20 20 20 20 20 20 20 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1.000 1,000 1,000 2.157 2,203 2,840 6.137 6.138 7.620 12.537 14.137 17.336 18,134 18,458 42,138 50.137 6.3 6.2 4.8 2.2 2.2 1.8 1.1 1.0 0.8 0.7 0.7 0.3 0.3 93.7 93.8 95.2 97.8 97.8 98.2 98.9 99.0 99.2 99.3 99.3 99.7 99.7 3,506 3,540 3,545 3,536 3,562 3,536 3,544 3,546 3,540 3,512 3,527 3,513 3.534 0 0 1 18 14 13 61 50 68 77 82 205 224 1 26,000 1 29,000 35 900,030 200 5,200,000 200 5,200,000 274 7,122,340 13,520,780 520 15,600,500 600 19,760,000 760 20,800,000 800 21,220.180 816 52.009.750 2,000 62.408,620 2,400 DSMHl DSMH2 SkSPl SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 100 100 100 100 100 100 100 100 100 100 100 100 100 1,000 1.000 1.000 1.000 1.000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 2.238 2.238 5.609 22.137 22.138 29.545 54.139 62.139 78.136 82,137 83,753 202,167 242,167 6.1 6.1 2.4 0.6 0.6 0.5 0.2 0.2 0.2 0.2 0.2 0.1 0.1 93.9 93.9 97.6 99.4 99.4 99.5 99.8 99.8 99.8 99.8 99.8 99.9 99.9 3,540 3,540 3,516 3,520 3,551 3,542 3,519 3,535 3,532 3,545 3,539 3,554 3.545 0 0 3 16 16 30 49 57 72 70 73 178 230 26,000 26,000 900,090 5.200.000 5,200,000 7,124,960 13,520,637 15,600,570 19,760.000 20.800.000 21.220.230 52.008,450 62,408,780 1 1 35 200 200 274 520 600 760 800 816 2,000 2,400 2 2 2 1,000 1,000 1,000 8.146 8.146 8,251 1.7 1.7 1.6 98.3 98.3 98.4 3.535 3.544 3,540 0 0 5 26,000 26,000 900,150 1 1 35 DSMHl 200 DSMH2 200 SkSPl 200 TC/week TCF % TCS defectives detected % 154 method fc vc cf SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 S1NG2 MILST D0UB2 MULT2 200 200 200 200 200 200 200 200 200 200 2 2 2 2 2 2 2 2 2 2 1,000 1,000 1.000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 DSMHl DSMH2 SkSPl SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 200 200 200 200 200 200 200 200 200 200 200 200 200 20 20 20 20 20 20 20 20 20 20 20 20 20 1,000 1.000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 DSMH2 DSMH1 SkSPl SINGl ChSPl SkSP2 DOUBl MULT1 ChSP2 SING2 MILST D0UB2 MULT2 200 200 200 200 200 200 200 200 200 200 200 200 200 100 1,000 100 1,000 100 1,000 100 1,000 100 1,000 100 1,000 100 1,000 100 1.000 100 1.000 100 1.000 100 1,000 100 1,000 100 1,000 DSMHl DSMH2 SkSPl SINGl ChSPl SkSP2 DOUBl 5 5 5 5 5 5 5 2 2 2 2 2 2 2 10,000 10,000 10,000 10,000 10,000 10,000 10,000 TC/week TCF % TCS defectives detected % 14 3,551 98.4 8,544 1.6 19 3,550 98.4 8,544 1.6 12 3,514 98.5 8,706 1.5 41 3,541 98.5 9,183 1.5 49 3.544 98.6 9,343 1.4 78 3.534 98.6 1.4 9,661 68 3,518 98.6 9.741 1.4 66 3,552 98.6 9.775 1.4 167 3,530 98.9 12.139 1.1 224 3,539 99.0 12.938 1.0 5,200,000 5,200,000 7,122,000 13,520,533 15,600,490 19,760.000 20.800.000 21,220,080 52,008.150 62.408.760 samp/ week 200 200 274 520 600 760 800 816 2,000 2,400 totalsamp 8,163 8.164 8.873 12,142 12,142 13,639 18,542 20.141 23,344 24,143 24,464 48,145 56,145 1.7 1.7 1.5 1.1 1.1 1.0 0.7 0.7 0.6 0.6 0.5 0.3 0.2 98.3 98.3 98.5 98.9 98.9 99.0 99.3 99.3 99.4 99.4 99.5 99.7 99.8 3,510 3,549 3,503 3,517 3,495 3,572 3,510 3,501 3,586 3,561 3,496 3,532 3,564 0 0 3 19 18 14 57 52 61 77 66 186 235 26,000 26,000 900,090 5,200,000 5,200,000 7,122,420 13,520,741 15,600,520 19,760,000 20,800,000 21,220,320 52,008,850 62.409,040 1 1 35 200 200 274 520 600 760 800 816 2.000 2.400 8,243 8,245 11,644 28,143 28,143 35,563 60,145 68,144 84.144 88.144 89.757 208,176 248,170 1.6 1.7 1.2 0.5 0.5 0.4 0.2 0.2 0.2 0.2 0.1 0.1 0.1 98.4 98.3 98.8 99.5 99.5 99.6 99.8 99.8 99.8 99.8 99.9 99.9 99.9 3,513 3,578 3,546 3,538 3,525 3,546 3.527 3,523 3,562 3,556 3,506 3,546 3,523 0 0 2 21 19 28 38 55 69 60 85 193 204 26,000 26,000 900,060 5,200,000 5,200,000 7,125,040 13,520.494 15.600.550 19.760.000 20.800.000 21,220.210 52,009.200 62,407.780 1 1 35 200 200 274 520 600 760 800 816 2,000 2,400 3,526 3,538 3,529 3,526 3,569 3,526 3,528 7 0 3 13 21 22 55 1.843,640 26.000 900,090 5,200,000 5,200,000 7,123,960 13,520.702 71 1 35 200 200 274 520 1,696 1,563 1,627 1,951 1,965 2,096 2,576 79.8 20.2 87.1 12.9 83.4 16.6 69.2 30.8 69.5 30.5 64.3 35.7 51.9 1 48.1 155 TC/week TCF % TCS defectives detected totalsamp samp/ week % 15.600,740 600 74 2.731 48.7 51.3 3,533 19,760,000 760 76 3,531 3,049 43.6 56.4 20,800,000 800 77 3,532 3,129 42.5 57.5 21,220,240 816 82 3,555 3,168 42.2 57.8 52,009,000 2,000 187 5,488 3,534 23.5 76.5 62,408,580 2,400 226 3,542 6,276 20.3 79.7 method fc vc Cf MULTl ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 5 2 2 2 2 2 2 10,000 10,000 10,000 10,000 10,000 10,000 DSMHl DSMH2 SkSPl SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 20 20 20 20 20 20 20 20 20 20 20 20 20 10,000 2,300 10,000 1,579 10,000 2,256 10,000 5,553 10,000 5,556 10,000 7,027 10,000 11,947 10,000 13,546 10,000 16,725 10,000 17,529 10,000 17,865 10,000 41,490 10,000 49.493 58.9 86.1 60.4 24.4 24.4 19.2 11.3 9.9 7.9 7.6 7.5 3.1 2.6 41.1 13.9 39.6 75.6 75.6 80.8 88.7 90.1 92.1 92.4 92.5 96.9 97.4 3,525 3,532 3,547 3,535 3,546 3,522 3,546 3,540 3,528 3,533 3,557 3,534 3,549 4 0 5 18 22 22 45 42 83 79 70 202 205 37 968,747 26.000 1 35 900.150 200 5,200.000 5,200,000 200 274 7,123,840 13,520.572 520 15,600,420 600 19,760,000 760 20,800,000 800 21,220,050 816 52,009,600 2,000 62,407,680 2,400 DSMHl DSMH2 SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 100 100 100 100 100 100 100 100 100 100 100 100 100 10,000 10,000 10,000 10,000 10,000 10,000 10.000 10,000 10,000 10,000 10,000 10,000 10,000 3.336 1,663 5.011 21.540 21.544 28.968 53.529 61,531 77,546 81,544 83,132 201,536 241,528 40.5 81.9 26.9 6.2 6.2 4.7 2.5 2.2 1.7 1.6 1.6 0.6 0.5 59.5 18.1 73.1 93.8 93.8 95.3 97.5 97.8 98.3 98.4 98.4 99.4 99.5 3,513 3,542 3,507 3,498 3,510 3,580 3,487 3,499 3,579 3,563 3,492 3,579 3,569 1 0 3 16 16 18 39 46 83 73 78 221 209 464,080 18 1 26,000 35 900,090 5,200,000 200 5,200,000 200 274 7,123,240 13,520,507 520 15,600,460 600 19,760,000 760 20,800,000 800 21,220,550 816 52,010,500 2,000 62,408.360 2,400 DSMHl DSMH2 SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST 50 50 50 50 50 50 50 50 50 50 50 2 2 2 2 2 2 2 2 2 2 2 10,000 10.000 10.000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 3,399 3,359 3,442 3,749 3,763 3,910 4,393 4,541 4,872 4,936 4,965 40.4 40.3 39.5 35.9 36.2 34.7 30.8 29.5 27.7 27.0 26.8 59.6 59.7 60.5 64.1 63.8 65.3 69.2 70.5 72.3 73.0 73.2 3,572 3,524 3,540 3,523 3,560 3,545 3,546 3,538 3,576 3.535 3.527 1 0 2 21 21 18 33 58 66 67 67 209.198 26.000 900.060 5.200.000 5.200.000 7.123.060 13.520.429 15,600,580 19,760,000 20,800,000 21,220,660 156 8 1 35 200 200 274 520 600 760 800 816 method fc vc D0UB2 MULT2 50 50 2 2 TC/week TCF % TCS defectives detected totalsamp samp/ week % 52,008,800 2,000 191 10,000 7,291 3,540 17.7 82.3 62,408,100 2,400 208 10,000 8,073 3,510 15.7 84.3 DSMH2 DSMH1 SkSPl ChSPl S1NG1 SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 20 20 20 20 20 20 20 20 20 20 20 20 20 10.000 10.000 10.000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 3,374 3,566 4,071 7,353 7,365 8,834 13.752 15.342 18,540 19,328 19,658 43,294 51,271 40.1 38.2 33.6 18.4 18.5 15.3 9.8 8.7 7.2 6.9 6.8 3.0 2.5 59.9 61.8 66.4 81.6 81.5 84.7 90.2 91.3 92.8 93.1 93.2 97.0 97.5 3,516 3,537 3,556 3,529 3,566 3,530 3,560 3,550 3,544 3,517 3,526 3,521 3,505 0 0 2 16 23 26 52 68 65 70 62 182 224 1 26,000 202,000 8 900,060 35 5,200,000 200 5,200,000 200 7,124,680 274 13,520,650 520 15,600,680 600 19,760,000 760 20,800,000 800 21,220.160 816 52,008,700 2,000 62,408,320 2,400 DSMHl DSMH2 SkSPl SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 100 100 100 100 100 100 100 100 100 100 100 100 100 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 4,081 3.462 6.829 23,345 23,355 30,756 55.344 63.341 79,332 83,339 84,943 203,339 243,322 33.3 39.3 19.8 5.8 5.8 4.4 2.4 2.1 1.7 1.6 1.6 0.6 0.5 66.7 60.7 80.2 94.2 94.2 95.6 97.6 97.9 98.3 98.4 98.4 99.4 99.5 3,530 3,537 3,530 3,510 3,541 3,535 3,525 3,530 3,526 3,536 3,531 3,557 3,565 1 0 6 18 25 19 43 57 70 62 90 168 225 7 179,164 1 26,000 35 900,180 5,200,000 200 5,200,000 200 274 7,123,420 13,520,546 520 15,600.570 600 19.760.000 760 20.800,000 800 21,220,080 816 52,007,900 2,000 62,408,320 2,400 DSMH2 DSMHl SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 200 200 200 200 200 200 200 200 200 200 200 200 200 2 2 2 2 2 2 2 2 2 2 2 2 2 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 9,374 9,410 9,470 9,758 9,764 9.916 10.390 10.551 10.863 10.936 10.983 13.305 14.077 14.5 14.4 14.3 13.8 13.9 13.6 12.9 12.7 12.3 12.1 12.2 9.7 9.0 85.5 85.6 85.7 86.2 86.1 86.4 87.1 87.3 87.7 87.9 87.8 90.3 91.0 3,546 3,533 3,526 3,522 3,543 3,521 3,543 3,543 3,522 3,520 3,569 3,542 3,520 0 0 4 11 17 24 55 51 51 67 80 175 225 26,000 1 60.000 2 900,120 35 5,200.000 200 5,200,000 200 7,124.260 274 13,520.702 520 15,600,510 600 19,760,000 760 20,800,000 800 21,220,050 816 52,008,350 2,000 62,408.460 2,400 DSMH2 200 20 10,000 9,393 14.5 85.5 3,549 0 cf 157 26,000 1 method fc vc cf DSMHl SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 200 20 200 20 200 20 200 20 200 20 200 20 200 20 200 20 200 20 200 20 200 20 200 20 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 TC/week TCF % TCS defectives detected totalsamp samp/ week % 2 9,557 60,000 14.1 85.9 3,509 0 35 10.083 13.3 86.7 900,090 3,499 3 200 13,341 10.0 90.0 5,200,000 3,485 18 200 13,354 10.1 89.9 5.200,000 3,513 14 14,870 274 9.2 90.8 3,574 19 7.123.360 19,746 6.8 93.2 13.520,455 520 3,513 36 21,342 6.2 93.8 15.600.520 600 3,518 52 24,558 5.5 94.5 3,582 19.760.000 760 73 25,354 5.3 94.7 3,556 58 20.800.000 800 25,653 5.2 94.8 21.220.180 816 3,503 68 49,316 2.6 97.4 3,576 200 52.009.700 2,000 57,294 2.2 97.8 254 3,573 62,409,400 2,400 DSMH2 DSMHl SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 200 200 200 200 200 200 200 200 200 200 200 200 200 100 100 100 100 100 100 100 100 100 100 100 100 100 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10.000 10,000 10,000 9,459 9,735 12,871 29,357 29,363 36,781 61,353 69,348 85,351 89,350 90.944 209,335 249,317 14.3 14.1 10.6 4.6 4.6 3.7 2.2 1.9 1.6 1.5 1.5 0.6 0.5 85.7 85.9 89.4 95.4 95.4 96.3 97.8 98.1 98.4 98.5 98.5 99.4 99.5 3,514 3,580 3,546 3,522 3,541 3,554 3,527 3,529 3,562 3,556 3,503 3,534 3.532 0 0 2 16 17 27 35 54 75 70 74 181 226 1 26,000 2 57,000 900,060 35 5,200,000 200 5,200,000 200 274 7,124,360 13,520,455 520 15,600.540 600 19.760.000 760 20.800.000 800 21,220,050 816 52,008,600 2,000 62,408,460 2,400 DSMH2 SkSPl SINGl DSMHl ChSPl SkSP2 D0UB1 MULT1 ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 2 2 2 2 2 2 2 2 2 2 2 2 2 50,000 6,996 50,000 7,051 50,000 7,364 50,000 7,385 50,000 7,433 50,000 7,449 50,000 7,929 50,000 8,075 50,000 8,364 50,000 8,468 50.000 8,477 50.000 10,626 50.000 11,436 97.1 96.2 91.8 91.4 91.9 89.9 84.4 82.7 79.4 78.7 78.4 60.5 56.3 2.9 3.8 8.2 8.6 8.1 10.1 15.6 17.3 20.6 21.3 21.6 39.5 43.7 3,533 3,527 3,531 3,532 3,566 3,518 3,522 3,521 3,523 3,546 3,522 3,529 3,558 0 1 14 23 13 34 44 50 68 79 67 188 212 1 26,000 900,030 35 5,200.000 200 5.676.311 218 5.200,000 200 7.126.020 274 13.520.572 520 15.600.500 600 19.760.000 760 20,800,000 800 21,220,580 816 52,009.150 2,000 62.408.040 2,400 DSMH2 SkSPl DSMHl ChSPl SING1 5 5 5 5 5 20 20 20 20 20 50.000 50,000 50,000 50,000 50,000 96.9 88.4 77.1 61.7 61.7 3.1 11.6 22.9 38.3 38.3 3,542 3,539 3,524 3,541 3.542 0 2 9 21 18 26.000 900.060 2,346,510 5,200,000 5,200,000 7,032 7,695 8,765 10,969 10,977 158 1 35 90 200 200 method fc vc SkSP2 DOUBl MULTl ChSP2 S1NG2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 20 50,000 20 50,000 20 50,000 20 50,000 20 50,000 20 50,000 20 50.000 20 50,000 TC/week TCF % TCS defectives detected totalsamp samp/ week % 274 12,402 54.2 45.8 7,126,560 3,532 38 17,303 38.7 61.3 13,520,663 520 3,538 53 18.924 35.5 64.5 15,600,530 600 3,549 53 22,108 30.3 69.7 3.554 19,760,000 760 66 22.900 29.3 70.7 20,800,000 800 3,536 52 23,237 28.9 71.1 21,220,130 816 3,567 76 46,640 13.8 86.2 3,540 52,008.950 2,000 195 54,563 11.6 88.4 3,526 221 62.408.360 2,400 DSMH2 SkSPl DSMHl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 5 5 100 100 100 100 100 100 100 100 100 100 100 100 100 7.116 10,399 11,219 26,885 26,906 34,439 58,840 66,823 82,947 86,970 88,403 206,722 246,692 95.8 64.8 60.2 24.9 24.9 19.9 11.3 9.9 8.1 7.8 7.5 3.1 2.6 4.2 35.2 39.8 75.1 75.1 80.1 88.7 90.1 91.9 92.2 92.5 96.9 97.4 3,544 3,505 3,514 3,497 3,508 3,580 3,501 3,501 3,574 3,579 3,487 3,553 3,563 0 2 4 21 21 25 50 59 66 59 62 180 205 1 26,000 35 900,060 43 1,109,758 200 5,200,000 200 5,200,000 274 7,124,500 13.520,650 520 15,600,590 600 19,760,000 760 20,800,000 800 21,220,080 816 52,008,550 2,000 62,408,120 2,400 DSMH2 SkSPl DSMH1 ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 MILST SING2 D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 2 2 2 2 2 2 2 2 2 2 2 2 2 50,000 8,775 50,000 8,883 50,000 9,003 50,000 9.140 50,000 9.194 50,000 9.317 50,000 9,756 50,000 9,916 50,000 10,207 50.000 10,254 50.000 10,304 50,000 12,468 50,000 13,213 77.2 76.6 76.3 73.7 73.9 72.6 68.8 67.7 65.5 64.6 65.0 51.9 48.5 22.8 23.4 23.7 26.3 26.1 27.4 31.2 32.3 34.5 35.4 35.0 48.1 51.5 3,521 3,541 3,577 3,530 3,548 3,542 3,542 3,541 3,548 3,520 3,556 3,528 3,547 0 4 4 26 16 25 51 50 72 78 71 166 214 26,000 1 35 900,120 39 1,007,968 5,200,000 200 5,200,000 200 7,124,500 274 13,520,650 520 15,600,500 600 19,760,000 760 21.220.000 816 20.800.000 800 52.008.000 2,000 62.408.200 2.400 DSMH2 SkSPl DSMHl SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 50 50 50 50 50 50 50 50 50 20 20 20 20 20 20 20 20 20 50,000 8,805 50,000 9.542 50,000 9,615 50,000 12,769 50,000 12,789 50,000 14,205 50,000 19,178 50,000 20,789 50,000 23,870 77.0 71.7 70.8 53.0 53.1 47.3 35.3 32.6 27.9 23.0 28.3 29.2 47.0 46.9 52.7 64.7 67.4 72.1 3,527 3,559 3,545 3,541 3,546 3,518 3,564 3,571 3,535 0 3 5 22 17 24 41 42 68 cf 50.000 50,000 50,000 50,000 50,000 50.000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 159 26.000 900,090 977,280 5,200,000 5,200,000 7,124,320 13,520,520 15,600,420 19,760,000 1 35 38 200 200 274 520 600 760 method fc vc cf SING2 MILST D0UB2 MULT2 50 50 50 50 20 20 20 20 50.000 50.000 50,000 50,000 TC/week TCF % TCS defectives detected totalsamp samp/ % week 24,623 26.9 73.1 3,515 72 20,800,000 800 24,987 26.7 73.3 3,529 65 21,220,380 816 48,410 13.2 86.8 3,509 181 52,008,700 2,000 56,329 11.2 88.8 3,520 234 62,408,840 2,400 DSMH2 DSMH1 SkSPl SING1 ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 50 50 100 100 100 100 100 100 100 100 100 100 100 100 100 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 8,908 11,697 12,262 28,737 28,777 36,160 60.677 68,661 84,699 88,720 90,275 208,518 248,435 76.4 58.0 55.4 23.4 23.5 18.7 11.0 9.7 7.9 7.6 7.4 3.1 2.6 23.6 42.0 44.6 76.6 76.5 81.3 89.0 90.3 92.1 92.4 92.6 96.9 97.4 3,539 3,528 3,533 3,523 3,542 3,536 3,521 3,524 3,545 3,542 3,529 3,548 3,543 0 3 3 21 19 25 52 63 63 49 69 178 216 1 26,000 29 757,090 35 900,065 200 5,200.000 5.200,000 200 274 7,124,500 13,520,676 520 15,600,610 600 19,760,000 760 20,800,000 800 21,220.690 816 52,008.500 2,000 62,408.140 2,400 DSMH2 SkSPl DSMHl SINGl ChSPl SkSP2 DOUBl MULT1 ChSP2 SING2 MILST D0UB2 MULT2 200 200 200 200 200 200 200 200 200 200 200 200 200 2 2 2 2 2 2 2 2 2 2 2 2 2 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 14,815 14,904 15,049 15,143 15,198 15,327 15,748 15,874 16,199 16,277 16,298 18,385 19,172 45.9 45.5 45.2 44.5 44.7 44.1 42.5 42.0 41.2 41.0 40.8 34.7 33.2 54.1 54.5 54.8 55.5 55.3 55.9 57.5 58.0 58.8 59.0 59.2 65.3 66.8 3,539 3,534 3,536 3,530 3,544 3,536 3,529 3,516 3,534 3,539 3,544 3,543 3,529 0 4 1 28 13 23 45 50 65 71 82 228 221 1 26,000 35 900,120 10 260,248 200 5,200,000 200 5,200,000 7,124.080 274 13.520.572 520 15,600,480 600 19,760,000 760 20,800,000 800 21,220,150 816 52,010,550 2,000 62,408,240 2,400 DSMH2 DSMHl SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 200 200 200 200 200 200 200 200 200 200 200 200 200 20 20 20 20 20 20 20 20 20 20 20 20 20 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 14,866 15,088 15,475 18,702 18,743 20,296 25,078 26,670 29,989 30,685 30,962 54,471 62,464 46.0 44.7 43.5 35.8 35.9 33.5 26.6 25.0 22.6 21.8 21.4 11.8 10.3 54.0 55.3 56.5 64.2 64.1 66.5 73.4 75.0 77.4 78.2 78.6 88.2 89.7 3,556 3,512 3,506 3,492 3,525 3.561 3,505 3,507 3,595 3,554 3,506 3,544 3,568 0 1 3 11 23 29 37 43 69 82 58 188 215 26,000 1 257,246 10 900,090 35 5,200,000 200 5,200.000 200 7.125,100 274 13,520,481 520 15,600,430 600 19,760,000 760 20,800,000 800 21,220,000 816 52.009,100 2,000 62,407,860 2,400 160 method fc vc cf DSMH2 DSMHl SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 200 200 200 200 200 200 200 200 200 200 200 200 200 100 100 100 100 100 100 100 100 100 100 100 100 100 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50.000 50.000 50,000 50,000 50,000 50,000 TC/week TCF % TCS defectives detected % 14,856 45.4 15,916 43.1 18,339 37.2 34.731 19.4 34,799 19.5 42,206 16.1 66,755 10.1 74,696 8.9 90,728 7.4 94,743 7.1 96,239 6.9 214,470 3.0 254,413 2.5 54.6 56.9 62.8 80.6 80.5 83.9 89.9 91.1 92.6 92.9 93.1 97.0 97.5 161 3,509 3,569 3,554 3,513 3,555 3,554 3,551 3,543 3,562 3,553 3,512 3,535 3,514 0 1 2 17 24 30 44 67 68 51 73 196 201 totalsamp samp/ week 1 26,000 10 248,236 35 900,055 200 5,200,000 200 5,200,000 274 7,125,400 13,520.572 520 15.600.650 600 19.760.000 760 20.800,000 800 21.220.180 816 52.009.500 2,000 62.407,780 2,400 APPENDDCD COMPARISON OF METHODS AT A 5-SIGMA REOC method fc vc ChSPl DSMH 5 5 2 2 SINGl 5 DSMH 5 1 DOUBl 5 MULTl 5 ChSP2 5 SING2 5 SkSP2 5 D0UB2 5 MILST 5 MULT2 5 SkSPl 5 2 2 2 2 2 2 2 2 2 2 2 TC/week TCF TCS defectives detected totalsamp samp/ ochrs % % week 1,000 7,835 55.2 44.8 20,136 185 10.055 432,366 1.72 1,000 7,873 55.2 44.8 20,094 428,264 185 10.058 1.7 1,000 7,899 55.2 44.8 20.143 185 10,048 429,420 1.72 1,000 7,923 55.1 44.9 20,209 10,065 582,895 251 1.64 1,000 8,127 53.7 46.3 20,262 10,112 1,103,627 475 1.45 1,000 8,210 52.7 47.3 19,982 10,175 1,236,560 545 1.42 1,000 8,455 50.4 49.6 20,035 10.189 1,591,521 689 1.4 1,000 8,536 50.2 49.8 20,007 10.190 1,659,915 724 1.38 9,114 46.0 54.0 19,971 1,000 10,156 1,434,990 613 2.31 1,000 10,790 38.6 61.4 19,917 10,501 4,081,169 1,805 1.33 1,000 10,835 38.3 61.7 19,923 10,524 4,149,858 1,831 1.33 1,000 11,159 36.9 63.1 20,054 10,649 4,938,174 2,164 1.31 1,000 11,288 34.2 65.8 20,074 10,027 284,161 109 7.12 SINGl DSMH 1 ChSPl DSMH 2 DOUBl MULTl ChSP2 SING2 SkSP2 D0UB2 MILST MULT2 SkSPl 5 5 20 20 1,000 1,000 37,384 37,738 11.5 88.5 11.5 88.5 20,129 20,107 10,047 10,041 434,452 429,117 186 185 5 5 20 20 1,000 1,000 37,757 37,792 11.6 88.4 11.5 88.5 20,278 20,120 10,048 10,043 433,863 431,181 186 185 5 5 5 5 5 5 5 5 5 20 20 20 20 20 20 20 20 20 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 40,669 40,688 44,921 44,998 51,348 67,913 68,492 72,808 76,762 10.6 10.4 9.5 9.6 8.3 6.1 6.0 5.7 4.9 89.4 89.6 90.5 90.4 91.7 93.9 94.0 94.3 95.1 20,013 19,910 19,780 20,024 20,098 19,986 19,963 19,979 19,884 10,126 10,185 10,202 10,169 10,153 10,514 10,479 10,602 10,021 DSMH 5 100 1,000 170,124 2.5 97.5 20,012 10,052 5 5 100 1,000 100 1,000 170,759 171.058 2.5 2.5 97.5 97.5 cf 1,084,958 474 1,256,708 546 1,543,728 687 1,649,813 724 1,425,976 613 4,135,254 1,807 4,179,469 1,825 4.876.921 2,160 284.408 110 427.341 2 ChSPl DOUBl MULT1 ChSP2 5 5 5 5 100 100 100 100 1,000 1,000 1,000 1,000 171,861 2.5 97.5 183,540 2.4 97.6 187,813 2.3 97.7 203,291 1 2.1 97.9 19,997 19,995 20,036 20,127 19.987 19.939 162 1.73 1.75 1.75 1.47 1.43 1.41 1.37 2.27 1.31 1.34 1.32 7.15 185 •1 1 SINGl DSMH 1.73 10,042 10,050 429.798 430.079 185 185 10,036 10,123 10,150 10,200 428.336 1.098.138 1,232.912 1.585,612 185 474 545 689 1.71 1.74 1.75 1.75 1.46 1.43 1.41 method fc vc cf SING2 5 SkSP2 D0UB2 MILST MULT2 SkSPl 5 5 5 5 5 100 100 100 100 100 100 1,000 1,000 1,000 1,000 1,000 1,000 DSMH 2 DSMH 1 SINGl ChSPl MULTl DOUBl ChSP2 SING2 SkSP2 D0UB2 MILST MULT2 SkSPl 50 2 1,000 9,846 43.9 56.1 20,134 10,046 433.081 185 50 2 1,000 9,859 43.7 56.3 19.996 10,058 427,679 185 50 50 50 50 50 50 50 50 50 50 50 2 2 2 2 2 2 2 2 2 2 2 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1.000 1.000 9,864 9,908 10,077 10.140 10.537 10,546 11,164 12,772 12,868 13,247 13,309 43.7 43.8 42.0 42.5 40.8 40.7 37.9 32.9 32.8 31.5 29.0 56.3 56.2 58.0 57.5 59.2 59.3 62.1 67.1 67.2 68.5 71.0 20.075 19.976 19.861 19.972 19,959 20,169 20,135 20,231 20,273 20,071 20,082 10,040 10,044 10,152 10,131 10,173 10,214 10,159 10,514 10,541 10,616 10,023 50 50 20 20 1,000 1,000 38,959 39,451 11.0 89.0 11.0 89.0 20,230 20,241 10,050 10,040 50 20 1,000 39,619 10.8 89.2 20,002 10,049 50 50 50 50 50 50 50 50 50 50 20 20 20 20 20 20 20 20 20 20 1,000 1,000 1,000 1,000 1.000 1,000 1,000 1,000 1,000 1,000 39,683 42,688 43,331 46,330 46,688 53,395 70,355 70.849 74.569 79.393 10.9 10.2 10.0 9.3 9.2 7.8 5.8 5.9 5.5 4.9 89.1 89.8 90.0 90.7 90.8 92.2 94.2 94.1 94.5 95.1 20,131 20,132 19,838 20,182 20,090 19,866 19,853. 20,058 19,929 20.186 10,054 10,134 10,146 10,195 10.184 10.163 10.516 10,495 10,580 10,034 ChSPl 50 100 1,000 DSMH 50 100 1,000 170.274 171,960 2.5 2.5 97.5 97.5 20,210 20,228 10,050 10,052 436,151 431,192 DSMH 50 100 1,000 2 SING1 50 100 1,000 DOUBl 50 100 1,000 172,920 2.5 97.5 19,968 10,047 428,580 185 173.186 2.5 97.5 185.817 1 2.3 97.7 19,931 20,160 10.048 10,121 429,023 1,096,936 185 474 SINGl DSMH 1 DSMH 2 ChSPl DOUBl MULTl ChSP2 SING2 SkSP2 D0UB2 MILST MULT2 SkSPl TC/week TCF % 207.838 2.1 240.017 1.8 322.807 1.3 328,025 1.3 345,255 1.2 366,080 1.0 TCS defectives detected % 97.9 20,099 10,193 98.2 20,043 10,178 98.7 20.117 10,524 98.7 19,978 10,538 98.8 20,005 10,637 99.0 20,099 10,041 totalsamp samp/ week 1,646.638 724 1,428,047 613 4,149,149 1,808 4,148,431 1,827 4,949.778 2,164 286.552 109 431,556 185 424,204 185 1,253,250 546 1,083,315 474 1,566,182 688 1,683,419 725 1,439,787 611 4,183,002 1,810 4,213,760 1,829 4,891,734 2,161 283,581 109 441.321 434.573 186 186 ochrs 1.37 2.3 1.33 1.36 1.32 7.21 1.71 1.72 1.75 1.73 1.44 1.45 1.41 1.41 2.35 1.32 1.33 1.31 7.13 1.74 4 11 433.385 186 432.336 185 1,086,263 474 1,221,609 545 1.589.011 689 1.674.940 725 1.422.256 613 4.103,598 1,806 4,147,391 1,827 4,895,726 2,161 284,325 109 186 185 ^ tl 1.71 1.75 1.74 1.46 1.42 1.42 1.38 2.27 1.34 1.33 1.31 7.24 1.73 1.71 163 1.73 1.75 1.45 method fc vc Cf MULTl ChSP2 SING2 SkSP2 D0UB2 MILST MULT2 SkSPl 50 50 50 50 50 50 50 50 100 100 100 100 100 100 100 100 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 SINGl ChSPl DSMH 2 DSMH 1 DOUBl MULTl ChSP2 S1NG2 SkSP2 D0UB2 MILST SkSPl MULT2 200 200 200 2 2 2 1,000 1,000 1,000 16,412 16.472 16.500 26.1 73.9 26.3 73.7 26.1 73.9 20.112 20.146 19,960 10.049 10.053 10.043 436,801 432,811 427,401 186 185 185 200 2 1,000 16,562 26.0 74.0 19,844 10.047 420,494 185 200 200 200 200 200 200 200 200 200 2 2 2 2 2 2 2 2 2 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 16,780 16,795 17,090 17,180 17,725 19,547 19,610 19,842 19,863 25.7 25.5 24.8 24.6 23.6 21.5 21.7 19.5 20.6 74.3 74.5 75.2 75.4 76.4 78.5 78.3 80.5 79.4 20,081 20,040 19,855 19,856 19,915 19,941 20,195 20,096 19,850 10,116 10,152 10,179 10,212 10,157 10,511 10,515 10,027 10,600 SINGl 200 20 DSMH 200 20 1,000 1,000 46,013 46,266 9.4 90.6 9.4 90.6 20,124 20,090 10,047 10,053 434,506 428,608 186 185 1.73 1.73 200 20 1,000 46,398 9.3 90.7 20,124 10,043 431,560 185 1.75 20 20 20 20 20 20 20 20 20 20 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 46,416 48,899 49,573 53,216 53,881 59,727 76,707 77.040 81.362 85.518 9.3 8.9 8.8 8.2 8.0 7.1 5.5 5.4 5.2 4.5 90.7 91.1 91.2 91.8 92.0 92.9 94.5 94.6 94.8 95.5 20,105 20.101 20.045 20.205 19,984 19,975 20,235 19,994 20,177 20,078 10,047 10,103 10,153 10,183 10,185 10,181 10,519 10,478 10,596 10,035 177,375 179,752 2.4 97.6 2.4 97.6 19,971 20,055 10,043 10,041 431,819 431,551 186 185 2.4 97.6 19.861 10,043 425,075 185 TC/week TCF % 186,608 2.3 206,416 2.1 211,428 2.0 238,305 1.7 324.450 1.3 330.110 1.3 345,998 1.2 373.116 1.0 TCS defectives detected totalsamp samp/ % week 97.7 19,933 10.151 1,261,281 546 97.9 20,096 10,193 1,569,406 688 98.0 19,866 10,201 1,646,867 724 98.3 19.933 10,136 1,442,614 613 98.7 20.001 10,500 4.114,234 1.806 98.7 19,940 10.525 4,124,949 1,831 98.8 20,208 10.635 4.984,691 2,165 99.0 19,746 10.024 110 283,125 ochrs 1.4 1.39 1.41 2.29 1.31 1.34 1.31 7.13 1.73 1.72 1.73 1,094.597 474 1.259.163 546 1.572.601 688 1.654.707 724 1.434.333 614 4.054.373 1,803 4.164.184 1,828 284,523 109 4,889,451 2,161 1.72 1.46 1.44 1.39 1.39 2.27 1.33 1.33 7.13 1.33 4 1 DSMH 2 ChSPl DOUBl MULT1 ChSP2 SING2 SkSP2 D0UB2 MILST MULT2 SkSPl 200 200 200 200 200 200 200 200 200 200 ChSPl 200 100 1,000 DSMH 200 100 1,000 2 DSMH 200 100 1,000 1 181,255 185 429,763 1,093,657 474 1,235,532 545 1,582,631 688 1,652,163 724 1,419,946 614 4,163,846 1,809 4,178,059 1,823 4,902,569 2,161 286,825 110 1.74 1.46 1.4 1.42 1.39 2.24 1.33 1.34 1.34 7.24 1.71 1.76 164 1.74 method fc vc Cf SINGl DOUBl MULT1 ChSP2 SING2 SkSP2 D0UB2 MILST MULT2 SkSPl 200 200 200 200 200 200 200 200 200 200 100 100 100 100 100 100 100 100 100 100 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 DSMH 2 SkSPl ChSPl SkSP2 DOUBl SINGl S1NG2 DSMH 1 MULTl ChSP2 MULT2 MILST D0UB2 5 2 10,000 41,319 5 5 5 5 5 5 5 2 2 2 2 2 2 2 5 5 5 5 5 2 2 2 2 2 DSMH 2 SINGl ChSPl DSMH 5 TC/week TCF TCS defectives detected totalsamp samp/ week % % 185 182,613 2.3 97.7 19,787 421,711 10.047 192,274 2.3 97.7 20,089 10.121 1,090,517 474 196,620 2.2 97.8 20,045 10.145 1,228,968 545 213,637 2.0 98.0 20.057 10.190 1,566,562 688 217,538 2.0 98.0 20.060 10.198 1,646,151 724 247,097 1.7 98.3 19.901 10,156 1,421,429 613 331,145 1.2 98.8 19.929 10,499 4,144,430 1.807 332,665 1.3 98.7 20.271 10,516 4,195,313 1.818 354,802 1.2 98.8 19.870 10,619 4,858,552 2.160 109 376,048 1.0 99.0 20.231 10,036 286,425 ochrs 1.74 1.46 1.43 1.4 1.41 2.26 1.32 1.33 1.32 7.24 19.8 80.2 19,914 9,986 12.127 1 10,000 10,000 10,000 10,000 10,000 10,000 10,000 46,252 83.7 16.3 46,675 92.4 7.6 46,791 89.5 10.5 46,828 91.9 8.1 46,900 92.4 7.6 46,924 90.9 9.1 47,112 92.2 7.8 19,959 20,070 19,859 20,150 19,968 19,933 20.129 10,031 10,040 10,176 10,149 10,051 10,184 10,097 281,208 431,234 1,421,368 1,102,937 423,863 1,655,180 842,989 110 185 615 475 185 724 365 10,000 10,000 10,000 10,000 10,000 47,596 47,745 48,096 48,109 48,562 8.1 8.9 14.6 14.0 13.6 20,172 20,090 19,923 19,806 20,039 10,139 10,171 10,591 10,502 10,518 1,251,886 546 1,568,524 688 4,912,797 2.162 4,128.515 1,836 4.098.104 1,805 20 10,000 75,897 55.9 44.1 19,892 10,045 430.368 185 20 10,000 76,391 20 10,000 76,743 20 10,000 77,041 56.3 43.7 55.8 44.2 56.0 44.0 20,074 19,883 19,924 10,051 10,043 10,062 431,971 425,280 618,682 185 185 271 10,135 10,153 10,170 10,181 10,173 10,531 10,550 10,032 10,598 1,115,530 1,269,507 1,563,286 1,655,770 1,443,664 4,088,372 4,190,225 280,722 4,947,719 475 546 688 724 611 1,806 1,831 110 2,163 1.59 1.45 1.43 1.39 1.4 2.3 1.31 1.35 6.89 1.31 10,052 10,055 429.833 431.119 185 185 1.74 1.73 112.26 7.09 1.72 2.25 1.46 1.72 1.38 1 52 1 5 5 5 91.9 91.1 85.4 86.0 86.4 4 1 DOUBl MULTl ChSP2 SING2 SkSP2 D0UB2 MILST SkSPl MULT2 5 5 5 5 5 5 5 5 5 20 20 20 20 20 20 20 20 20 45.1 45.8 48.2 48.7 52.6 60.8 61.2 65.5 61.8 20,291 20,143 19,997 20,049 20,117 19,880 19,944 19,745 20,243 ChSPl SINGl 5 5 100 10,000 208.798 20.5 79.5 100 10,000 209,856 20.7 79.3 19,959 20.172 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 78,832 79,317 83,524 84,164 88,662 105.291 105,952 110.253 110.459 54.9 54.2 51.8 51.3 47.4 39.2 38.8 34.5 38.2 165 mXj^ 1.45 1.41 1.29 1.35 1.33 1.74 1.73 1.75 method fc DSMH 2 DSMH 1 DOUBl MULT1 ChSP2 SING2 SkSP2 D0UB2 MILST MULT2 SkSPl 5 TC/week TCF TCS defectives detected totalsamp samp/ week % % 221 100 10,000 210.602 20.2 79.8 19,806 506,160 10,053 5 100 10,000 211.131 20.6 79.4 20,004 10,052 5 5 5 5 5 5 5 5 5 100 100 100 100 100 100 100 100 100 80.6 80.4 82.3 82.3 84.5 88.5 88.5 89.3 90.5 20,087 20,405 19,979 20,130 20,196 20,060 20,004 19,793 19,987 10,131 10,153 10,166 10,203 10,180 10,494 10,536 10,551 10,030 DSMH 2 SkSPl SINGl DSMH 1 MULTl SkSP2 SING2 ChSPl ChSP2 MILST DOUBl D0UB2 MULT2 50 2 10,000 43,102 19.1 80.9 19,920 9,970 12.107 1 50 50 50 2 2 2 10,000 47,379 10,000 48,303 10,000 48,468 80.3 19.7 88.4 11.6 85.4 14.6 20,024 19,821 19,932 10,026 10,042 10,076 286,345 424,066 721,541 109 185 303 50 50 50 50 50 50 50 50 50 2 2 2 2 2 2 2 2 2 vc cf 1.67 DSMH 50 2 ChSPl 50 SINGl 50 DSMH 50 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 221,687 223,043 242,017 247.328 277.971 360.788 368,020 383,202 400,446 19.4 19.6 17.7 17.7 15.5 11.5 11.5 10.7 9.5 506,544 221 1,099.013 475 1.282,908 547 1.575.905 688 1,644,197 723 1,424,170 611 4,150,335 1,807 4,125,771 1,842 4,881,135 2,160 109 284,826 50 50 50 50 50 50 50 50 50 1.66 1.46 1.44 1.4 1.38 2.27 1.34 1.33 1.34 7.15 112.61 7.13 1.76 25 48,509 87.9 12.1 48,950 85.9 14.1 49,048 87.1 12.9 49,252 88.7 11.3 49,273 87.4 12.6 49,339 82.4 17.6 49,533 88.2 11.8 49.812 82.7 17.3 50.919 82.2 17.8 19,994 19,802 19,865 20,201 20,057 19.724 20,034 19,893 20,252 10,146 10.126 10.176 10,048 10,160 10,506 10,123 10,519 10,614 20 10,000 77.663 55.2 44.8 20,247 10,049 20 10,000 78,170 20 10,000 78,417 20 10,000 78,970 54.8 45.2 54.8 45.2 54.6 45.4 20,021 20,007 20,004 10,050 10,044 10,062 46.3 47.1 49.3 49.7 53.4 61.4 62.1 62.6 65.7 20,316 19.972 20.097 20.165 20.102 19.995 19,810 20,357 20,079 10,116 10,169 10,198 10,218 10,149 10,496 10,524 10,600 10,022 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 1,260,317 546 1,414,807 615 1,640,126 723 430,620 185 1,582,648 689 4,144,959 1,828 1,074,257 473 4,109,213 1,806 4,985,186 2,165 442,044 431,647 429,468 579,470 20 20 20 20 20 20 20 20 20 10,000 10,000 10.000 10,000 10,000 10,000 10,000 10,000 10.000 81,679 82,142 84,939 85,668 91,480 107,403 108,630 112,421 113,183 53.7 52.9 50.7 50.3 46.6 38.6 37.9 37.4 34.3 166 1.45 2.24 1.4 • 1.75 1.41 1.33 1.46 1.32 1.33 186 185 185 251 A 1 D0UB1 MULTl ChSP2 SING2 SkSP2 D0UB2 MILST MULT2 SkSPl ochrs 1,104,122 475 1,229,617 545 1,583,768 689 1,674.536 725 1.430,459 612 4,138.419 1,807 4.129,595 1,833 5,024,740 2,167 282,979 109 1.74 1.74 1.73 1.64 1.46 1.41 1.41 1.4 2.28 1.33 1.35 1.33 7.06 method fc vc cf TC/week TCF TCS defectives detected totalsamp samp/ week % % ChSPl 50 100 10,000 209,919 20.6 79.4 20,136 185 432.366 10,055 DSMH 50 100 10,000 210,016 20.4 79.6 19,985 222 10,068 514.253 1 DSMH 50 100 10,000 211,733 20.5 79.5 20,044 221 510,921 10,053 2 SING1 50 100 10,000 214,646 20.2 79.8 19,987 185 423,948 10,047 DOUBl 50 100 10,000 223,618 19.4 80.6 20,047 10.120 1,087,703 474 MULT1 50 100 10,000 228,578 19.0 81.0 20,042 10.144 1,244.777 545 ChSP2 50 100 10,000 243,444 17.5 82.5 20,035 10.189 1.591.521 689 SING2 50 100 10,000 246,472 17.7 82.3 20,247 10.163 1.676.184 725 SkSP2 50 100 10,000 279,032 15.1 84.9 19,976 10,179 1.432,316 617 D0UB2 50 100 10,000 363.130 11.3 88.7 19,849 10.511 4,095,757 1,806 MILST 50 100 10,000 366.939 11.3 88.7 19,923 10,524 4,149,858 1,831 MULT2 50 100 10,000 384,910 10.7 89.3 19.942 10,598 4,916,781 2.162 109 50 100 10,000 400,218 284,932 SkSPl 9.5 90.5 19.969 10,043 ochrs 1.72 1.66 1.67 1.76 1.42 1.44 1.4 1.38 2.27 1.32 1.33 1.32 7.11 200 2 10,000 49,368 16.9 83.1 20.115 9,973 12,137 1 113.02 200 2 10,000 50,014 24.5 75.5 19,954 10,028 282,791 35 65.61 200 200 200 200 200 200 200 200 200 200 200 2 2 2 2 2 2 2 2 2 2 2 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10.000 77.8 71.2 77.5 78.1 77.4 76.8 75.7 76.9 72.8 73.3 72.5 22.2 28.8 22.5 21.9 22.6 23.2 24.3 23.1 27.2 26.7 27.5 19,923 20,294 20,007 20,187 19,961 20,225 20,074 19,878 19,891 20,207 19,980 10,041 10.027 10.126 10,048 10,159 10,211 10,152 10,187 10,523 10,496 10,553 DSMH 200 20 10.000 84,835 50.2 49.8 19,830 10,057 507.739 221 1.64 2 SINGl 200 20 10.000 85,046 DSMH 200 20 10,000 85.501 50.1 49.9 50.1 49.9 19,817 19,768 10,038 10,067 425.328 537.995 185 237 1.75 1.63 49.4 51.1 51.0 53.5 53.1 56.9 63.8 63.4 68.2 19,973 20,049 20,192 19,801 20,221 20,082 19,970 19,970 19,796 10,042 10,136 10,171 10,180 10,199 10,174 10,496 10,477 10,031 425.208 1.098,062 1,270,130 1,546,494 1,676,809 1,446,424 4,188,133 4,080,212 284,886 185 474 547 687 725 612 1.829 1.804 110 1.73 1.46 1.42 1.4 1.38 2.3 1.33 1.31 7.03 DSMH 2 DSMH 1 SINGl SkSPl DOUBl ChSPl MULTl SING2 SkSP2 ChSP2 MILST D0UB2 MULT2 55,095 55,244 55.359 55.359 55.454 55,563 55,772 56,013 56,744 57,240 57,268 185 427,132 110 285.936 1.091.972 474 435.148 186 1.245,553 546 1.702.631 726 1.441,034 613 1,546,246 687 4,163,345 1.836 4,184,902 1.809 4,907.841 2.162 1.75 7.02 1.46 1.73 1.42 1.39 2.28 1.41 1.35 1.33 1.31 A 1 ChSPl DOUBl MULTl ChSP2 SING2 SkSP2 MILST D0UB2 SkSPl 200 200 200 200 200 200 200 200 200 20 20 20 20 20 20 20 20 20 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 85,545 87,531 87,935 91,898 92,486 97,150 114.392 114,759 118,199 50.6 48.9 49.0 46.5 46.9 43.1 36.2 36.6 31.8 167 method fc vc cf TC/week TCF TCS defectives detected totalsamp samp/ % week % MULT2 200 20 10.000 119,083 35.0 65.0 20,032 10,611 4,878,665 2,160 DSMH o SINGl ChSPl DSMH ochrs 1.33 200 100 10,000 217,438 19.8 80.2 20,007 10,055 510,712 221 1.64 200 100 10.000 218.256 19.9 80.1 200 100 10,000 218,319 19.9 80.1 200 100 10,000 219,005 20.0 80.0 20,052 20,085 20,147 10,041 10,044 10,065 427,892 427,664 509,458 185 185 221 1.73 1.72 1.67 DOUBl MULTl ChSP2 SING2 SkSP2 D0UB2 MILST MULT2 SkSPl 200 200 200 200 200 200 200 200 200 100 100 100 100 100 100 100 100 100 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 231,160 236,144 251,290 255,360 283.179 369.838 372,590 390,687 406,052 18.7 18.6 17.2 16.4 14.8 11.3 11.1 10.5 9.4 81.3 81.4 82.8 83.6 85.2 88.7 88.9 89.5 90.6 20.098 20.051 20.111 19,704 19,891 20,008 19,952 20,031 19,921 10,126 10,136 10,184 10,192 10,159 10,504 10,511 10,628 10,037 1,096,051 474 1,231,121 545 1,585,279 689 1.641,887 723 1,426,365 614 4,117,617 1,806 4,172,882 1,829 4,946,968 2,164 109 282,258 1.46 1.44 1.4 1.4 2.23 1.32 1.33 1.31 7.05 SkSPl MULT2 SkSP2 MULTl D0UB2 ChSPl DSMH 2 D0UB1 DSMH 1 SING1 MILST ChSP2 SING2 5 5 5 5 5 5 5 2 2 2 2 2 2 2 50,000 50,000 50,000 50,000 50,000 50,000 50,000 200,286 214.152 215,438 215,798 216,318 217,445 218,082 96.3 96.7 97.7 98.2 97.0 98.3 98.4 3.7 3.3 2.3 1.8 3.0 1.7 1.6 20,005 20,087 20,002 19,956 20,066 19,814 20,158 10,035 10,600 10,151 10,128 10,495 10,043 10,049 283,162 110 4,954,803 2,163 1,434,407 613 1,266,070 546 4,120,405 1,806 423,144 185 353,471 150 7.02 1.32 2.26 1.44 1.31 1.74 1.83 5 5 2 2 50,000 218.339 98.3 50,000 218.842 98.3 1.7 1.7 20,059 19,809 10,138 10.095 1,096,897 824,080 474 365 1.47 1.5 5 5 5 5 2 2 2 2 50,000 50,000 50,000 50,000 98.4 1.6 97.0 3.0 98.1 1.9 98.1 1.9 19,953 20,301 20,059 20,103 10,048 10,528 10,190 10,169 425,798 4,212.703 1.571.092 1.661.512 185 1,833 688 725 1.75 1.32 1.42 1.4 DSMH 5 20 50,000 245.632 86.5 13.5 20,026 10,051 435.317 186 1.73 5 20 50,000 248.463 86.1 13.9 19,989 10,110 834.657 362 1.52 5 5 5 5 5 5 5 5 20 20 20 20 20 20 20 20 13.3 14.3 15.9 13.0 14.6 16.1 18.0 27.4 20,171 20,070 19,884 20,492 19,981 19,812 20,126 19,881 10.043 10,123 10.197 10,052 10.163 10.168 10,163 10,033 185 432.700 1,093,343 474 1,566,322 688 439,479 186 1,232,777 545 1,612,873 722 1,435,211 613 283,129 1 110 1.74 1.46 1.41 1.74 1.41 1.38 2.27 6.98 1 1 2 DSMH 1 SING1 D0UB1 ChSP2 ChSPl MULTl SING2 SkSP2 SkSPl 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 219.013 219.216 220.372 220.880 250.414 251,701 253,007 253,599 254,089 257,230 259,639 262,744 86.7 85.7 84.1 87.0 85.4 83.9 82.0 72.6 168 method fc vc MULT2 D0UB2 MILST 5 5 5 20 50,000 270,936 74.8 25.2 20 50,000 271,227 76.3 23.7 20 50,000 272,617 76.2 23.8 19,916 19,897 20,075 10,600 10,523 10,573 DSMH 5 100 50,000 380,077 56.1 43.9 20,029 10,093 668,241 287 1.59 SINGl DSMH 2 ChSPl D0UB1 MULTl ChSP2 SING2 SkSP2 D0UB2 MILST MULT2 SkSPl 5 5 100 50,000 382,004 56.9 43.1 100 50,000 383,334 56.2 43.8 20,212 19,916 10,041 10,051 433,935 506,935 185 221 1.72 1.68 5 5 5 5 5 5 5 5 5 5 100 100 100 100 100 100 100 100 100 100 43.6 45.8 45.6 47.9 48.6 52.9 60.7 60.7 62.2 65.2 20,067 19,919 20,129 20,094 19,984 19,913 19,927 20,004 20,064 20,067 10,051 10,120 10,147 10,150 10,196 10,178 10,497 10,496 10,582 10,031 DSMH 2 SkSPl MILST MULT2 SkSP2 D0UB2 D0UB1 ChSPl MULTl ChSP2 DSMH 50 2 50,000 77,006 54.7 45.3 20,164 9,984 50 50 50 50 50 50 50 50 50 50 2 2 2 2 2 2 2 2 2 2 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50.000 50.000 95.3 96.0 95.8 96.8 96.1 97.3 97.4 97.3 97.2 97.4 4.7 4.0 4.2 3.2 3.9 2.7 2.6 2.7 2.8 2.6 19,809 19,905 20,035 20,020 20,164 19,714 19,797 19,964 19,982 19.910 10,035 10.496 10,632 10,162 10.546 10.145 10,041 10,144 10,177 10,099 110 283,448 4,158,962 1.827 4,920,712 2,163 1,445,828 613 4,167,267 1,809 1,065,874 473 419,649 185 1,246,032 545 1,570,798 688 831.465 365 1 SINGl 50 SING2 50 2 2 50,000 223.902 97.5 2.5 50,000 224,154 97.2 2.8 20,150 20,155 10,045 10,202 429,066 1,653,987 185 724 1.76 1.41 50 50 50 50 20 20 20 20 50,000 50,000 50,000 50,000 14.4 14.1 15.3 15.0 19,775 19,930 19,898 19,891 10,038 10,058 10,152 10,025 425,289 429,729 1,252.302 260.280 185 185 546 113 1.72 1.71 1.42 2.02 20 50,000 253,780 83.4 16.6 20 50,000 253.942 85.6 14.4 19,837 20,080 10,189 10,105 1.569,206 815,280 688 355 1.39 1.52 20 50,000 255.041 85.2 14.8 20 50,000 255,783 83.2 16.8 20,211 19,823 10,119 10,184 1,102,486 1,637,827 475 723 1.46 1.37 cf % 1 samp/ week 4,977,339 2,164 4,088,493 1,805 4,192,257 1,834 TC/week TCF TCS defectives detected totalsamp % ochrs 1.33 1.33 1.33 1 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 383,536 392,007 401,142 416,256 419,123 446,322 526.070 532,656 550,107 552,644 198,688 215,345 215,690 215,722 217,337 218,333 220,748 220,795 220,985 221.168 56.4 54.2 54.4 52.1 51.4 47.1 39.3 39.3 37.8 34.8 185 429,109 1,094,543 474 1,246,558 545 1,577,664 689 1.642.941 724 1.425.059 616 4.116,388 1,806 4,146,900 1,825 4,922,489 2,162 109 285,147 12,093 1 1.74 1.47 1.45 1.41 1.36 2.27 1.33 1.35 1.32 7.14 111.84 7.01 1.32 1.31 2.27 1.3 1.47 1.75 1.43 1.41 1.5 A SINGl ChSPl MULT1 DSMH 2 ChSP2 50 DSMH 50 A 1 DOUBl 50 SING2 50 247,780 248,066 250,855 251,292 85.6 85.9 84.7 85.0 169 method fc vc Cf SkSP2 SkSPl D0UB2 MILST MULT2 50 50 50 50 50 20 20 20 20 20 50,000 50,000 50,000 50,000 50,000 SINGl DSMH 1 DSMH 2 ChSPl DOUBl MULTl SING2 ChSP2 SkSP2 D0UB2 MILST MULT2 SkSPl 50 100 50,000 382.371 56.6 43.4 50 100 50,000 385,047 56.0 44.0 20,187 20,097 10,049 10,079 434.352 654.470 186 282 1.71 1.6 50 100 50,000 385,972 56.0 44.0 19,991 10,046 508,895 221 1.68 50 50 50 50 50 50 50 50 50 50 100 100 100 100 100 100 100 100 100 100 44.1 45.6 46.5 49.2 48.3 52.6 61.0 61.2 62.1 65.1 19,897 20,079 19,886 19,924 19,998 20.017 19,821 19,876 20,215 20,191 10,044 10,128 10,154 10.197 10.177 10.158 10,489 10,518 10,570 10,038 DSMH 2 DSMH 1 SkSPl MULT2 SkSP2 ChSPl MILST D0UB2 DOUBl MULTl SINGl SING2 ChSP2 200 2 50,000 82,484 50.3 49.7 20,058 9,979 12,154 1 112.06 200 2 50,000 200,400 91.4 8.6 19,893 10,098 784,233 293 5.19 200 200 200 200 200 200 200 200 200 200 200 2 2 2 2 2 2 2 2 2 2 2 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 7.6 7.1 6.1 5.4 6.8 6.8 5.5 5.5 5.4 5.6 5.6 19,991 20,027 19,797 19,817 20,108 19,888 19,828 20,093 19,970 20,159 20,272 10,030 10,568 10,157 10,059 10,546 10,472 10,135 10,157 10,036 10,169 10,151 DSMH 2 SINGl ChSPl DOUBl DSMH 1 MULTl 200 20 50,000 255.076 82.9 17.1 19,877 10,033 262.656 113 1.98 16.3 16.5 17.3 16.6 20,086 19,929 20,080 20,091 10,055 10,046 10,133 10,073 432,472 423,897 1,103,279 739,213 185 185 475 320 1.74 1.74 1.47 1.59 200 20 50,000 263,044 82.6 17.4 19,925 10,154 1,224,878 544 1.43 TC/week TCF % 260,532 81.0 268,122 72.0 274.395 75.9 275.575 75.9 279.093 74.8 TCS defectives detected % 19.0 19,958 10.170 28.0 19,892 10.030 24.1 19,953 10.513 24.1 20,085 10.509 25.2 20,107 10.594 totalsamp samp/ week 1,428,652 616 110 282,019 4,091,757 1,805 4,177.993 1,824 4.927.333 2,163 ochrs 2.29 7.02 1.31 1.33 1.3 Cm 200 200 200 200 20 20 20 20 50,000 50,000 50,000 50,000 50,000 50.000 50.000 50.000 50.000 50.000 50,000 50,000 50,000 50,000 387,178 397,945 398,645 418,659 419,742 443.451 528,319 534,207 550,035 555,502 210,373 222,080 222,533 224,203 224,618 224,788 225,582 227,404 227,972 230.768 231.140 257.057 258.383 258,611 260,291 55.9 54.4 53.5 50.8 51.7 47.4 39.0 38.8 37.9 34.9 92.4 92.9 93.9 94.6 93.2 93.2 94.5 94.5 94.6 94.4 94.4 83.7 83.5 82.7 83.4 170 421,250 185 1,089,848 474 1,245,280 545 1,654,122 724 1,556,225 688 1,433,480 611 4,084,832 1,805 4,147,830 1,836 5,011.995 2,166 286,745 109 281.641 110 4.958,558 2,164 1,427,034 619 426,412 185 4,196,945 1,837 4,052,870 1,803 1,077,395 474 1,262,437 546 426,509 185 1,660,624 724 1.597,545 689 1.73 1.45 1.45 1.39 1.4 2.27 1.34 1.35 1.32 7.15 6.98 1.3 2.24 1.74 1.35 1.31 1.43 1.43 1.74 1.38 1.4 method fc vc cf SkSP2 SING2 ChSP2 SkSPl D0UB2 MILST MULT2 200 200 200 200 200 200 200 20 20 20 20 20 20 20 50,000 50,000 50,000 50,000 50.000 50.000 50.000 ChSPl DSMH 2 SINGl DSMH 11 MULTl DOUBl ChSP2 SING2 SkSP2 D0UB2 MILST SkSPl MULT2 200 100 50,000 388,761 54.7 45.3 200 100 50,000 389,791 55.1 44.9 19,892 19,942 200 100 50,000 391,264 54.7 45.3 200 100 50,000 391,751 54.8 45.2 200 100 200 100 200 100 200 100 200 100 200 100 200 100 200 100 2001100 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 TC/week TCF % 263,090 79.0 265,193 81.4 267,104 81.6 274.117 70.6 281,275 74.2 281,404 74.0 285,821 73.0 403.707 408,159 422,074 428,645 454,510 534,972 542,396 557,987 561,582 52.5 53.6 50.9 50.1 46.5 38.5 38.7 33.8 37.5 ochrs 1,443,736 1,649,798 1,540,325 283,351 4,125,060 4,172,071 4,891,787 samp/ week 612 724 687 109 1,807 1,824 2,161 10,059 10,049 428,027 510.197 185 221 1.72 1.66 19,804 20,000 10,063 10,079 421,584 628,829 185 272 1.71 1.62. 19,735 20,120 20,088 19,935 20,070 19,762 19,974 19,723 20,068 10,164 10,129 10,174 10,210 10.166 10.475 10.515 10.039 10,567 TCS defectives detected % 21.0 19,986 10,177 18.6 20,049 10,203 18.4 19,940 10,171 29.4 20,065 10,026 25.8 20,059 10,529 26.0 20,052 10,529 27.0 20,083 10,640 47.5 46.4 49.1 49.9 53.5 61.5 61.3 66.2 62.5 171 totalsamp 1,231,057 545 1,082,503 474 1,590.116 689 1.637.519 723 1.440,785 615 4.065,261 1.803 4,112,301 1.826 282,648 110 4,872,790 2.160 2.27 1.39 1.39 7.04 1.33 1.34 1.3 1.44 1.45 1.39 1.41 2.32 1.33 1.35 6.96 1.31 APPENDDC E COMPARISON OF METHODS AT A 6-SIGMA - REOC DSMH 2 DSMH 11 5 2 TC/week TCF TCS defectives detected % % 1.000 443 14.9 85.1 1,614 824 5 2 1,000 455 SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 5 5 5 5 5 5 2 2 2 2 2 2 2 2 2 2 2 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 DSMH 1 DSMH 2 SkSPl SING1 ChSPl SkSP2 D0UB1 MULTl ChSP2 SING2 MILST D0UB2 MULT2 5 20 1,000 5 5 5 5 5 5 5 5 5 5 5 20 20 20 20 20 20 20 20 20 20 20 1,000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 3,098 4,728 4,734 6,892 11,062 12,636 15,870 16,630 17,344 40,631 48,573 DSMH 2 DSMH 1 SkSPl ChSPl SINGl SkSP2 5 100 1,000 7,655 5 100 1,000 7,759 0.9 99.1 1,638 5 5 5 5 100 100 100 100 1,000 1,000 1,000 1,000 14,828 22,614 22,696 33,741 0.5 0.3 0.3 0.2 99.5 99.7 99.7 99.8 1,678 1,694 1,665 1,683 cf totalsamp 239,834 samp/ ochrs week 5.66 CM Cm 1,622 831 549 711 713 938 1,350 1,507 1,829 1,909 1,972 4,303 5,100 11.9 9.1 9.2 7.5 5.1 4.4 3.7 3.7 3.3 1.5 1.3 88.1 90.9 90.8 92.5 94.9 95.6 96.3 96.3 96.7 98.5 98.7 1,626 1,617 1,625 1,715 1,700 1,641 1,644 1,718 1,625 1,695 1,741 834 829 832 839 854 854 853 852 843 887 904 1,748 3.8 96.2 1,637 813 5.42 359,457 CO CO 14.4 85.6 o to vc o method fc 435,431 2,419,868 200 2,413,872 200 3,480,097 280 6,365,656 519 7,101,225 599 8,860,385 759 9,771,113 799 10,069,626 833 24,657,596 1,997 29,530,866 2,396 13.09 1.75 1.73 5.53 1.45 1.48 1.41 1.4 1.38 1.32 1.37 2.78 504,385 41 5 20 1,000 1,783 3.9 96.1 1,672 839 2.3 97.7 1.4 98.6 1.4 98.6 1.0 99.0 0.6 99.4 0.5 99.5 0.4 99.6 0.4 99.6 0.4 99.6 0.2 99.8 0.1 99.9 1,697 1,641 1,650 1,685 1,665 1,658 1,654 1,746 1,628 1,670 1,623 829 835 836 829 839 841 844 829 858 871 905 0.9 1,624 814 99.1 3.03 482.778 40 438.784 36 2,418,069 200 2,386,897 200 3,410,847 281 6,407,875 519 7,232,844 599 8,968,118 759 9,987,445 799 10.051,772 833 24,447,150 1.997 28.890,560 2.396 488,817 13.3 1.75 1.74 5.18 1.44 1.36 1.4 1.34 1.4 1.3 1.29 2.95 40 172 829 832 835 852 832 487,236 440,678 2,415,874 2,353,365 3,370,147 2.9 40 *tVJ 36 200 200 281 13.84 1.75 1.74 5.52 method fc DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 DSMH 1 DSMH 2 SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 S1NG2 MILST D0UB2 MULT2 DSMH 1 DSMH 2 SkSPl SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2: MULT2 TCS (defectives detected totalsamp samp/ week % 6,298,114 841 99.9 1,651 519 7,022,367 859 1,646 99.9 599 9,131,826 759 843 1,637 99.9 9,647,668 799 1,621 850 99.9 10,058,745 833 848 1,718 99.9 23,641.938 1,997 864 1,617 100 29.257.238 2,396 896 100 1,623 vc cf 5 •100 5 5 5 5 5 5 100 100 100 100 100 100 1.000 1.000 1.000 1.000 1.000 1.000 1,000 50 2 1,000 2,250 3.0 97.0 50 2 1,000 2,252 3.0 97.0 50 50 50 50 50 50 50 50 50 50 50 2 2 2 2 2 2 2 2 2 2 2 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 2,369 2,514 2,517 2,744 3,148 3,308 3,629 3,705 3,781 6,105 6,900 2.8 2.6 2.8 2.4 2.0 2.0 1.9 1.7 1.7 1.0 0.9 97.2 97.4 97.2 97.6 98.0 98.0 98.1 98.3 98.3 99.0 99.1 1.652 1,626 1,683 1,649 1,595 1,647 1,673 1,631 1,644 1,642 1,671 845 829 816 844 845 849 858 842 864 891 893 50 20 1,000 3,516 2.0 98.0 1,683 812 502,961 1,000 3,583 2.0 98.0 1,725 854 487,717 TC/week TCF % 54,251 0.1 62,283 0.1 78,193 0.1 82,160 0.1 85,693 0.1 202,107 0.0 241,806 0.0 1,676 831 ochrs 1.41 1.37 1.45 1.33 1.46 1.26 1.29 5.84 249,552 20 1.649 837 20 430,890 36 2,459,530 200 2,493,945 200 3,383,912 281 6,138,163 519 7,236.738 599 9.192.773 759 9.853.125 799 9.898,040 834 23.742,529 1,997 28,887,001 2,396 13.46 1.7 1.75 5.54 1.44 1.35 1.31 1.4 1.4 1.28 1.27 2.92 40 50 20 50 50 50 50 50 50 50 50 50 50 50 20 1,000 20 1,000 20 1,000 20 1,000 20 1,000 20 1,000 20 1,000 20 1,000 20 1,000 20 1,000 20 1,000 4,947 6,514 6,550 8,725 12,867 14,471 17,676 18.463 19,174 42,445 50,355 1.3 1.0 1.0 0.8 0.5 0.5 0.4 0.4 0.3 0.2 0.1 98.7 99.0 99.0 99.2 99.5 99.5 99.6 99.6 99.7 99.8 99.9 1,655 1,643 1,656 1,708 1,642 1.642 1,665 1,646 1,612 1.731 1.696 839 824 853 841 834 857 853 850 855 885 861 DSMH 50 100 1,000 1 DSMH 50 100 1,000 2 SkSPl 50 IOC 1,000 SINGl 50 100 1.000 9.514 0.7 99.3 1.664 835 9,530 0.7 99.3 1,631 829 481,813 17,106 24,299 0.4 0.3 99.6 99.7 1,636 1,681 852 822 419,620 2,453,443 173 5.77 241,110 2.95 40 438,638 36 2,479.545 200 2.400.643 200 3.443.806 281 6.304,435 519 7,128,665 599 9,005,830 759 9,626.867 799 9.790.572 834 24.461.461 1,997 30,074,048 2,396 13.3 1.72 1.72 5.31 1.41 1.47 1.43 1.41 1.4 1.32 1.28 2.96 486,691 40 2.91 40 36 13.28 200 1 1-7 method fc vc cf ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 100 100 100 100 100 100 100 100 100 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 DSMH 2 DSMH 1 SkSPl ChSPl SING1 SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 200 2 1,000 TC/week TCF % 24,396 0.3 35,604 0.2 55,990 0.1 64.107 0.1 79.964 0.1 83.971 0.1 87.342 0.1 203.932 0.0 243.608 0.0 8.233 0.8 TCS defectives detected totalsamp sampy % week 99.7 1,669 840 2.418,198 200 99.8 1,633 843 3.355,933 281 99.9 1,731 843 6,440,645 519 99.9 1,617 852 7,023,456 599 99.9 1,634 840 9,390,771 759 99.9 1,749 837 10,027,654 799 99.9 1,681 858 10,217,176 833 100 1,642 865 24,177,078 1,997 100 1,690 878 29,891,001 2,396 ochrs 99.2 3.01 1,646 843 485,126 1.75 5.4 1.44 1.47 1.4 1.49 1.38 1.31 1.37 A f\ 200 2 1,000 8.235 0.9 99.1 1,720 837 492,821 40 2.9 A f\ 200 200 200 200 200 200 200 200 200 200 200 40 451,074 13.5 36 2,398,954 200 1.71 2,376,900 200 1.72 3.347,215 281 5.32 6,147,099 519 1.47 7,361,855 599 1.41 9,173,643 759 1.32 9,662,692 799 1.34 9,815,509 834 1.33 24,067,410 1,997 1.27 28,506,908 2,396 1.26 2 2 2 2 2 2 2 2 2 2 2 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 8.379 8.532 8.534 8,762 9,165 9,323 9,646 9,719 9,792 12.119 12.916 0.8 0.8 0.8 0.8 0.7 0.7 0.7 0.7 0.7 0.6 0.5 99.2 99.2 99.2 99.2 99.3 99.3 99.3 99.3 99.3 99.4 99.5 1,697 1,658 1,666 1,673 1,636 1,689 1,696 1,627 1,640 1,665 1,669 833 843 849 850 854 845 841 847 871 861 891 200 20 1,000 9.564 0.8 99.2 1,735 830 494.407 40 2.97 200 20 1,000 9.602 0.7 99.3 1,634 841 472.214 40 2.93 200 200 200 200 200 200 200 200 200 200 200 1,000 1,000 1,000 1,000 1,000 1.000 1.000 1.000 1.000 1,000 1,000 11.101 12.547 12.571 14.729 18.880 20,456 23,673 24.456 25,141 48,446 56,382 0.6 99.4 0.5 99.5 0.5 99.5 0.5 99.5 0.4 99.6 0.3 99.7 0.3 99.7 0.3 99.7 0.3 99.7 0.1 99.9 0.1 99.9 1,631 1,647 1,608 1,701 1,656 1,720 1,651 1,664 1,654 1,677 1,690 852 836 846 834 851 839 854 832 836 880 878 DSMH 200 100 1,000 2 DSMH 200 100 1,000 1 15,528 0.4 99.6 1,649 823 484,330 40 2.94 15,681 0.4 99.6 1,642 837 477,691 40 2.97 DSMH 2 DSMH 1 SkSPl ChSPl SINGl SkSP2 D0UB1 MULTl ChSP2 SING2 MILST D0UB2 MULT2 20 20 20 20 20 20 20 20 20 20 20 174 422,249 36 13.38 2.415,672 200 1.66 2,314,704 200 1.73 3,482.980 280 5.4 6,251,123 519 1.37 7,458,253 599 1.42 9,172,601 759 1.46 9,859,351 799 1.39 10,227,345 833 1.42 24,368,253 1,997 1.25 30,052,596 2,396 1.26 TCS defectives detected totalsamp samp/ week % 36 438,930 99.7 829 1,688 200 2,420,051 857 1,694 99.8 2,334,412 200 842 1,621 99.8 3,440,077 281 840 99.8 1,643 6,414,425 519 837 99.9 1,706 7,224,719 599 857 99.9 1,645 9,093,216 759 846 99.9 1,620 9,720,761 799 833 99.9 1,632 9,729,017 834 862 1,645 99.9 24,173,033 1,997 853 100 1,635 28,562,083 2,396 892 100 1,594 ochrs method fc vc cf SkSPl (DhSPI SINGl SkSP2 iDOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 200 200 200 200 200 200 200 200 200 200 200 100 100 100 100 100 100 100 100 100 100 100 1,000 1.000 1.000 1.000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 DSMH 2 SkSPl DSMH 5 2 10,000 1,101 65.5 34.5 1,707 824 244,854 20 6.02 5 5 2 2 10,000 10,000 1,135 1,165 57.4 42.6 57.9 42.1 1,626 1,651 834 824 435,431 1,333,948 36 109 13.09 3.02 1 ChSPl 5 SINGl 5 SkSP2 5 DOUBl 5 MULT1 5 ChSP2 5 SING2 5 MILST 5 D0UB2 5 MULT2 5 2 2 2 2 2 2 2 2 2 2 10,000 10,000 10,000 10,000 10,000 10,000 10.000 10,000 10,000 10,000 1,297 1,330 1,573 1,967 2,134 2,438 2,503 2,555 4,855 5,686 50.2 51.3 44.9 34.9 32.6 27.8 26.6 25.3 12.7 11.5 49.8 48.7 55.1 65.1 67.4 72.2 73.4 74.7 87.3 88.5 1,617 1,676 1,715 1,693 1,711 1,644 1,669 1,625 1,618 1.706 829 846 839 856 837 853 849 843 888 911 2,419,868 200 2,429,398 200 3,480,097 280 6,335,384 519 7,517,876 599 8,860.385 759 9,845,976 799 10,069,626 833 23,633,779 1,997 29,123,558 2,396 TC/week TCF % 22,354 0.3 30,433 0.2 30,511 0.2 41,509 0.2 62.063 0.1 69,964 0.1 86,009 0.1 89,946 0.1 93,669 0.1 209,888 0.0 249,663 0.0 13.42 1.68 1.73 5.54 1.44 1.35 1.36 1.38 1.36 1.3 1.33 A 1.75 1.75 5.53 1.48 1.4 1.41 1.28 1.38 1.31 1.29 DSMH 2 DSMH 1 SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2! MULT2 5 20 10,000 2,386 27.9 72.1 1.623 839 470,957 40 2.99 5 20 10,000 2.664 24.5 75.5 1.632 834 718,640 59 2.64 5 5 5 5 5 5 5 5 5 5 5 20 20 20 20 20 20 20 20 20 20 20 DSMH 5 100 10,000 10,000 3,808 10,000 5,347 10,000 5,348 10,000 7,493 10,000 11,660 10,000 13,260 10,000 16,456 10,000 17,233 10,000 17,993 10,000 41,218 10,000 49,113 18.6 12.7 12.8 8.7 5.7 5.1 4.0 3.9 3.8 1.5 1.2 81.4 87.3 87.2 91.3 94.3 94.9 96.0 96.1 96.2 98.5 98.8 1.680 1,645 1,670 1,633 1,659 1,654 1,636 1,669 1,670 1,634 1.611 829 828 834 822 840 820 844 838 870 882 882 8,364 8.3 91.7 1,672 836 175 36 430,046 2,404,572 200 2,433,977 200 3,494,045 280 6,346,134 519 7,331,367 599 9,120,074 759 9,878,071 799 9,797,741 834 23,749,764 1,997 29,243,459 2,396 480,243 1 40 13.4 1.74 1.76 5.63 1.48 1.48 1.36 1.35 1.36 1.3 1.37 2.84 method fc 2 DSMH tfC cf TC/week TCF TCS defectives (detected % % 5 100 10,000 8,474 8.5 31.5 1,735 818 5 5 5 5 5 5 5 5 5 5 5 '100 -10,000 '100 •10,000 100 10,000 100 10,000 100 10,000 100 10,000 100 10,000 100 10,000 100 10,000 100 10,000 100 10,000 15,024 23,018 23,253 34,222 54,862 62,748 78,654 82.757 86.139 202.616 242,389 4.7 2.8 3.0 1.9 1.2 1.1 0.9 0.9 0.8 0.3 0.3 95.3 97.2 97.0 98.1 98.8 98.9 99.1 99.1 99.2 99.7 99.7 1,696 1,640 1,669 1,645 1,651 1,690 1,685 1,747 1,674 1,659 1,674 814 830 844 835 849 859 835 855 853 876 881 totalsamp 553,290 isamp/ ochrs week 2.84 43 A 1 SkSPl S1NG1 ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 DSMH 1 DSMH 2 SkSPl ChSPl SING1 SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 36 449.618 2.475,550 200 2,384,880 200 3.431,776 281 6,350,501 519 7,342,224 599 9,463,417 759 9,989,671 799 10,260,151 833 24,142,531 1,997 29,541,441 2,396 14.01 1.66 1.76 5.43 1.51 1.39 1.34 1.36 1.4 1.27 1.34 50 2 10,000 2,868 23.2 76.8 1,638 831 406,936 33 5.63 50 2 10,000 2,892 24.5 75.5 1,692 830 242,951 20 5.94 50 50 50 50 50 50 50 50 50 50 50 2 2 2 2 2 2 2 2 2 2 2 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 2,989 3,126 3,142 3,342 3,725 3,944 4,272 4,340 4,365 6,692 7,467 23.4 21.6 22.1 20.0 17.2 17.8 16.6 16.1 14.9 9.7 8.5 76.6 78.4 77.9 80.0 82.8 82.2 83.4 83.9 85.1 90.3 91.5 1,694 1,652 1,696 1,649 1,618 1.705 1.731 1.702 1.625 1,679 1,655 831 834 829 834 855 852 850 826 849 885 872 50 20 10,000 4.207 16.1 83.9 1,652 847 475,118 40 2.86 50 20 10,000 4.343 16.6 83.4 1,712 828 580.162 47 2.91 50 50 50 50 50 50 50 50 50 50 20 20 20 20 20 20 20 20 20 20 1,670 1,720 1.661 1,656 1,685 1.622 1,674 1,618 1,712 1,673 851 836 856 831 835 851 839 831 844 866 423,086 36 13.69 DSMH 2 DSMH 1 SkSPl SING1 ChSPl SkSP2 DOUB'1 MULTl ChSPJI SING2! MILS!• DOUB 2 10,000 10.000 lO.OOC1 10,00C1 10,00C) 10,00C) 10,00C) 10,00() 10,00CD 10,00(3 5.741 12.1 87.9 7,168 10.0 90.0 9.5 90.5 7,172 7.3 92.7 9,379 13,492 5.1 94.9 15,060 4.4 95.6 18,237 3.7 96.3 19.034 3.4 96.6. 19.749 3.5 96.£. 43.014 1.5 98.£i 176 36 440,929 2,419,067 200 2,497,207 200 3,420.170 281 6,191,253 519 7,270,788 599 9,419,752 759 9,988,173 799 9,926.830 834 24,360,280 1,997 29,651,148 2,396 2,470,659 200 2,370,718 200 3,400,429 281 6,371,692 519 6,998,493 599 9,471,137 759 9,671,934 799 10,327,260 833 24,655,489 1,997' 13.16 1.75 1.69 5.44 1.41 1.42 1.44 1.35 1.39 1.31 1.28 1.72 1.76 5.53 A O/^ 1.39 1.39 A O ^ 1.36 1.38 1.4 1.34 A A method fc vc % % MULT2 50 20 10,000 50.940 1.2 98.8 1.635 879 50 100 10,000 10,197 6.7 93.3 1.683 826 527,691 42 2.83 50 100 10,000 10,212 6.7 93.3 1.659 842 479,289 40 2.89 SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 S1NG2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 50 50 50 50 100 100 100 100 100 100 100 100 100 100 100 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 16,576 4.1 24,875 2.8 24,912 2.8 36,152 1.8 56,736 1.1 64,586 1.1 80.569 0.9 84.602 0.8 87.902 0.8 204.378 0.3 244,208 0.3 95.9 97.2 97.2 98.2 98.9 98.9 99.1 99.2 99.2 99.7 99.7 1.665 1.693 1.697 1.645 1.622 1.721 1.757 1.639 1.708 1,658 1,656 829 824 824 844 859 841 847 862 855 863 890 DSMH 1 DSMH 2 SkSPl SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 200 2 10,000 92.2 1,656 843 DSMH 2 DSMH 1 SkSPl SING1 ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 DSMH cf TC/week TCF TCS defectives detected 1 samp/ ochrs week 29,018,796 2,396 1.29 totalsamp 1 DSMH 2 8,860 7.8 36 442,440 2,499,564 200 2,476.056 200 3,391,959 281 6,148,183 519 7,386,602 599 9,600,483 759 9,506,476 799 10,394,326 832 24,627,616 1,997 29,323,785 2,396 13 1.73 1.73 5.47 1.47 1.4 1.39 1.37 1.37 1.27 1.32 2.85 502,170 42 200 2 10,000 8,860 7.8 92.2 1,671 844 200 200 200 200 200 200 200 200 200 200 200 2 2 2 2 2 2 2 2 2 2 2 10,000 10,000 10,000 10.000 10.000 10.000 10.000 10,000 10,000 10,000 10,000 9,002 9,150 9.152 9.374 9,799 9,974 10,217 10,329 10,372 12,682 13,470 7.5 92.5 7.5 92.5 7.5 92.5 7.3 92.7 7.2 92.8 7.3 92.7 6.3 93.7 6.5 93.5 6.2 93.8 5.0 95.0 4.6 95.4 1,637 1,649 1,676 1,667 1,684 1,748 1,633 1,663 1,625 1.636 1.633 835 844 852 816 827 817 843 842 860 868 859 200 20 10,000 10,212 6.6 93.4 1.650 846 2.93 40 426,099 36 13.57 1.7 2,349,308 200 2,391,844 200 1.67 3,479,959 280 5.66 6,325,269 519 1.43 7,708,561 599 1.41 9.334,963 759 1.41 9,722,580 799 1.37 9,885,136 834 1.38 24,368,604 1,997 1.29 29,730,454 2,396 1.27 478,164 477,286 3 40 200 20 10,000 200 200 200 200 200 200 200 200 20 20 20 20 20 20 20 20 10,276 10,000 11.572 10,000 13.162 10,000 13.179 10,000 15,426 10,000 19.498 10,000 21.087 10,000 24.251 10,000 25.072 6.5 6.0 5.2 5.3 4.3 3.5 3.2 2.8 2.6 93.5 94.0 94.8 94.7 95.7 96.5 96.8 97.2 97.4 177 1.639 1.670 1,664 1,694 1,648 1,679 1,658 1,686 1,654 851 822 836 839 862 840 839 840 860 3.08 492,149 439,969 2.405.334 2,427,716 3.304.508 6,312.483 7,201,571 9,592,485 9,578,149 42 36 200 200 281 519 599 759 799 13.63 1.67 1.7 5.35 1.48 1.47 1.39 1.38 method fc vc Cf TC/week TCF % MILST 200 20 10,000 25,750 2.7 D0UB2 200 20 10,000 49,049 1.3 MULT2 200 20 10,000 56,982 1.1 TCS defectives detected totalsamp samp/ % week 97.3 1,670 824 10,259.163 833 98.7 1,682 887 24,063,559 1,997 98.9 1,682 888 29.170.789 2,396 ochrs DSMH 11 DSMH 2 SkSPl SING1 ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 95.9 2.85 200 100 10,000 16,132 4.1 1,656 829 504.650 1.46 1.29 1.31 41 200 100 10.000 16,300 4.2 95.8 1,652 840 200 200 200 200 200 200 200 200 200 200 200 100 100 100 100 100 100 100 100 100 100 100 10.000 10.000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 22,775 30,917 31,029 42.114 62.752 70.503 86.665 90.590 93.980 210.493 250.286 3.1 2.2 2.2 1.7 1.1 1.0 0.7 0.8 0.7 0.3 0.3 96.9 97.8 97.8 98.3 98.9 99.0 99.3 99.2 99.3 99.7 99.7 1,683 1,682 1,686 1,701 1,691 1,744 1,583 1,675 1,658 1,682 1,667 834 829 837 837 856 837 850 841 846 879 898 DSMH 2 SkSPl ChSPl SINGl SkSP2 DSMH 1 DOUBl MULTl ChSP2 MILST SING2 D0UB2 MULT2 5 2 50,000 3.666 89.6 10.4 1,629 832 242.522 5 5 5 5 5 2 2 2 2 2 50,000 50,000 50,000 50,000 50,000 4.010 4.033 4,161 4,277 4,332 87.8 83.9 84.4 79.9 81.9 12.2 16.1 15.6 20.1 18.1 1,694 1,639 1,685 1,682 1,719 853 841 837 810 839 427.797 2,355,622 2,410.715 3,574,160 3,337,289 5 5 5 5 5 5 5 2 2 2 2 2 2 2 50,000 50,000 50,000 50,000 50,000 50,000 50,000 4,711 4,961 5,101 5,206 5,227 7,551 8,238 72.8 71.0 65.6 63.3 64.8 43.9 38.9 27.2 29.0 34.4 36.7 35.2 56.1 61.1 1,682 1,702 1,674 1,651 1,676 1,714 1,674 847 838 838 867 840 876 872 269 6,320,010 519 1.5 7,347,034 599 1.41 9,486,313 759 1.34 9,915,321 834 1.36 9,853,198 799 1.37 25,240,913 1.997 1.36 29,952,600 2.397 1.3 DSMH 2 SkSPl DSMH 1 ChSPl SINGl SkSP2 D0UB1 MULTl 5 20 50,000 5,081 67.0 33.0 1,670 836 2.94 40 433.464 36 12.74 2.487,032 200 1.67 2,433,596 200 1.77 5.7 3,485.718 280 6,353,097 519 1.48 7,620,326 599 1.41 8,903,598 759 1.34 9,735,470 799 1.41 10,089,082 833 1.36 24,594,372 1.997 1.3 28,619,853 2.396 1.33 470.797 5.81 20 20 20 20 20 50.000 50,000 50,000 50.000 50.000 6,456 6,701 8,019 8,091 10,144 14,346 16,088 53.7 46.3 53.2 46.8 41.8 42.0 32.5 23.2 21.8 58.2 58.0 67.5 76.8 78.2 178 1,682 1,702 1,657 1,652 1,640 1,630 1,714 CO 5 5 5 5 5 20 50,000 20 50,000 O <D C\J 5 5 200 200 280 490,163 822 831 443,232 1,446,211 844 860 820 846 842 2,423,479 2,328,654 3,486.971 6.120.310 7.447.219 12.6 1.79 1.78 5.53 1.69 2.89 40 36 118 200 200 280 519 599 13.2 2.27 1.69 1.65 5.53 1.48 1.47 TCS defectives detected totalsamp samp/ week % 9,277.233 759 1,707 859 82.0 10.004,334 799 837 1,652 83.6 10,249,845 833 842 1.662 83.8 23,766,967 1,997 874 92.6 1.648 29,604,419 2,396 915 1.701 93.8 method fc vc cf ChSP2 SING2 MILST D0UB2 MULT2 5 5 5 5 5 20 20 20 20 20 50,000 50,000 50.000 50,000 50,000 DSMH 2 DSMH 1 SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 5 100 50,000 11.131 30.8 69.2 1.670 5 100 50,000 12,740 26.4 73.6 1.664 5 5 5 5 5 5 5 5 5 5 5 100 100 100 100 100 100 100 100 100 100 100 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 17,281 25,987 26,063 36,895 57,635 65,405 81,394 85,520 88,875 205,153 244,964 19.9 13.7 13.3 9.0 6.0 5.1 4.0 4.0 3.8 1.5 1.2 2 50,000 50,000 50,000 DSMH 50 2 SkSPl DSMH 1 SINGl ChSPl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 DSMH 2 DSMH 1 SkSPl SINGl ChSPl 1SkSP2 TC/week TCF % 19,263 18.0 19,820 16.4 20,584 16.2 43,826 7.4 51,687 6.2 845 1.38 1.37 1.36 1.37 1.31 2.81 480,706 40 813 2.68 770,833 61 \j 1 450,099 36 2.439,502 200 2,426,493 200 3,420,168 281 6,368,981 519 7,300,631 599 9,101,567 759 9,786,524 799 10,207,576 833 23,401,416 1,997 27,223,436 2,396 80.1 86.3 86.7 91.0 94.0 94.9 96.0 96.0 96.2 98.5 98.8 1.682 1,696 1,683 1,657 1,694 1.652 1.617 1,696 1,669 1,620 1,598 815 828 843 844 840 842 843 853 850 907 905 5,616 61.2 38.8 1,681 833 246,667 5,735 5,768 60.1 39.9 60.3 39.7 1,661 1,698 826 844 433,729 1,072,491 87 2,347,649 200 2,425,038 200 3,473,371 280 6,439,824 519 7,315,381 599 9,019,626 759 9,806,645 799 10,418,478 832 24.397,381 1,997 30,132,264 2,396 50 50 2 2 50 50 50 50 50 50 50 50 50 50 2 50,000 2 50,000 2 50,000 2 50,000 2 50,000 2 50,000 2 50,000 2 50,000 2 50,000 2 50,000 5,864 5,993 6,069 6,643 6,648 6,878 6,888 7,035 9,311 10,111 58.2 59.1 56.1 53.6 51.2 48.2 47.1 47.3 35.1 32.4 41.8 40.9 43.9 46.4 48.8 51.8 52.9 52.7 64.9 67.6 1,649 1,688 1,669 1,717 1,683 1,647 1,628 1,678 1,670 1,696 847 828 826 834 851 859 831 845 871 872 50 20 50,000 7,163 51.2 48.8 1,732 834 50 20 50,000 7,643 44.4 55.6 1.655 843 50 50 50 50 20 20 20 20 8,385 9,804 9,895 12,098 39.1 33.9 34.8 28.2 60.9 66.1 65.2 71.8 1,624 1,644 1.675 1.660 865 838 829 841 50,000 50,000 50,000 50,000 ochrs 179 13.15 1.74 1.78 5.34 1.39 1.39 1.4 1.35 1.41 1.29 1.31 5.71 20 36 12.95 4.75 1.72 1.77 5.27 1.47 1.46 1.35 1.4 1.32 1.29 1.34 2.92 489,602 40 2.54 957,174 415,653 2,420,915 2,455,623 3,374,999 80 36 200 200 281 13.39 1.79 1.73 5.4 method fc vc cf DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 50 50 50 50 50 50 50 20 20 20 20 20 20 20 50,000 50,000 50,000 50.000 50.000 50,000 50,000 DSMH 2 DSMH 1 SkSPl ChSPl SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 50 100 50,000 TC/week TCF % 16,119 20.4 17.732 18.9 21.169 16.9 21.806 15.7 22,250 14.3 45,587 7.1 53,589 6.1 12,921 TCS defectives detected totalsamp samp/ week % 79.6 6,047,335 519 1,615 849 81.1 7,284,510 599 1,655 842 83.1 8,990,443 759 1,693 846 1,704 9,733,030 799 84.3 868 85.7 10,241,788 833 1,629 844 25,106,776 1,997 92.9 1,679 864 29,132,914 2,396 93.9 1.671 871 26.0 74.0 1.657 853 ochrs 1.48 1.45 1.31 1.35 1.37 1.29 1.21 2.84 478,645 40 50 100 50,000 14,312 23.9 76.1 1,665 827 50 50 50 50 50 50 50 50 50 50 50 100 100 100 100 100 100 100 100 100 100 100 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50.000 19,628 27.555 27,610 39,009 59,364 67,070 83,332 87,317 90.577 207.109 246.771 17.4 12.0 12.0 8.9 5.6 4.6 4.0 3.8 3.8 1.6 1.3 82.6 88.0 88.0 91.1 94.4 95.4 96.0 96.2 96.2 98.4 98.7 1,667 1,665 1,656 1,662 1,622 1,566 1,688 1,666 1,682 1,682 1,651 834 832 836 827 838 848 853 871 831 858 888 DSMH 1 SkSPl DSMH 2 ChSPl SINGl SkSP2 DOUBl MULTl MILST ChSP2 S1NG2 D0UB2 MULT2 200 2 50,000 11,585 29.2 70.8 1,662 838 DSMH 2 DSMH 1 SkSPl ChSPl 200 20 50,000 28.6 71.4 30.2 69.8 1,637 1,705 817 826 11,766 28.1 71.9 12,004 29.5 70.5 12,273 29.2 70.8 12,455 26.9 73.1 12,555 26.3 73.7 12,930 24.8 75.2 13,008 26.4 73.6 13,080 26.2 73.8 15,370 21.6 78.4 16,122 20.3 79.7 1,654 1,714 1,696 1,645 1,643 1,640 1.698 1,653 1,701 1,669 839 841 854 857 849 870 862 843 856 896 12,895 26.2 73.8 1.643 837 200 20 50,000 13,244 26.1 73.9 1.687 200 20 50,000 200 20 50,000 14,418 15,834 24.1 75.9 21.2 78.8 1.687 1.645 200 200 200 200 200 200 200 200 200 200 200 200 2 2 2 2 2 2 2 2 2 2 2 2 50,000 50,000 50,000 50.000 50.000 50.000 50.000 50,000 50,000 50,000 50,000 50,000 11,651 11,692 2.75 690,063 56 437,021 36 2,518,684 200 2,470,666 200 3,364,841 281 6,129,307 519 6,962,804 599 9,407.163 759 9,497,697 799 10,374,880 832 24,536,809 1.997 28,683,066 2.396 2.76 688,282 440,229 498,099 13.11 1.8 1.73 5.52 1.44 1.43 1.41 1.33 1.42 1.27 1.3 57 36 40 2,462,576 200 2.461,348 200 3,306,185 281 6,097,968 519 7,206,779 599 10,016,582 834 9,241,090 759 9,448,752 799 25.417,748 1,997 28.315,714 2,396 13.83 2.99 1.67 1.8 5.26 1.51 1.39 1.28 1.37 1.39 1.3 1.33 2.86 477.343 40 180 845 852 831 2.82 655.930 431,112 2,426,369 54 36 200 13.15 1.74 method fc vc cf TC/week TCF TCS defectives detected % SINGl SkSP2 DOUBl MULTl ChSP2 SING2 MILST D0UB2 MULT2 200 200 200 200 200 200 200 200 200 20 20 20 20 20 20 20 20 20 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 DSMH 200 100 50,000 2 % 16,015 22.1 77.9 17,985 18.2 81.8 22,296 15.7 84.3 23,858 14.5 85.5 26,851 12.2 87.8 27,905 12.6 87.4 28,554 12.1 87.9 51,647 6.3 93.7 59,542 5.4 94.6 1,694 1,630 1,698 1,695 1,640 1,698 1,687 1,685 1,688 829 826 845 853 835 847 849 877 862 18,905 1.683 823 18.5 81.5 totalsamp samp/ week 2,438,363 200 3.445,080 281 6,344,707 519 7,309,395 599 9,356,327 759 9,699,317 799 10,095,203 833 24,709,109 1,997 30,673,198 2,397 1.73 5.76 1.41 1.43 1.4 1.37 1.38 1.34 1.38 2.92 491,786 40 Cm DSMH 200 11 SkSPl 200 SINGl 200 ChSPl 200 SkSP2 200 DOUBl 200 MULTl 200 ChSP2 200 SING2 200 MILST 200 D0UB2 200 MULT2 200 ochrs 100 50,000 100 100 100 100 100 100 100 100 100 100 100 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 50,000 19,755 16.8 83.2 25,523 13.9 86.1 33,495 9.5 90.5 33,783 10.1 89.9 44,667 7.6 92.4 65,342 5.1 94.9 73,307 4.7 95.3 89,336 3.8 96.2 93,219 3.5 96.5 96,686 3.3 96.7 213,124 1.5 98.5 252,745 1.3 98.7 181 1,619 1,708 1,623 1,671 1,671 1,675 1,674 1,635 1,637 1,621 1,637 1,680 830 825 843 845 841 848 830 836 850 850 865 868 2.95 580,812 445,694 2,438,285 2,413,572 3,419,468 6,385,284 7,347,550 9,049,175 9,644,711 9,924.954 23,896,044 29,890,767 49 36 200 200 281 519 599 759 799 834 1.997 2,396 13.36 1.68 1.77 5.28 1.4 1.43 1.44 1.36 1.38 1.31 1.25 APPENDIX F A SAS PROGRAM FOR THE FINE-TUNING OF DSM-HQ, STAGE 1 * SIMULATION PROGRAM - Stage 1; %let sigma3 = 0.06807; %let sigma4 = 0.006210; %let sigmaS = 0.000233; %let sigma6 = 0.0000034; %let lotsize = 1000; %let lotsperday = 8; %let daysperweek = 5; %let weeksperyear = 26; %let years = 1; %let sampobs = 1 ; %let allseries = %eval(&lotsize*&lotsperday*&daysperweek*&weeksperyear) ; %let numlots = %eval(&allseries/&lotsize); %let itiaxsamplesize = 10; * Macro generate creates two arrays. The first array is the Poisson * is the Os and Is for the appropriate sigma level and the second * array is the observation number for each of the Os and Is; %MACRO generate(sigmalevel); DATA productionphasel; ARRAY dpl{&numlots} dl - d&numlots; ^defects per lot; ARRAY next{4} nxl-nx4; ^used to check if next 4 are oc; ARRAY samptest{&maxsamplesize} stl - st&maxsamplesize; *sample to be destructively tested; ARRAY series{&allseries} si - s&allseries; ^production series; ARRAY re_oc{50} reocl - reocSO; *used to decide if error is random event or out-of-control; ARRAY fcost{3} fcl - fc3; *fixed cost; fcl = 5; fc2 = 50; fc3 = 200; ARRAY vcost{3} vcl - vc3; *variable cost; vcl = 2; vc2 = 20; vc3 = 100; ARRAY cfield{3} cfl - cf3; *cost of finding defect in the field; cfl = 1000; cf2 = 10000; cf3 = 50000; do pp = 1 to 3 do qq = 1 to 3 do rr = 1 to 3 fc = fcost{pp}; vc = vcost{qq}; 182 cf = cfield{rr}; do minimumsampling = 1 to 15; TCperweek = 0; week for all series; totaldefectives = 0; all series; totaidefeetivesdetected all series; totalsamples = 0; series; totalweeks = 0; all series; totalunits = 0; produced; totalrandomevents = 0; all series; totaloutofcontrol = 0; events in all series; oc_totaldays = 0; days OC; oc_number = 0; the process went OC; needmoreweeks = 0; *Total cost per *Total defectives in 0; do nn = 1 to 1000; ro = 0; oc_point = 0 *Total defectives detected in *Total samples in all *Total weeks in *Total units *Total random events in *Total out-of-control *cumulative number of *number of times *out-of-control point in the series; *place where sampling method oc_found = 0 finds out-of-control; *place where binary search start_oc = 0 finds out-of-control; *binary search high; hi = 0; *binary search low; lo = 0; *number of days the sampling oc_daystodectect = 0; method took to detect OC; *random binary do jj = 1 to &allseries; generator according to sigma level; seriesljj} = RANBIN(32,1,&sigmalevel); end; do jj = 1 to 50; *determines if next error is OC or RE; re_oc{jj} = RANBIN{51,1,0.5); *for RE and OC combined; re_oc{jj} = 1; *for only random events; re oc{jj} = 0; *for only out-ofcontrol; end; *start of series is 0, avoids out of si = 0; range in binary search; 183 defectives = 0; *number of defectives produced; do jj = 1 to &allseries; if series{jj} > 0 then do; *if defective then; ro = ro + 1; *for random and oc combined; ro = 1; *for random or out of control; defectives = defectives + 1; ^defective counter; if re_oc{ro} = 0 then goto outofcontrol; end; end; outofcontrol: ; oc_point = jj; if oc_point < &allseries then do; do jj = oc_point to &allseries; seriesijj} = 1; end; end; invalid = 0; *used when function does not have an intercept; intercept = 0; ^intercept which helps determine samplesize; startint = 0; *functions starting intercept; endlambda = 0; *ending lambda in the series; alpha = 1; beta = 294117; lambda = alpha/beta; randomevents = 0; *number of random events detected in the series; outofcontrol = 0; *number of out-of-control events in the series; oc_daystodetect = 0; *number of days to detect the out-of-control; samplesize = 0; *sample size for the lot(s); aa = 0 bb = 0 cc = 0 sfc = 0; SVC = 0; tsvc = 0; rfc - 0 rvc = 0, ofc = 0 ovc = 0, nd = 0; m = 0; orwc = 0; 184 SC = 0; REC = 0 OCC = 0 TCS = 0 do jj = 1 to 4; next{jj} = 0; end; do jj = 1 to &numlots; dpHjj} = 0; end; % dsmhq; otherseries: ; numberoflots = b/&lotsize; * number of lots in the series; totalunits = totalunits + totalseries; *total units accumulated; if oc_found > oc_point then do; oc_daystodetect = (oc_found oc__point) / (&lotsize * &lotsperday) ; oc_totaldays = oc_daystodetect + oc_totaldays; oc_number = oc_number + 1; end; sfc = fc * numberoflots; * sampling fixed cost; rfc = fc * randomevents; * random event fixed cost; rvc = vc * randomevents * 4; * random event variable cost; ofc = fc * outofcontrol; * out-of-control fixed cost; ovc = vc * nd; * out-of-control variable cost; orwc = m*vc/2; * out-of-control rework cost; SC = sfc + tsvc; * Sampling Cost; REC = rfc + rvc; * Total Random Event Cost; OCC = ofc + ovc + orwc; * Total Out-ofcontrol Cost; TCS = SC + REC + OCC; * Total Cost of Sampling; TCF = (defectives - defectivesdetected) * cf; * Total Cost of Field; TC = TCS + TCF; * Total Cost; endlambda = alpha/beta; TCperweek = TCperweek + TC; totaldefectives = totaldefectives + defectives; totaldefectivesdetected = totaldefectivesdetected + defectivesdetected; totalrandomevents = totalrandomevents + randomevents; totaloutofcontrol = totaloutofcontrol + outofcontrol; 185 keep samplesize startint intercept oc_found start_oc totalseries totalunits oc_daystodetect oc_totaldays oc__number defectivesdetected TC; keep minimumsampling invalid alpha beta lambda endlambda samplesize startint intercept fc vc cf TCS TCF TC defectives defectivesdetected; output; end; if b > &allseries then needmoreweeks + 1; totalweeks = totalunits/(&lotsize*&lotsperday*&daysperweek); TCperweek = TCperweek/totalweeks; keep fc vc cf minimumsampling TCperweek totalweeks totaldefectives totaldefectivesdetected totalsamples totalrandomevents totaloutofcontrol needmoreweeks; output; end; end; end; end; %mend generate; * Macro dsmhq performs divisions of the series into lots and calls; * the Macro sample to sample the lot and the Macro analyzedsmhq to perform; * the analysis of the sampling method; %MACRO dsmhq; n = 0; a = 1; lot(s) tested; b = a + &lotsize - 1; acceptedlots = 0; rejectedlots = 0; defectivesdetected = 0; detected in this lot(s); ci = fc + 4*vc; random event; ^beginning of *end of the lot{s) tested; *number of defectives *cost in-house for a do until (b > &allseries); do mm = 1 to &maxsamplesize; samptest{mm} = 0; end; xtralots = 0; aa = vc/(cf-ci); bb = (fc + (vc^beta))/(cf-ci); cc = ((fc*beta)/(cf-ci)) - alpha; if (bb**2) > (4*aa*cc) then do; intercept = (-bb + SQRT((bb**2)(4*aa*cc)))/(2*aa); if a = 1 then startint = intercept; end; 186 else do; invalid = invalid + 1; intercept = &maxsamplesize; end; if intercept <= 0 then intercept = 0.020; *tests at a min rate of once/week; * days_to_minsampling = b/(&lotsize*&lotsperday); * keep intercept totalsamples samplesize fc vc cf alpha beta defectivesdetected b days_to_minsampling; * output; * if intercept <= 0 then goto otherseries; *to determine when process reached one per week sampling; samplesize = ceil(intercept); if intercept >= Smaxsamplesize then samplesize = &maxsamplesize; select; when(minimumsampling = 1) do; * one/week; if 0 <= intercept <= 0.025 then xtralots = 39 * &lotsize; if 0.025 < intercept <= 0.03125 then xtralots = 31 * &lotsize; if 0.03125 < intercept <= 0.0625 then xtralots = 15 * &lotsize; if 0.0625 < intercept <= 0,125 then xtralots = 7 * &lotsize; if 0.125 < intercept <= 0.25 then xtralots = 3 * Slotsize; if 0.25 < intercept <= 0.5 then xtralots = &lotsize; if 0.5 < intercept then xtralots = 0; end; when(minimumsampling = 2) do; * one/2 days; if 0 < intercept <= 0.0625 then xtralots = 15 * &lotsize; if 0.0625 < intercept <= 0.125 then xtralots = 7 * &lotsize; if 0.125 < intercept <= 0.25 then xtralots = 3 * &lotsize; = &lotsize; if 0.25 < intercept <= 0.5 then xtralots if 0.5 < intercept then xtralots = 0; end; when(minimumsampling = 3) do; * one/day max; if 0 < intercept <= 0.125 then xtralots 7 * &lotsize; 187 if 0.125 < intercept <= 0.25 then xtralots = 3 * Slotsize; if 0.25 < intercept <= 0.5 then xtralots = &lotsize; if 0.5 < intercept then xtralots = 0; end; when (minimumsampling = 4) do; * one/4 lots; if 0 < intercept <= 0.25 then xtralots = 3 *• &lotsize; if 0.25 < intercept <= 0.5 then xtralots = &lotsize; if 0.5 < intercept then xtralots = 0; end; when(minimumsampling = 5 ) do; * one/2 lots; if 0 < intercept <= 0.5 then xtralots = Slotsize; if 0.5 < intercept then xtralots = 0; end; when(minimumsampling = 6) xtralots = 0 ; * one/lot; when(minimumsampling = 7) do; *min 2 samples/lot; if intercept < 2 then samplesize = 2; xtralots = 0; end; when(minimumsampling = 8 ) do; *min 3 samples/lot; if intercept < 3 then samplesize = 3; xtralots = 0; end; when(minimumsampling = 9) do; *min 4 samples/lot; if intercept < 4 then samplesize = 4; xtralots = 0; end; when(minimumsampling = 10) do; *min 5 samples/lot; if intercept < 5 then samplesize = 5; xtralots = 0; end; when(minimumsampling = 11) do; *min 6 samples/lot; if intercept < 6 then samplesize = 6; xtralots = 0; 188 end; when (minimumsampling = 12) do; *min 7 samples/lot; if intercept < 7 then samplesize = 7; xtralots = 0; end; when(minimumsampling = 13) do; *min 8 samples/lot; if intercept < 8 then samplesize = 8; xtralots = 0; end; when (minimumsampling = 14) do; *min 9 samples/lot; if intercept < 9 then samplesize = 9; xtralots = 0; end; when(minimumsampling = 15) do; *min 10 samples/lot; if intercept < 10 then samplesize = 10; xtralots = 0; end; end; SVC = vc * samplesize; ^sampling variable cost; tsvc = tsvc + svc; *total sampling variable cost; b = b + xtralots; *end of the lot(s) tested if there are additional lots; totalseries = b; *totalsereis is b if out-of-control; %sainple; *takes a sample from lot(s) examined from a to b; % analyzedsmhq; beta = beta + samplesize; a = &lotsize + xtralots + a; b = a + &lotsize - 1; if &allseries < b then totalseries = &allseries; end; %mend dsmhq; * MACRO sample performs a random sample size from the start of the lot(s) "a" to the end of the lots "b" and stores the values on the array samptest; %MACRO sample; do i = 1 to samplesize; sampobs = (floor((b-a+1)*RANUNI(99)+a-l)); if sampobs > &allseries-8 then do; *to avoid out of range if b>allseries; i = i-1; 189 goto tryother; end; if sampobs <= 8 then do; *to avoid out of range in OC/RE search; i = i-1; goto tryother; end; do k = 1 to samplesize; if samptest{k} = sampobs then do; i = i-1; goto tryother; *if item is already part of sample; end; end; samptest{i} = sampobs; tryother: ; end; %mend sample; * MACRO analyzedsmhq accumulates the number of total samples and defectives detected; * if there is a defective, alpha is adjusted and the Macro determinetype is called; %MACRO analyzedszahq; n = n + 1; do mm = 1 to samplesize; totalsamples = totalsamples + 1; if series{samptest{mm}} > 0 then do; dpl{n} = dpl{n} + 1; defectivesdetected = defectivesdetected + 1; alpha = alpha + 1; % determinetype; end; end; if dpl{n} > 0 then rejectedlots = rejectedlots + 1; else acceptedlots = acceptedlots + 1; %mend analyzedsmhq; * MACRO determinetype is used to determine if the defective is a random event or an out-of-control situation. If out-of-control the Macro binarysearch is called; %MACRO determinetype; totalnext = 0; next{l} = samptest{mm} + 1 next {2} == samptest {mm} + 2 next{3} = samptest{mm} + 4 next{4} = samptest{mm} + 8 totalnext = series{next{1}} +series{next{2}} +series{next{3}} +series{next{4}}; 190 if totalnext > 3 then do; outofcontrol = outofcontrol + 1; nxtl=series{nxl} nxt2=series{nx2} nxt3=series{nx3} nxt4=series{nx4] keep randomevents outofcontrol nxl nx2 nx3 nx4 nxtl nxt2 nxt3 nxt4; output; %hlna2:ysearch; end; else randomevents = randomevents + 1; * nxtl=series{nxl} nxt2=series{nx2} * nxt3=series{nx3} nxt4=series{nx4] keep randomevents outofcontrol nxl nx2 nx3 nx4 nxt1 nxt2 nxt3 nxt4; output; %mend determinetype; * MACRO binarysearch starts with oc_found which is the point where the sampling method found the out-of-control situation and determines where the OC began; %MACRO binarysearch; hi = samptest{mm}; oc_found = hi; lo = 0; kk = 0; point = 0; check_oc = hi; start_oc = 0; do until (series{check_oc} < 1 ) ; check_oc = samptest{mm} - 2**kk; if check_oc < 1 then check__oc = 1; kk = kk + 1; end; lo = check_oc; do until (lo > hi); point = floor((lo+hi) * 0.5); select; when(series{point} < 0.5) do; lo = point + 1; nd = nd + 1; end; when(series{point} >= 0.5) do; hi = point - 1; nd = nd + 1; end; end; end; start_oc = point; 191 m = b - start_oc - nd; goto otherseries; %mend binarysearch; generate(&sigma6) ; 192 APPENDDC G A SAS PROGRAM FOR COMPARISON OF DSM-HQ WITH EXISTING METHODS, STAGE 2 * SIMULATION PROGRAM - Stage 2 - Comparison of all Methods; %let sigma3 = 0.066807; %let sigma4 = 0.006210; %let sigmaS = 0.000233; %let sigma6 = 0.0000034; %let lotsize = 1000; %let lotsperday = 8; %let daysperweek = 5; %let weeksperyear = 26; %let years = 1; %let sampobs = 1; %let allseries = %eval(&lotsize*&lotsperday*&daysperweek*&weeksperyear); %let numlots = %eval(&allseries/&lotsize); %let maxsamplesize = 10; %let allsamplesize = 80; * Macro generate creates two arrays. The first array is the Poisson * is the Os and Is for the appropriate sigma level and the second * array is the observation number for each of the Os and Is; %MACRO generate(sigmalevel); DATA productionphase2; REOC = 1 ; only, 2 for REOC combined; * 1 for RE sig = 5; if ssigmalevel < 0.000100 then sig = 6; ARRAY dpl{&numlots} dl - d&numlots; ^defects per lot; ARRAY next{4} nxl-nx4; *used to check if next 4 are oc; ARRAY samptest{&allsamplesize} stl - st&allsamplesize; *sample to be taken by each method; ARRAY series{&allseries} si - s&allseries; ^production series; ARRAY re_oc{50} reocl - reocSO; *used to decide if error is random event or out-of-control; ARRAY MILSTDhistory{10} MShl - MShlO; *MILSTD-105 sampling history; ARRAY fcost{3} fcl - fc3; 193 *fixed cost; fcl = 5; fc2 = 50; fc3 = 200; ARRAY vcost{3} vcl - vc3; ^variable cost; vcl = 2; vc2 = 20; vc3 = 100; ARRAY cfield{3} cfl - cf3; *cost of finding defect in the field; cfl = 1000; cf2 = 10000; cf3 = 50000; method = "DSMHl"; do pp = 1 to 3 do rr = 1 to 3 do qq = 1 to 3 fc = fcost{pp}; vc = vcost{rr}; cf = cfield{qq}; do samplingmethod = 1 to 13; TCperweek = 0; week for all series; *Total cost per TCFperweek = 0 TCSperweek = 0 OCCperweek = 0 orwcperweek = 0; TCFpercent = 0; TCSpercent = 0; totaldefectives = 0; *Total defectives in all series; totaldefectivesdetected = 0; *Total defectives detected in all series; totaldefectivesnotdetected = 0; totalsamples = 0; *Total samples in all series; totalweeks = 0 ; *Total weeks in totalunits = 0; *Total units all series; produced; totalrandomevents = 0; *Total random events in all series; totaloutofcontrol = 0; events in all series; oc_totaldays = 0; days OC; oc_number = 0; the process went OC; oc_average = 0; *Total out-of-control ^cumulative number of *number of times totalacceptedlots = 0; totalrejectedlots = 0; needmoreweeks = 0; do nn = 1 to 1000; ro = 0; 194 oc_point = 0; *out-of-control point in the series; oc_found = 0; *place where sampling method finds out-of-control; start_oc = 0; *place where binary search finds out-of-control; ^i ~ 0; *binary search high; lo = 0; *binary search low; oc_daystodectect = 0; ^number of days the sampling method took to detect OC; do jj = 1 to &allseries; *random binary generator according to sigma level; series{jj} = RANBIN(32,1,&sigmalevel); end; do jj = 1 to 50; ^determines if next error is OC or RE; if REOC = 2 then re_oc{jj} = RANBIN(51,1,0.5); *for RE and OC combined; if REOC = 1 then re_oc{jj} = 1; *for random events only; * if REOC = 1 then re_oc{jj} = 0; *for out-of-control only; end; si = 0; *start of series is 0, avoids out of range in binary search; defectives = 0 ; *number of defectives produced; do jj = 1 to &allseries; if series{jj} > 0 then do; *if defective then; if REOC = 2 then ro = ro + 1; *for random and OC combined; if REOC = 1 then ro = 1; *for random or out of control; defectives = defectives + 1; *defective counter; if re oc{ro} = 0 then goto outofcontrol; end; end; outofcontrol: ; oc_point = jj; if oc_point >= &allseries then needmoreweeks + 1; if oc_point < &allseries then do; do jj = oc_point to &allseries; series{jj} = 1; end; end; invalid = 0; *used when function does not have an intercept; intercept = 0; determine samplesize; *intercept which helps 195 startint = 0; *functions starting intercept; randomevents = 0; ^number of random events detected in the series; outofcontrol = 0; *number of out-of-control events in the series; oc_daystodetect = 0; *number of days to detect the out-of-control; samplesize == 0; ^sample size for the lot(s); sfc = 0; SVC = 0; tsvc = 0; rfc = 0 rvc = 0 ofc = 0 ovc = 0 nd = 0; m = 0; orwc = 0; SC = 0; REC = 0 OCC = 0 TCS = 0 a = 1; *beginning of lot(s) tested; b = a + &lotsize - 1; *end of the lot(s) tested; acceptedlots = 0; rejectedlots = 0; aefectivesdetected = 0; deteced in this lot(s); n = 0; count defects per lot; ci = fc + 4*vc; ^number of defectives do jj = 1 to 4; next{jj} = 0; end; do jj = 1 to &numlots; dpl{jj} = 0; end; select; when(samplingmethod = 1) do; beta = 15; method = "DSMHl"; minimumsampling = 1 ; %RE;andCX:; % dsmhq; end; when(samplingmethod = 2 ) do; if sig = 5 then beta - 4292; if sig = 6 then beta = 294117; minimumsampling = 1; 196 *used to method = "DSMH2"; %PEandOC; % dsmhq; end; when(samplingmethod = 3 ) do; method = "SINGl"; samplesize = 5; % singlesampling; end; when(samplingmethod = 4 ) do; method = "SING2"; samplesize = 20; % singrlesamplingr; end; when(samplingmethod = 5) d o ; method = "DOUBl"; samplesize = 13; %dovhlesampllng; end; when(samplingmethod = 6 ) do; method = "D0UB2"; samplesize = 50; % doublesampllng; end; when(samplingmethod = 7 ) do; method = "MULTl"; samplesize = 5; %inul tlplesas^llng; end; when(samplingmethod = 8 ) do; method = "MULT2"; samplesize = 20; %iziul tlplesampllng; end; when(samplingmethod = 9) do; method = "SkSPl"; samplesize = 5; i_sksp = 9; inverse_f_sksp = 5; %SkSP2; end; when(samplingmethod = 10) do; method = "SkSP2"; samplesize = 20; i_sksp = 15; inverse_f_sksp = 2; %SkSP2; end; when(samplingmethod = 11) do; method = "ChSPl"; samplesize = 5; %ChSPl; end; 197 when(samplingmethod = 12) do; method = "ChSP2"; samplesize = 19; % ChSPl; end; when(samplingmethod = 13) do; method = "MILST"; samplesize = 5; %M1LSTD105; end; end; otherseries: ; numberoflots = b/&lotsize; * number of lots in the series; totalunits = totalunits + totalseries; *total units accumulated; if oc_found > oc_point then do; oc_daystodetect = (oc_found oc_point)/(&lotsize * &lotsperday); oc_totaldays = oc_daystodetect + oc_totaldays; oc_number = oc_number + 1; end; sfc = fc * numberoflots; * sampling fixed cost; rfc = fc * randomevents; * random event fixed cost; rvc = vc * randomevents * 4; * random event variable cost; * out-of-control ofc = fc * outofcontrol; fixed cost; * out-of-control ovc = vc * nd; variable cost; * out-of-control orwc = m*vc/2; rework cost; * Sampling Cost; SC = sfc + tsvc; REC = rfc + rvc; * Total Random Event Cost; OCC = ofc + ovc + orwc; * Total Out-ofcontrol Cost; TCS = SC + REC + OCC; * Total Cost of Sampling; TCF = (defectives - defectivesdetected) * cf; * Total Cost of Field; TC = TCS + TCF; * Total Cost; endlambda = alpha/beta; TCperweek = TCperweek + TC; TCFperweek = TCFperweek + TCF; TCSperweek = TCSperweek + TCS; orwcperweek = orwcperweek + orwc; totaldefectives = totaldefectives + defectives; 198 totaldefectivesdetected = totaldefectivesdetected + defectivesdetected; totalrandomevents = totalrandomevents + randomevents; totaloutofcontrol = totaloutofcontrol + outofcontrol; totalacceptedlots = totalacceptedlots + acceptedlots; totalrejectedlots = totalrejectedlots + rejectedlots; keep samplesize startint intercept oc_found start_oc totalseries totalunits oc_daystodetect oc_totaldays oc_number defectivesdetected TC; keep minimumsampling invalid alpha beta lambda endlambda samplesize startint intercept fc vc cf TCS TCF TC defectives defectivesdetected; * output; end; totalweeks = totalunits/ (&lotsize*&lotsperday*&daysperweek) ; totaldefectivesnotdetected = totaldefectives totaldefectivesdetected; TCperweek = TCperweek/totalweeks; TCFperweek = TCFperweek/totalweeks; TCSperweek = TCSperweek/totalweeks; OCCperweek = OCCperweek/totalweeks; orwcperweek = orwcperweek/totalweeks; TCFpercent = TCFperweek/TCperweek; TCSpercent = TCSperweek/TCperweek; if oc_number > 0 then oc_average = oc_totaldays/oc_number; keep totalrejectedlots totalacceptedlots samplingmethod method fc vc cf TCperweek TCFperweek TCSperweek TCFpercent TCSpercent OCCperweek orwcperweek oc_average totalweeks totaldefectives totaldefectivesdetected totaldefectivesnotdetected totalsamples totalrandomevents totaloutofcontrol sig REOC needmoreweeks; output; end; end; end; end; %mend generate; * Macro dsmhq performs divisions of the series into lots and calls; * the Macro sample to sample the lot and the Macro analyzedsmhq to perform; * the analysis of the sampling method; %MACRO dsmhq; aa = 0; bb = 0; cc = 0; 199 endlambda = 0; *ending lambda in the series; alpha = 1; lambda = alpha/beta; do until (b > Sallseries); do mm = 1 to &allsamplesize; samptest{mm} = 0; end; xtralots = 0; aa = vc/(cf-ci); bb = (fc + (vc*beta))/(cf-ci); cc = ((fc*beta)/(cf-ci)) - alpha; if (bb**2) > (4*aa*cc) then do; intercept = (-bb + SQRT((bb**2)(4*aa*cc)))/(2*aa); if a = 1 then startint = intercept; end; else do; invalid = invalid + 1; intercept = &maxsamplesize; end; * days_to_minsampling = b/(&lotsize*&lotsperday); * keep intercept totalsamples samplesize fc vc cf alpha beta defectivesdetected b days__to__minsampling; * output; * if intercept <= 0 then goto otherseries; *to determine when process reached one per week sampling; if intercept <= 0 then intercept = 0.020; *tests at a min rate of once/week; samplesize = ceil(intercept); if intercept >- &maxsamplesize then samplesize = &maxsamplesize; select; when(minimumsampling = 1 ) do; * one/week; if 0 <= intercept <= 0.025 then xtralots = 39 * &lotsize; if 0.025 < intercept <= 0.03125 then xtralots = 31 * &lotsize; if 0.03125 < intercept <= 0.0625 then xtralots = 15 * Slotsize; if 0.0625 < intercept <= 0.125 then xtralots = 7 * &lotsize; if 0.125 < intercept <= 0.25 then xtralots = 3 * &lotsize; if 0.25 < intercept <= 0.5 then xtralots = &lotsize; if 0.5 < intercept then xtralots = 0; end; 200 when(minimumsampling = 2) do; * one/2 days; if 0 < intercept <= 0.0625 then xtralots = 15 * &lotsize; if 0.0625 < intercept <= 0.125 then xtralots = 7 * &lotsize; if 0.125 < intercept <= 0.25 then xtralots = 3 * &lotsize; if 0.25 < intercept <= 0.5 then xtralots = &lotsize; if 0.5 < intercept then xtralots = 0; end; when(minimumsampling = 3) do; * one/day max; if 0 < intercept <= 0.125 then xtralots = 7 * &lotsize; if 0.125 < intercept <= 0.25 then xtralots = 3 * &lotsize; if 0.25 < intercept <= 0.5 then xtralots = &lotsize; if 0.5 < intercept then xtralots = 0; end; when(minimumsampling = 4) do; * one/4 lots; if 0 < intercept <= 0.25 then xtralots = 3 * Slotsize; if 0.25 < intercept <= 0.5 then xtralots = &lotsize; if 0.5 < intercept then xtralots = 0; end; when(minimumsampling = 5 ) do; * one/2 lots; if 0 < intercept <= 0.5 then xtralots = &lotsize; if 0.5 < intercept then xtralots = 0; end; when(minimumsampling = 6) xtralots = 0; * one/let; when(minimumsampling = 7) do; *min 2 samples/lot; if intercept < 2 then samplesize = 2; xtralots = 0; end; when(minimumsampling = 8 ) do; *min 3 samples/lot; if intercept < 3 then samplesize = 3; xtralots = 0; end; when(minimumsampling = 9) do; *min 4 samples/lot; if intercept < 4 then samplesize = 4; xtralots = 0; end; 201 when(minimumsampling = 10) do; *min 5 samples/lot; if intercept < 5 then samplesize = 5; xtralots = 0; end; when(minimumsampling = 11) do; *min 6 samples/lot; if intercept < 6 then samplesize = 6; xtralots = 0; end; end; SVC = vc * samplesize; ^sampling variable cost; tsvc = tsvc + svc; *total sampling variable cost; b = b + xtralots; *end of the lot(s) tested if there are additional lots; totalseries = b; *totalsereis is b if out-of-control; %sainple; *takes a sample from lot(s) examined from a to b; %ana2yzemethod; if dpl{n} > 0 then rejectedlots •= rejectedlots + 1; else acceptedlots = acceptedlots + 1; beta = beta + samplesize; a = &lotsize + xtralots + a; b = a + &lotsize - 1; if &allseries < b then totalseries = &allseries; end; %mend dsmhq; * MACRO sample performs a random sample of size "samplesize" from the start of the lot(s) "a" to the end of the lots "b" and stores the values on the array samptest; san^le; do i = 1 to samplesize; tryother = 0; sampobs = (floor((b-a+1)*RANUNI(99)+a-l)); if 8 < sampobs <= &allseries-8 then do; do k = 1 to samplesize; if samptest{k} = sampobs then tryother tryother + 1; end; end; else tryother = tryother + 1; if tryother = 0 then samptest{i} = sampobs; else i = i - 1; end; %mend sample; * MACRO analyzedsmhq accumulates the number of total samples and defectives detected; * if there is a defective, alpha is adjusted and the Macro determinetype is called; %MACRO 202 %MACRO analyzemethod; n = n + 1; do mm = 1 to samplesize; totalsamples = totalsamples + 1; if series{samptest{mm}} > 0 then do; dpl{n} = dpl{n} + 1; defectivesdetected = defectivesdetected + 1; alpha = alpha + 1; % de termine tj^e; end; end; %mend analyzemethod; * MACRO determinetype is used to determine if the defective is a random event or an out-of-control situation. If out-of-control the Macro binarysearch is called; %MACRO determinetype; totalnext = 0; next{l} = samptest{mm} + 1 next{2} = samptest{mm} + 2, next{3} = samptest{mm} + 4 next{4} = samptest{mm} + 8 totalnext = series{next{1}} +series{next{2}} +series{next{3}} +series{next{4}}; if totalnext > 3 then do; outofcontrol == outofcontrol + 1; rejectedlots = rejectedlots + 1; %blnarysearch; end; else randomevents = randomevents + 1; %mend determinetype; * MACRO binarysearch starts with oc_found which is the point where the sampling method found the out-of-control situation and determines where the OC began; %MACRO binarysearch; hi = samptest{mm}; oc_found = hi; lo = 0; kk = 0; point = 0; check_oc = hi; start_oc = 0; do until (series{check_oc} < 1 ) ; check_oc = samptest{mm} - 2**kk; if check oc < 1 then check oc = 1; 203 kk = kk + 1; end; lo = check_oc; do until (lo > hi); point = floor((lo+hi) * 0.5); select; when(series{point} < 0.5) do; lo = point + 1; nd = nd + 1; end; when(series{point} >= 0.5) do; hi = point - 1; nd = nd + 1; end; end; end; start_oc = point; m = b - start_oc - nd; goto otherseries; %mend binarysearch; %MACRO REandOC; if REOC = 2 then do; if sig = 5 then do; %PE0C_5slgma; end; if sig = 6 then do; %REOC_6slgma; end; end; %mend REandOC; %MACRO REOC_6slgma; minimumsampling = 6; if vc = 2 AND fc < 55 then minimumsampling = 5; %mend RE0C_6sigma; %MACRO KE0C_5slgma; if cf = 1000 then minimumsampling = 10; else if vc = 2 then minimumsampling = 1; else if vc = 100 then minimumsampling = 11; else if fc = 5 then minimumsampling = 10; else if cf = 50000 then minimumsampling = 8; else if fc = 50 then minimumsampling = 10; else if fc = 200 then minimumsampling = 11; else minimumsampling = 200; %mend RE0C_5sigma; %MACRO slnglesample; do mm = 1 to &allsamplesize; samptest{mm} = 0; end; svc = vc * samplesize; tsvc = tsvc + svc; totalseries = b; %sample; % analyzemethod; %mend slnglesample; %MACRO single sampling; do until (b>&allseries); % slnglesample; 204 if dpl{n} < 1 then do; acceptedlots = acceptedlots + 1; accepted = 1; end; else do; rejectedlots = rejectedlots + 1; accepted = 0; end; a = &lotsize + a; b = a •: &lotsize - 1; if &allseries < b then totalseries = &allseries; end; %inend singlesampling; %MACRO doublesamplingr; do until (b>&allseries); % double sample; a = &lotsize + a; b = a + &lotsize - 1; if &allseries < b then totalseries = &allseries; end; %mend doublesampllng; %MACRO doublesample; accepted = 0; % slnglesample; if dpl{n} = 0 then do; acceptedlots = acceptedlots + 1; accepted = 1; end; else if dpl{n} >= 2 then do; rejectedlots = rejectedlots + 1; accepted = 0; end; else do; n = n - 1; % slnglesample; if dpl{n} = 1 then do; acceptedlots = acceptedlots + 1; accepted = 1; end; else do; rejedtedlots = rejectedlots + 1; accepted = 0; end; end; %MACRO %mend doublesample; multlplesampllng; do until (b>&allseries); ^multiple sample; a = &lotsize + a; b = a + &lotsize - 1; if &allseries < b then totalseries = &allseries; end; %mend multiplesampling; 205 %MACRO multlplesample; accepted = 0; % slnglesample; if dpl{n} >=2 then do; rejectedlots = rejectedlots + 1; accepted = 0; end; else do; n = n - 1; % slnglesample; if dpl{n} >= 2 then do; rejectedlots = rejectedlots + 1; accepted = 0; end; else do; n = n - 1; % slnglesan^le; if dpl{n} = 0 then do; acceptedlots = acceptedlots + 1; accepted = 1; end; else if dpl{n} >== 2 then do; rejectedlots = rejectedlots + 1; accepted = 0; end; else do; n = n - 1; % slnglesample; if dpl{n} = 0 then do; acceptedlots = acceptedlots + 1; accepted = 1; end; else if dpl{n} >= 3 then do; rejectedlots = rejectedlots + 1; accepted = 0; end; else do; n = n - 1; %slnglesaix^le; i f dpl{n) = 1 then do; acceptedlots = acceptedlots + 1; accepted = 1; end; else if dpl{n} >= 3 then do; rejectedlots = rejectedlots + 1; accepted = 0; end; else do; n = n - 1; % slnglesample; if dpl{n} = 1 then do; acceptedlots = acceptedlots + 1; accepted = 1 ; 206 end; else if dpl{n} >= 3 then do; rejectedlots = rejectedlots + 1; accepted = 0; end; else do; n = n - 1; % slnglesample; if dpl{n} = 2 then do; acceptedlots = acceptedlots + 1; accepted = 1; end; else do; rejectedlots = rejectedlots + 1; accepted = 0; end; end; end; end; end; end; end; %mend multlplesample; %MACRO SkSP2; skiplotcounter = 1; xtralots = 0; do until (b>&allseries); % slnglesample; if dpl{n} = 0 then do; skiplotcounter = skiplotcounter + 1; acceptedlots = acceptedlots + 1; end; else do; skiplotcounter = 1; xtralots = 0; rejectedlots = rejectedlots + 1; end; if skiplotcounter >= i_sksp then do; xtralots = inverse_f_sksp * &lotsize; acceptedlots = acceptedlots + inverse_f_sksp; n = n + inverse_f_sksp; end; a = &lotsize + xtralots + a; b = a + &lotsize - 1; if &allseries < b then totalseries = &allseries; end; %mend SkSP2; %MACRO ChSPl; chainlotcounter = 1; i_chsp = 4; do until (b>&allseries); 207 % slnglesample; if dpl{n} = 0 then do; chainlotcounter = chainlotcounter + 1; acceptedlots = acceptedlots + 1; end; else if dpl{n} >= 2 then do; chainlotcounter = 0; rejectedlots = rejectedlots + 1; end; else if chainlotcounter >= i_chsp then do; chainlotcounter = 0; acceptedlots ~ acceptedlots + 1; end; else do; chainlinkcounter = 0; rejectedlots = rejectedlots + 1; end; a = &lotsize + a; b = a + Slotsize - 1; if sallseries < b then totalseries = &allseries; end; %mend ChSPl; %MACR0 MILSTD105; level = 2; do mm = 1 to 10; MJLSTDhistory{mm} = 0; end; do until (b>&allseries) ; select; when(level = 2) do; % normalInspection; end; when(level = 1) do; %reducedlnspectlon; end; when(level = 3) do; % t l g h t e n e d l n s p e c t l o n ; end; end; a = &lotsize + a; b = a + &lotsize - 1; if &allseries < b then totalseries = &allseries; end; %mend MILSTD105; %MACRO reducedlnspectlon; samplesize = 20; % singles allele; if accepted = 0 then level = 2; %mend reducedinspection; %MACRO normal Inspect Ion; samplesize = 50; % doublesa mple; do ni = 10,9,8,7,6,5,4,3,2; MILSTDhistory{ni} = MILSTDhistory{ni-l}; end; MShl = accepted; normalcounter = 0; do ni = 1 to 10; normalcounter = normalcounter + MILSTDhistory{ni}; end; 208 if normalcounter = 10 then level = 1; normalcounter = 0; do ni = 1 to 5; normalcounter = normalcounter + MILSTDhistory{ni}; end; if normalcounter <= 3 then level = 3; %mend normalinspection; %MACR0 tlghtenedlnspectlon; samplesize = 80; % doublesample; do ni = 5,4,3,2; MILSTDhistory{ni} = MILSTDhistory{ni-l}; end; MShl = accepted; tightenedcounter = 0; do ni = 1 to 5; tightenedcounter = tightenedcounter + MILSTDhistory{ni}; end; if tightenedcounter = 5 then level = 2; %mend tightenedinspection; %generate(&sigma6); 209