This action might not be possible to undo. Are you sure you want to continue?

**Reliability Analysis Center
**

RAC is a DoD Information Analysis Center sponsored by the Defense Technical Information Center and operated by IIT Research Institute, Rome, NY

Ordering No.: STAT

Practical Statistical Tools for the Reliability Engineer

Q rh rqÃi)

SryvhivyvÃ6 hyvÃ8rr ! ÃHvyyÃT rr SrÃI`ÃÃ "##%( %

Vqr Ã8 hpÃ)

9rsrrÃTyÃ8rr Ã8yi 9T88QGD QPÃ7Ã %&#Ã7vyqvtÃ!

8yiÃPCÃÃ#"! %$

**Reliability Analysis Center
**

S68ÃvÃhÃ99ÃDs
hvÃ6hyvÃ8rr
Ã
rqÃiÃurÃ9rsrrÃUrpuvphy Ds
hvÃ8rr
ÃhqÃr
hrqÃiÃDDUÃSrrh
puÃDvrÃSrÃI`

6 rqÃs ÃQiyvpÃSryrhrÃ9v vivÃVyvvrq

The information and data contained herein have been compiled from government and nongovernment technical reports and from material supplied by various manufacturers and are intended to be used for reference purposes. Neither the United States Government nor IIT Research Institute warrant the accuracy of this information and data. The user is further cautioned that the data contained herein may not be used in lieu of other contractually cited references and specifications. Publication of this information is not an expression of the opinion of The United States Government or of IIT Research Institute as to the quality or durability of any product mentioned herein and any use for advertising or promotional purposes of this information in conjunction with the name of The United States Government or IIT Research Institute without written permission is expressly prohibited.

-ii-

00 20. reliability growth testing. to Washington Headquarters Services. Belvoir. statistical quality control. AGENCY USE ONLY (Leave Blank) 2. including the time for reviewing instructions. PERFORMING ORGANIZATION REPORT NUMBER Reliability Analysis Center 201 Mill Street Rome. demonstrating reliability. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. sampling. SPONSORING/MONITORING AGENCY REPORT NUMBER Defense Technical Information Center (DTIC-AI) 8725 John J. 0704-0188 Public reporting burden for this collection of information is estimated to average 1 hour per response. REPORT DATE 3. General probability and statistical concepts are explained and specific statistical tools are introduced by considering their applications to measuring reliability. NUMBER OF PAGES 120 Probability Statistics Reliability 17. SUBJECT TERMS 15.00 U. 2-89) Prescribed by ANSI Std. Arlington. Suite 1204. Form Approved OMB No. distribution unlimited. SECURITY CLASSIFICATION OF ABSTRACT 16. VA 22060-6218 11.S. NY 13440-6916 9. 14. DISTRIBUTION/AVAILABILITY STATEMENT 12b. including suggestions for reducing this burden. Directorate for Information Operations and Reports. PRICE CODE $75. existing data sources. and process improvement. ABSTRACT (Maximum 200 words) Unclassified This report provides basic instruction in statistics and its applications to reliability engineering. Kingman Road. DC 20503. 12a. FUNDING NUMBERS Practical Statistical Tools for the Reliability Engineer 6. LIMITATION OF ABSTRACT NSN 7540-01-280-5500 Standard Form 298 (Rev. searching.. SECURITY CLASSIFICATION OF REPORT 18. and completing and reviewing the collection of information. 201 Mill Street. $85. DISTRIBUTION CODE Approved for public release. AUTHOR(S) 65802S Anthony Coppola 7. NY 13440-6916. Rome. (Price: $75. Z39-18 298-102 -iii- . SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) STAT 10. Paperwork Reduction Project (0704-0188). REPORT TYPE AND DATES COVERED September 1999 4.). gathering and maintaining the data needed. Send comments regarding this burden estimate or any other aspect of this collection of information.S. Suite 0944 Ft.REPORT DOCUMENTATION PAGE 1. Washington. TITLE AND SUBTITLE 5. SECURITY CLASSIFICATION OF THIS PAGE 19. 1215 Jefferson Davis Highway.00 Non-U. SUPPLEMENTARY NOTES: SP0700-97-D-4006 Hard copies available from the Reliability Analysis Center. and to the Office of Management and Budget. 13. VA 22202-4302.

maintainability and quality data to enhance these data collection efforts. IIT Research Institute This material may be reproduced by or for the US Government pursuant to the copyright license under the clause at DFARS 252.The Reliability Analysis Center (RAC) is a Department of Defense Information Analysis Center sponsored by the Defense Technical Information Center. The RAC addresses both military and commercial perspectives and includes such reliability related topics as testability.iitri.mil © 1999.227-7013 (Oct. many are also targeted to managers and designers. as well as the components used in them. maintainability and quality information pertaining to systems and products. government laboratories and equipment users (government and industry).af. training. Total Quality Management and lifetime extension.org ALL OTHER REQUESTS SHOULD BE DIRECTED TO: Air Force Research Laboratory AFRL – Information Directorate Attn: R. and operated by IIT Research Institute (IITRI). and responses to technical and bibliographic inquiries. REQUESTS FOR TECHNICAL ASSISTANCE AND INFORMATION ON AVAILABLE RAC SERVICES AND PUBLICATIONS MAY BE DIRECTED TO: Reliability Analysis Center 201 Mill Street Rome. Automatic distribution lists.org http://rac. NY 13441-4505 Telephone: DSN: TeleFax: E-Mail: (315) 330-4857 587-4857 (315) 330-7647 hyler@rl. While most are intended to meet the needs of reliability practitioners. voluntary data submittals and field failure reporting systems supplement an intensive data solicitation program. The data contained in the RAC databases is collected on a continuous basis from a broad range of sources. including testing laboratories. analyze and disseminate reliability. RAC also offers reliability consulting. Users of RAC are encouraged to submit their reliability. device and equipment manufacturers. 1988) -iv- . RAC publishes documents for its users in a variety of formats and subject areas. managed by the Air Force Research Laboratory (formerly Rome Laboratory). Hyle 525 Brooks Road Rome. NY 13440 General Information: Product Ordering: Training Inquires: TQM Inquiries: Technical Inquiries: TeleFax: DSN: E-Mail: Internet: (888) RAC-USER (888) 722-8737 (800) 526-4802 (800) 526-4803 (800) 526-4804 (315) 337-9933 (315) 337-9932 587-4151 rac@iitri. RAC is chartered to collect.

many people. including engineers. Reliability Analysis Center (RAC) • 201 Mill Street. Rome.STAT Preface v PREFACE Statistical tools are powerful aids to reliability engineering and related disciplines. discussion of statistical theory will be kept to a minimum. However. NY 13440-6916 • 1-888-RAC-USER . and useful tools will be demonstrated by showing their practical application to various reliability engineering tasks. consider the process of mastering statistics to be painful. This book is an attempt to provide the reliability practitioner a reasonable capability in the use of statistical tools without the pain. For this reason.

Reliability Analysis Center (RAC) • 201 Mill Street.vi Practical Statistical Tools for the Reliability Engineer STAT This page intentionally left blank. NY 13440-6916 • 1-888-RAC-USER . Rome.

2 Ways to Measure Spread ..........3 The Weibull ...............2....................................2..................................................................................... 1............................1 The Standard Normal Distribution .......... 2........................................................2 The Versatile Weibull Distribution ...................................3.............................3................2.. 2........................1.... 2..............3 Introduction to Distributions................... 3..........1 Zero Failure Tests .................................. Page 1 1 2 3 4 6 7 7 8 9 11 12 13 13 13 15 16 18 19 20 22 22 22 23 23 23 25 25 25 27 27 28 32 33 33 36 40 41 41 43 43 44 46 48 2........................... 3..... 3.............. 1......... 4................................. 4.......2....................2........... 5...2 When Events are Mutually Exclusive.......3... 2........................ 3....................4 Measuring Reliability of "One-Shot" Products.....................................2...................................................................................... 3.......................3 The Hypergeometric Distribution ...1...........2 Various Other Useful Distributions.........STAT Table of Contents vii TABLE OF CONTENTS 1......... NY 13440-6916 • 1-888-RAC-USER ......................................................1 Testing for Trends.........................................................................................................................................2.................2........................................... SOME DISTRIBUTIONS AND THEIR USES..................................... 3.........................................4 Other Test Considerations ....................................2 The Exponential...........................................1......2..2........................1 Bayes’ Theorem................................................ 3. 4............................... 4...........0 3...............1 The Binomial Distribution.......................................... 3...............3 When Events are Not Independent ............2............................................................ 5............................... INTRODUCTION TO STATISTICS ................................0 5..............................................................................................2 The Normal Distribution’s Role in Sampling ............................................................. 3........... 3...... In Brief.....6 The Chi-Square Distribution......... 3........2......4 The Student t...............................1 The Lognormal ...............2......... 3........................................................................................................................................3..................2 Tests Allowing Failures .1..................2 Confidence Limits when the Failure Rate is Constant .................................1 Many Ways to be "Average".....0 WHAT YOU NEED TO KNOW ABOUT PROBABILITY..5 The F Distribution..................... 1.................. 1..2....... 4........................................................................ 4....................3 Measuring Reliability of Repairable Systems............1 General Principles......................................................... 5...............................................................1 Sequential Tests: A Short Cut... 3.....................1..........................................................5 For Further Study..1 The Normal Distribution...........2........................................................................................ 3...... 4......................................1 Controlling the Producer’s Risks ...........4 In Summary...0 Reliability Analysis Center (RAC) • 201 Mill Street..................................................3 Testing Under the Exponential Distribution......................................................2...............................0 4.............1 When Events are Independent ....... 3.......................................2 Continuous Distributions ..........................................................................2................. 5....................... 2................................... Rome........... 1.............2 The Poisson Distribution ..................4 Testing Hypotheses..............................................................................................1 Discrete Distributions . 5..........2....................................1 Caveats.... 5....................................3 In Summary......................................... 3........... DEMONSTRATING RELIABILITY ....... MEASURING RELIABILITY.....................

...................................... 6................................ 7........... but Risky.........0 RELIABILITY GROWTH TESTING ... NY 13440-6916 • 1-888-RAC-USER ..3. Appendix F: Critical Values of the F Distribution for Tests of Significance................................ 7.................................................. 7.........................3 Range Charts..................3 How Strong is the Correlation? .0 USING STATISTICS TO IMPROVE PROCESSES ..... 7......................................................................... Appendix A: Poisson Probabilities ....0 CLOSING COMMENTS........................ 7....................... Appendix B: Cumulative Poisson Probabilities...............1 Proportions.................................1...............3....... 9.....1................................................ 7................5................5 Controlling Attributes.4 Interpreting Control Charts ................1. Appendix D: The Chi-Square Distribution ...... 6..................1 Saturated Arrays: Economical.......................................5.....2 Rates ....................3.................................................................................1 Control Charts........................................................................ 7.......1 Measuring Quality from Samples ................................................................ 7.................................................2 Measuring Process Performance............................... 7. 7................2 Demonstrating Acceptability Through Sampling . Rome......1 Least Square Regression ........................................2 AMSAA Growth Analysis.........................................................................3........... 7.............................. 8............2 Testing for Robustness ................... 7.............................. 8...................................................viii Practical Statistical Tools for the Reliability Engineer STAT TABLE OF CONTENTS (CONT’D) 6............................................................................................1 Measuring Process Capability................ 6...................... 8........ 8............................................ 7.........................................................................3..............6....... 7...............2 Is There Really a Difference? ..................................................................3.....................1 Duane Growth Analysis..............1 Designing Experiments.......................................2 Control Charts for Variables..............................................1.........................3......................................................................... 8.............. 8................3........................................................................ Appendix E: The Student t Distribution......................3 Statistical Quality Control .................................................................1 Caveats............0 SAMPLING (POLLING) AND STATISTICAL QUALITY CONTROL........................................6 Caveat: "In Control" May Not Be "In-Spec" ...................................................................................... Appendix C: The Standard Normal Distribution .................6.3....................... Page 49 49 50 52 55 55 58 58 61 62 63 64 66 66 67 68 68 68 70 71 71 75 75 76 79 83 85 89 93 97 101 105 Reliability Analysis Center (RAC) • 201 Mill Street...............3........ 7................................................

............ Standard Normal Distribution ...................................................................................................................................................................... Continuous Distribution Curve for Height....... Process Capability (Cp) Chart . Run Chart ............. Practical O-C Curve ............................................................................................................................................................... Scattergram of Data in Table 8-10 ............................................... Ideal O-C Curve ..................................................................................................................................................................................................................................................................................... NY 13440-6916 • 1-888-RAC-USER ......................................... Rome........................................ Probability of Failure as Represented by the Area Under the Probability Density Function.............................................................................................................................................................. Process Performance (Cpk) Chart .................................................................. X and R Chart Combination .............. Distribution of Heights......... More Detailed Distribution of Heights.... "p chart" for Different Sample Sizes ................................... Page 10 18 18 19 21 27 32 44 47 52 59 59 62 63 64 67 69 70 80 81 Reliability Analysis Center (RAC) • 201 Mill Street.................. Weibull Plot ................................................................STAT Table of Contents ix LIST OF FIGURES Figure 2-1: Figure 3-1: Figure 3-2: Figure 3-3: Figure 3-4: Figure 4-1: Figure 4-2: Figure 5-1: Figure 5-2: Figure 6-1: Figure 7-1: Figure 7-2: Figure 7-3: Figure 7-4: Figure 7-5: Figure 7-6: Figure 7-7: Figure 7-8: Figure 8-1: Figure 8-2: Distribution of Heads in Two Tosses of a Coin .......................................................... Scattergram............................................ Typical Duane Plot................................... Control Chart........................................................ Devising a Reliability Test ....................................................................................................................................................................... Typical Sequential Test .........................

.......... 2................... Page 4 5 6 7 9 9 10 16 21 22 25 30 31 31 33 35 37 40 46 47 47 51 53 54 56 65 72 72 73 74 75 75 76 76 79 81 82 Reliability Analysis Center (RAC) • 201 Mill Street.............................. NY 13440-6916 • 1-888-RAC-USER ................... Spread Analysis..................................................................................................................... Paired Data ................................................. Comparison of Estimates ....................... Critical Values of z...................................................... Rome.... Confidence Interval Formulas .................... Growth Data Revisited ........................................... Critical Values of z................................................................................................. Orthogonal Array........................................................................................................................................... Extracts from Appendix A ................................................................................................... Completed Data Table............. Fixed-Time Reliability Tests............................................................................................................................................... Salary Data ................................................................................. Converted Data................................................................................. Data Analysis .................................................. Chi-Square Values................................................. Defect Data......................................... Failure Data .......................................05 Significance ...... Life Data..........................x Practical Statistical Tools for the Reliability Engineer STAT LIST OF TABLES Table 1-1: Table 1-2: Table 1-3: Table 2-1: Table 2-2: Table 2-3: Table 2-4: Table 3-1: Table 3-2: Table 3-3: Table 3-4: Table 4-1: Table 4-2: Table 4-3: Table 4-4: Table 4-5: Table 4-6: Table 4-7: Table 5-1: Table 5-2: Table 5-3: Table 6-1: Table 6-2: Table 6-3: Table 7-1: Table 7-2: Table 8-1: Table 8-2: Table 8-3: Table 8-4: Table 8-5: Table 8-6: Table 8-7: Table 8-8: Table 8-9: Table 8-10: Table 8-11: Known Data......................................................................................... Experimental Data...................................... Testing for Robustness ........................................................................................................ Summary of Distributions ........... Sequential Tests............. Sample Test Results ............................................. Critical Values for the Laplace Statistic.................................................................................................... Growth Data ................................................................................................................................................................................................................................................................... Expanded Test Matrix ................................................ Critical Values for F at 0.............................................................0 Discrimination Ratio ................................................................................. Summary of Section 1 ................................................................................................ Two Factor Orthogonal Array.................................................................................................................................................................................................................................................................................................................................................. Standard Normal Distribution Data.................................................................................................................................................................. Ordered Data ........ Statistical Constants ................................................ Three Factor Full-Factorial Array ...................... Sequential Test Plan for 10% Risks...................................................................................... Saturated Array (Table 8-2 Modified)...................................................................... Comparison of Results .....................................................................................

STAT

Table of Contents

xi

INTRODUCTION This book presents some basic material on probability and statistics and provides examples of how they are used in reliability engineering. To keep the book short and uncomplicated, not all subjects will be treated in detail and many more topics were ignored. Nevertheless, this text should help the novice reliability engineer understand the utility of probability and statistics, and can serve as a quick reference and refresher for the experienced engineer. It is important to remember that reliability engineering is not just the application of probability and statistics, and probability and statistics are not exclusively dedicated to reliability engineering. Reliability engineering is the science of designing products and processes to be reliable. Probability and statistics are simply tools that can help evaluate, predict, and measure reliability, among other uses. However, it is important for every reliability engineer to be able to use these tools effectively.

Reliability Analysis Center (RAC) • 201 Mill Street, Rome, NY 13440-6916 • 1-888-RAC-USER

xii

Practical Statistical Tools for the Reliability Engineer

STAT

This page intentionally left blank.

Reliability Analysis Center (RAC) • 201 Mill Street, Rome, NY 13440-6916 • 1-888-RAC-USER

STAT

Section 1: What You Need to Know About Probability

1

1.0

WHAT YOU NEED TO KNOW ABOUT PROBABILITY

Statistical inferences will often be expressed as probabilities. Probability can be defined as our degree of belief that an event will occur (e.g., I think I have an even chance of finishing this book on schedule). In statistical analysis, however, probability is usually defined as the expected frequency that an outcome will occur (e.g., nine times out of ten the actual value of a product’s failure rate will be within a range calculated by a particular method). This frequency may be stated as a percentile (e.g., 90% of the time the failure rate will actually be within the calculated range) or its decimal equivalent (e.g., there is a probability of 0.90 that the failure rate is really in the calculated range). Probability may also be referred to as "confidence" (e.g., there is a 90% confidence that the failure rate is within the calculated range). In dealing with probabilities, some useful relationships can be applied, depending on certain assumptions. 1.1 When Events are Independent If we know the probabilities of two events happening, and can assume that the events are independent (i.e., the occurrence of one does not increase or decrease the probability that the other will occur), then the probability of both events happening is: P(a and b) = P(a)P(b) where: P(a and b) = probability of both event "a" and event "b" happening P(a) = probability of event "a" happening P(b) = probability of event "b" happening For example, suppose an airplane uses a satellite receiver to track its position from the Global Positioning System (GPS) and can also track its position from a radio direction finder (RDF) receiver. If the GPS receiver fails once in 100 flights, the probability of losing the GPS tracking capability is 0.01 per flight. If the RDF receiver fails once every fifty flights, the probability of losing the RDF tracking capability is 0.02 per flight. If we can assume that failure of one does not affect the other (this is not a trivial assumption: both could fail simultaneously from some common cause, for example, lightning hitting the aircraft) then the probability of losing both position tracking systems on the same flight is: P = (0.01)(0.02) = 0.0002 Or, two times in 10,000 flights both position tracking systems will be out of service. (1-2) (1-1)

Reliability Analysis Center (RAC) • 201 Mill Street, Rome, NY 13440-6916 • 1-888-RAC-USER

0002.02) + (0. By convention.0 . whether it refers to a failure event or a non-failure event. Often.0198 + 0. Continuing the position tracking example.Q(g)Q(r) = 1.01) = 0. and Q(i) the probability of failure.99. Hence P(i) can be the probability of no failure or the probability of failure in the component identified as (i). an event can be defined either as the occurrence of a failure or as the lack of a failure.01 probability of no failure in RDF receiver = (1 .02) = 0.02 (1-5) Hence: P(s) = (0. and the assumption of independence.99)(0.0. the probability of either of two events happening is: P(a or b) = P(a) + P(b) (1-3) A useful fact is that the sum of the probabilities of all possible outcomes of an event must equal unity.99 probability of failure in GPS receiver = 0.e. NY 13440-6916 • 1-888-RAC-USER . the probability that both the GPS and RDF receivers will fail. The other possibility is Q(g)Q(r).9702 + 0.01)(0.98 probability of failure in RDF receiver = 0.98) + (0.01) or 0. Thus: P + Q = 1 or P = (1 .0002 = 0.0. Rome.01.0.9998 (1-6) Each of these events is mutually exclusive and they constitute all the "successful" situations. Reliability Analysis Center (RAC) • 201 Mill Street. which was computed by Equation 1-2 as equal to 0.99)(0. P(i) is generally used when only one notation is needed.Q) or Q = (1 .2 When Events are Mutually Exclusive When probabilities are mutually exclusive (i. when both notations are used in one formula.0098 = 0.0 . Further. then the probability of no failure is (1 .2 Practical Statistical Tools for the Reliability Engineer STAT 1. Since P + Q = 1. the probability of a flight without total loss of position tracking capability would be: P(s) = P(g)P(r) + P(g)Q(r) + Q(g)P(r) where: P(s) P(g) Q(g) P(r) Q(r) = = = = = probability of success (no total loss of all position tracking systems) probability of no failure in GPS receiver = (1 ..9998 (1-7) Note: in this example. the probability that an event will occur (P) plus the probability that it will not occur (Q) must also equal one.0. the occurrence of one event precludes the other). it is much easier to calculate one of the parameters (P or Q) than the other.98) = 0. P(i) is usually the probability of success (no failure). one parameter can always be found from the other. From Equation 1-4: P(s) = 1. since there are no other possibilities.P) (1-4) If the probability of a GPS receiver failure is 0.

Then. Similarly.STAT Section 1: What You Need to Know About Probability 3 We are following this convention. Suppose one-tenth of all failures in the GPS are due to external events. P(s) = P(g) + P(r) . Then: P(a and b) = P(a)P(b|a) where: P(a and b) = the probability that both events "a" and "b" will occur P(a) = the probability that event "a" will occur P(b|a) = the probability that event "b" will occur. Q(g)Q(r) in Equation 1-7 is identical in meaning to P(a)P(b) in Equation 1-1.1 and 1. The rationale for this is that P(g) includes all of the cases in which the GPS receiver is operating. shown in Equation 1-8.P(g)P(r) = P(g)P(r) + P(g)Q(r) + Q(g)P(r) which is the same result as Equation 1-5. defined as the conditional probability that event "b" will occur. we need a new term: P(b|a). Thus. Finally.3 When Events are Not Independent In the examples given in Sections 1.2. This is not always true. like lightning strikes. P(r) includes all of the cases in which the RDF receiver is operating. Rome. a failure of the GPS receiver does not change the probability that the RDF receiver will also fail. NY 13440-6916 • 1-888-RAC-USER . This is easily proven by decomposing P(g) and P(r) into mutually exclusive events: P(g) = P(g)P(r) + P(g)Q(r) P(r) = P(r)P(g) + P(r)Q(g) = P(g)P(r) + Q(g)P(r) Substituting Equations 1-9 and 1-10 into Equation 1-8 we get: P(s) = P(g)P(r) + P(g)Q(r) + P(g)P(r) + Q(g)P(r) . P(g) + P(r) twice counts the times that both the GPS and RDF receivers are operating. including both the times the GPS receiver is operating and the times it fails. even though this reverses the meaning of P(i) from the previous example. including both the times that the RDF receiver is operating and the times that it has failed.P(g)P(r) where all terms are as defined for Equation 1-5. and so these times must be subtracted to yield P(s). which also take out the RDF. consideration of mutually exclusive events leads to another solution for the case of independent events. given that event "a" occurs (1-12) (1-11) (1-9) (1-10) (1-8) Reliability Analysis Center (RAC) • 201 Mill Street. given that event "a" has occurred. our calculation of the probability of both receivers failing becomes more complicated. First. 1.

4 Practical Statistical Tools for the Reliability Engineer STAT Since "a" and "b" are arbitrary labels. where we assumed independence.001 [0.20 0. thus eliminating any effects of GPS failure.20 0. Since the events are not independent.90(0. Hence: P(b|a) = 0.017] = 0. any RDF failures must be from other causes. was 19 failures in 1.019)] = 0. exclusive of simultaneous failures. the overall probability of both units failing must go up. and Equations 1-12 and 1-13 would be identical in form to Equation 1-1. Suppose we found that the failure rate for the RDF. This means that for 90% of the GPS failures. Equation 1-12 can also be written: P(a and b) = P(b)P(a|b) (1-13) If the events were independent.10 + 0. Assume we have gathered some statistics as shown in Table 1-1.1 Bayes’ Theorem A noted derivation from conditional probabilities is Bayes’ Theorem.0). Table 1-1: Known Data Mission Profile Combat Training Transport Percent of Sorties Using Mission Profile 0. the probability of both receivers failing is: P(a and b) = P(a) P(b|a) = P(a)[0. If P(a and b) is the probability of both the GPS and RDF receivers failing.000 flights = 0. is higher in Equation 1-2 than it is in Equation 1-15.3.00017 = 0.00027 (1-15) (1-14) This result is higher than the 0. let us consider a radar installed in an aircraft. using the censored data.10 0. and event b represent a radar failure.019)] = 0.60 Probability of Radar Failure During Mission 0. we can convert the data in Table 1-1 to terms of probabilities and conditional probabilities. We know that one-tenth (10% or 0. To determine this probability.90(0. we must do a little more work.0002 found by Equation 1-2. P(b|a) = P(b) and P(a|b) = P(a). even though the RDF failure rate.019) Using Equations 1-12 and 1-14. Reliability Analysis Center (RAC) • 201 Mill Street.01 [0. which have some probability of occurrence whether or not the GPS has failed. Rome.1 + 0.90(0. and P(a) is the probability of the GPS failing. When some failure mechanisms take out both units simultaneously.019. NY 13440-6916 • 1-888-RAC-USER . we could search our records using just those flights when there was no failure of the GPS. 1.05 Letting event ai represent the probability of a sortie using a specific mission profile. For our discussion.10 (1) + 0.10) of all GPS failures are caused by factors that also kill the RDF (probability of RDF failure = 1.0001 + 0.10(1) + 0. Table 1-2 shows the converted data. then P(b|a) is the probability of the RDF failing on a flight when the GPS failed.20 0.

60 P(b|ai) 0. and there is a different conditional probability of event "b" happening for each event in set "a". P(a1|b) is the probability of the returned airplane having flown a combat profile.03 (1-21) 0. we must derive Bayes’ Theorem.44 0. We can use the information to calculate the probability that the sortie was a combat mission. given that it came back with a failed radar.20 0.20) + (0.04 = 0.05 Suppose we are given the information that an aircraft came back from a sortie with a failed radar.20)(0. To do so. NY 13440-6916 • 1-888-RAC-USER .20) 0.05) 0.20 0.02 + 0. Equation 1-20 is Bayes’ Theorem.20 0. Since P(a and b) = P(a|b) P(b) = P(b|a) P(a). The other terms are quantified in Table 1-2.10) + (0.09 Reliability Analysis Center (RAC) • 201 Mill Street. the solution we seek is: P(a1|b) = P(b|a1) P(a1)/P(b) (1-18) (1-17) (1-16) Since event "a" is a set of mutually exclusive events.04 = (0. Rome.20)(0. it follows that: P(ai) P(b) = P(ai|b) P(b) = P(b|ai) P(ai) Therefore: P(ai|b) = P(b|ai) P(ai)/P(b) Since P(a1) is the probability of a combat mission.STAT Section 1: What You Need to Know About Probability 5 Table 1-2: Converted Data i 1 2 3 Mission Profile Combat Training Transport P(ai) 0. Substituting the data from Table 1-2 into Equation 1-20: P(a 1 | b) = (0.10 0.04 + 0. the total probability of event "b" happening is: P(b) = Σ P(b|ai) P(ai) Substituting Equation 1-19 into Equation 1-18: (1-19) P(a 1 | b) = P(b | a1 ) P(a1 ) ∑ P(b | a i ) P(a i ) (1-20) In the equation.60)(0.20)(0.

because it comes after the new data is considered.20. NY 13440-6916 • 1-888-RAC-USER . because it comes before the gathering of additional data (i. Table 1-3: Summary of Section 1 When events are: Independent Mutually exclusive And the following apply: The occurrence of one event has no effect on the occurrence of the other The occurrence of one event precludes the other P(a) = probability event "a" occurs Q(a) = probability event "a" does not occur The occurrence of one event may affect the other One event may have several different outcomes. Some applications that depend on "subjective priors" (i.4 In Summary Table 1-3 summarizes the material presented in this section. The 0. that the radar failed on the mission).20 figure is called the "prior" estimate.e.44. The information that the radar failed raises our estimate to 0. each affecting the other event differently You can use: P(a and b) = P(a) P(b) P(a or b) = P(a) + P(b) .P(a and b) P(a or b) = P(a) + P(b) P(a) + Q(a) = 1 P(a) = 1 . we would have estimated the probability of the aircraft having just flown a combat profile at 0.Q(a) Q(a) = 1.44 figure is called the "posterior" estimate. Bayes’ Theorem is the foundation of both useful and dubious analyses combining "prior" data with information from statistical sampling to produce a "posterior" estimate..P(a) P(a and b) = P(a) P(b|a) = P(b) P(a|b) Not independent P(a 1 | b) = P(b | a 1 ) P(a 1 ) ∑ P(b | a i ) P(a i ) Reliability Analysis Center (RAC) • 201 Mill Street. and the 0. the known information is an opinion or assumption rather than a conclusion from a set of statistics) can be questionable. Rome. The theorem is quite correct.6 Practical Statistical Tools for the Reliability Engineer STAT Without the knowledge of the radar failure. 1.. based on the data in the second column of Table 1-1.e.

1 Many Ways to be "Average" The word "average" is used to describe a central point for the values taken by a random variable.) A variety of statistical tools are used to convert statistics (i. and the different definitions are based on the different uses of the measure..STAT Section 2: Introduction to Statistics 7 2.g. The term “random variable” refers to the function that describes the values the statistics can take. (A case could be made that the constraints on a random variable make it neither truly random nor greatly variable. One measured value is called a "statistic" and more than one are called "statistics". The company is interested in what it pays for reliability engineers and calculates the "average" using the arithmetic mean (sum the values and divide by the number of data points). the average salary for a reliability engineer) or with providing the means for making useful inferences about something (e. However.. but we shall have to take the term as given. The engineers usually figure the average salary as $40. 1.000 20.) The company claims its reliability engineers earn an average of $50.000 40.000. 2. To illustrate. ordered from smallest to largest. The number of heads in two tosses of a coin is a random variable that can take the values 0.000 125. These products are based on measured values.. Reliability Analysis Center (RAC) • 201 Mill Street.000 20.0 INTRODUCTION TO STATISTICS "Statistics" as a mathematical discipline is concerned with describing something in useful numerical terms (e.000 50. such as the number of defects in a sample of parts.e.000 75. measurements) into the desired descriptions or inferences. Common to all of these tools will be the terminology and concepts discussed in this section. For example.000. more specific measures are needed in statistical analysis.000 What is the average salary for a reliability engineer in the company? (We suggest you form your own opinion before going on. Rome. the outcome of a toss of a coin is a random variable that can take the value "heads" or "tails". NY 13440-6916 • 1-888-RAC-USER . And they are all right! The differences arise from different definitions of "average". This gives their cost per capita for reliability engineers. or 2. These will be discussed in the following sections organized by the use of the results. consider the following (completely contrived) data shown in Table 2-1 which we will say represents the salaries of reliability engineers in a mythical company. the percent of a product that will be outside specified limits). and the company union claims it's $20.000. Table 2-1: Salary Data $20.g.

However. our results would tend to zero. i. sum these results and divide by the number of units measured.g.2 Ways to Measure Spread Besides measuring central tendency. In most of the methods described below. Hence. We seldom reject a group (or a "lot") of products because some units measure outside the specified range. we would want the mean length of a production sample to be close to one inch. which do exist but will not be needed for our purposes.. but the individual rods varied from one-half inch to one and one-half inches in length. suppose we measured the length of a sample of rods. we could not merely subtract each measurement from the mean (or the mean from each measurement). we would probably not be happy. "rods shall be one inch long plus or minus 0. the range. The union uses the mode (the value most frequently measured) as its "average" because it represents the most people. We do reject groups of products that we feel have too many units outside the specified range. NY 13440-6916 • 1-888-RAC-USER . the 50th percentile). This establishes a desired limit on spread. if we were producing rods designed to be one inch long. Thus the definition of "average" depends on its use. if the mean length were one inch. as shown in Table 2-2.0011 inches always bad?).001 inch"). However. Any "average" measure not using the arithmetic mean will be noted when it occurs.0009 inches always good and 1. we are usually concerned with measures of spread. perhaps. and others. the arithmetic mean will be the definition of choice. in specifying or measuring product parameters. defined as the difference between the highest and lowest values of the parameter of interest. we need another measure of spread. if a somewhat arbitrary one (is 1. For example. To determine what proportion of the product group is outside the specified range.e. or spread. harmonic mean. The most obvious measure of spread is. Rome." ignoring the existence of the geometric mean. and calculated the difference between each measurement and the mean.8 Practical Statistical Tools for the Reliability Engineer STAT The engineers are interested in their standing vis a vis each other and compute the "average" based on the median (the value for which there are as many data points above as below. we often want to measure variation. Specified values most often include a stated range or "tolerance" around the design value (e. 2. Because some measurements will be higher than the mean and others lower. For example. We will call this simply "the mean.. measure the parameter of interest in each unit of the sample and compare these measurements to the mean value of the measurements. One way of measuring spread would be to take a sample of the product. Reliability Analysis Center (RAC) • 201 Mill Street.

Gaming examples are quite analogous to some engineering applications. One measurement of spread. ∑ (x . summing the squares and dividing by the number of measurements. where we are interested in the odds of making a wrong decision based on a set of statistics.50 5.50 0 0 We therefore need a better way to measure spread.10 0 +0. called a histogram.3 Introduction to Distributions We have discussed the random variable for the number of heads occurring in two tosses of a coin. the more spread there is to the data. called the variance. A more convenient measure is the standard deviation.50 -0. One such experiment would not be any help.000.00 1. so we could establish appropriate odds.10 +0. is formulated by subtracting the mean value from each measurement (x .90 1. is shown in Figure 2-1. We could use the absolute values of the differences between the individual measurements and the mean.x )2. say 1. There are several ways to do this. The larger the standard deviation.STAT Section 2: Introduction to Statistics 9 Table 2-2: Spread Analysis Sample 1 2 3 4 5 Sum of all Data Mean Value Rod Length in Inches 0. Suppose we are interested in wagering on the outcome of two tosses. but a more common alternative is to use squared values.00 1.00 Deviation from Mean Value -0.) Returning to our example.10 1. we could experiment by tossing a coin twice and counting the number of heads. because it graphically shows the relative frequencies. NY 13440-6916 • 1-888-RAC-USER .x )2/n. Table 2-3: Experimental Data Outcome (Number of Heads) 0 1 2 Frequency (in 1000 Experiments) 238 509 253 A better way.x) n 2 = (2-1) 2. but many replications. which is merely the square root of the variance (Equation 2-1). One is simply as shown in Table 2-3. squaring the results (x . which was actually created to analyze gaming odds.x ).50 0. (Note: the use of gambling examples is common in statistics. ∑ (x . We would then have to organize these statistics into a useful format. would give a useful set of statistics. We would want to know the probabilities of all the possible outcomes. Rome. Reliability Analysis Center (RAC) • 201 Mill Street.

The probability of getting two heads is (0.5 on each toss.0.5)(0.10 Practical Statistical Tools for the Reliability Engineer STAT 509 Frequency In 1.50.5) = 0.5) + (1 . if the computed probability holds. we shall compute the same probabilities from a theoretical approach.5)(1 .5). as shown in Table 2-4.25.0.25 The experimental results show a little bias towards getting more heads than tails. Let us now compare these to our experimental results.38 against $7.0. From probability theory.0. This means the probability of a head is 0. but the longer expression shall be used in the following formulas so that it will be clear which probability is meant. For example. and that our method of tossing is also unbiased. which is the same as 0. We would therefore expect heads or tails to be equally likely.5.62 and should break even in the long run.5) = 0. meaning that it has no bias towards falling either heads or tails.238 to (1 .0. and the area of each column equal to the probability of the outcome it represents. using probability theory.0.238 so the odds are 0. Table 2-4: Comparison of Results Outcome (Number of Heads) 0 1 2 Experimental Results 0.000 Trials 253 238 0 1 2 Number of Heads Figure 2-1: Distribution of Heads in Two Tosses of a Coin The histogram of Figure 2-1 shows a frequency distribution.25 0. which always exists.253 Theoretical Results 0.5)(1 . Rome. The next step towards determining the odds on each outcome would be to convert the frequency distribution to a probability distribution. This could be done by dividing the value of each column by the number of experiments.25. NY 13440-6916 • 1-888-RAC-USER . the probability of no heads in two tosses is (1 . Many of the Reliability Analysis Center (RAC) • 201 Mill Street. The probability of one head in two tosses is the sum of the probability of getting heads on the first toss and not on the second and the probability of getting heads on the second toss and not the first: (0.238 0. One betting on the outcome could put up $2. for comparison. This makes the total area enclosed by the data equal to one. but would you bet on it? The difference could just be experimental error.762. and the probability of not getting a head is (1 .509 0.5) = 0. Let us assume our coin is fair. the probability of no heads is 0. The odds for a fair bet are then simply the ratio of the probability of the outcome to one minus the probability. Now.238) or 0.5)(0.50 0.

Many different statistical distributions will be useful in making such decisions. This is another correct result. the alternate hypothesis is that it does not have that MTBF. times between failures). This is called the null hypothesis.. etc. It involves using statistical analysis to come to a conclusion to accept or reject a stated hypothesis with a known risk of being in error. With each null hypothesis is an associated alternate hypothesis. failure rate. This is called a Type II error and its probability of occurrence is conventionally called " β ". the "producer's risk" defined in Section 5 and Section 7 would be equal to α since it is the probability that a customer will not accept a good product. This is called a Type I error in hypothesis testing. The "consumer's risk" discussed in Sections 5 and 7 represents the probability of accepting a bad product. will be discrete distributions. " α " is the probability that we would conclude the product is not acceptable. Alternate hypothesis is true and is accepted (null hypothesis rejected). and is equal to β when the rejected alternate hypothesis is that the product is bad (i. If the null hypothesis is that the product has a parameter (e. • • • Reliability Analysis Center (RAC) • 201 Mill Street. The probability of it occurring is conventionally called " α ". Alternate hypothesis is true and is rejected (null hypothesis accepted). In this case.3. We will sort these out in Section 3.g. Rome.. having only integers (e. failure rate. but whether we like it or not. In any event.e. If the null hypothesis is that a product has a poor MTBF. number of defects. et al) which is acceptable.). there are other ways. we do want a correct conclusion. we would probably prefer it to be refuted.. Null hypothesis is true and is rejected. Instead of trying to decide whether or not we have a fair coin. has an unacceptable MTBF. which may be the desired result or not. Some. and whether we prefer the null hypothesis or the alternate hypothesis. This is usually merely the negation of the null hypothesis. number of failures) as possible outcomes. number of defects. NY 13440-6916 • 1-888-RAC-USER . 2.g. If the null hypothesis is that a product has a certain MTBF. we will try to decide such things as how many failures it will take to convince us that a product is not as reliable as we expected. we want a correct result.STAT Section 2: Introduction to Statistics 11 methods described in this book will be concerned with separating experimental error from true indications of bias. In sequential reliability testing (Section 5. MTBF. which has some applications to reliability engineering that we will cover later in Sections 4.4 Testing Hypotheses One of the specialties of statistics is called hypothesis testing. like the one shown in our example. each hypothesis test can have any of four results: • Null hypothesis is true and is accepted. Others will be continuous distributions. 5 and 7. However. A hypothesis might be that a product has a given mean time between failures (which can be one we want it to have or one we don’t want). where an infinite delineation in outcomes is possible (for example.1) the null hypothesis is that a product has a specific MTBF that is considered desirable and the alternate hypothesis is that it has a specific MTBF that is considered undesirable. This is a correct result.

the hypothesis test is performed by defining an acceptable value of " α " and comparing sample data to a distribution representing the expected results when the null hypothesis is true. DC. New York. as we shall see in Sections 5 and 7. the null hypothesis is accepted. Kiemele & S. Advanced: • • Methods for Statistical Analysis of Reliability & Life Data.R. NY. Statistical Methods in Engineering and Manufacturing.S. Milwaukee.J. Military Handbook 338-1B. Schmidt. If the distribution would produce the sample data with a probability of α or less. New York. Singpurwalla. NY 13440-6916 • 1-888-RAC-USER . 1997. M. ranked by increasing level of difficulty. Hypothesis tests can also be defined based on " β " and on both risks. The Cartoon Guide to Statistics. Intermediate: • • Basic Statistics. 1974. John E. HarperCollins Publishers. Stephen S. Willoughby. Department of Defense.12 Practical Statistical Tools for the Reliability Engineer STAT Basically. Electronic Reliability Design Handbook. are: Introductory: • • Probability and Statistics. the data is considered sufficiently unlikely to have come from the distribution and the null hypothesis is rejected. Air Academy Press. Silver Burdett Company. 1968. Larry Gonik & Woollcott Smith. 1990. John Wiley & Sons. Colorado Springs. Otherwise. R. Schaefer & N. 2. WI. 1993. Rome. Quality Press. Washington. NY. N. U. 1990.5 For Further Study Some recommended references in probability and statistics. Mann.D. Reliability Analysis Center (RAC) • 201 Mill Street.E.R. NJ. Brown. CO. Morristown.

or one success.1 Discrete Distributions Discrete distributions are concerned with random variables which are integers.x" failures.x)! (3-1) is a counting formula giving the total number of different ways one can have exactly "x" successes in "n" trials. There could be "x" successful trials followed by "n . and more than one distribution may describe a set of data.1. such as the probability of no failures. However.1" successes.x) n! p x q (n . We will use these distributions to determine the probabilities that certain outputs will be experienced. and the usefulness of the results depends on the assumption being reasonable.x) x! (n . Rome. of less than "x" failures or more than "y" failures. (Note: success of a trial could be the occurrence of a product failure and failure of a trial the non-occurrence of a product failure. there is more than one way for this to happen. It assumes that the probability of an event is the same in every trial. The term n! x! (n . Since each way is presumably equally likely and its probability of Reliability Analysis Center (RAC) • 201 Mill Street. This. Either a person is over six feet tall or he is not.p) is simply the probability of "x" successes and "n . then "x . Each of these is a mutually exclusive way of getting the result of interest. Analysis of a distribution is invariably advantageous. etc.1 The Binomial Distribution The binomial distribution. either a product has failed or it has not. There are non-parametric methods that do not assume any distribution for the data. requires the assumption of a distribution for the data. or vice-versa). etc.STAT Section 3: Some Distributions and Their Uses 13 3.x" failures in The product p times q "n" trials. however. 3. as the name implies. depending on the problem addressed. the distribution of the lengths of a lot of rods may follow a normal distribution. The most useful discrete distributions are the binomial and the Poisson. The hypergeometric distribution is also of interest to reliability engineers. of course. "n .0 SOME DISTRIBUTIONS AND THEIR USES Statistical analysis of practical problems often requires consideration of the distribution of the data. So there are a variety of distributions we may find useful. Of more interest. Then: f(x) = where: f(x) = probability of exactly "x" events occurring in "n" trials p = probability of a successful trial (success = the event happened) q = probability of an unsuccessful trial = (1 . As an example.x)! (3-2) x (n . is concerned with "yes-no" outcomes.x" failures. either because it is simpler or because it gives more precise results. NY 13440-6916 • 1-888-RAC-USER . No one distribution fits all data. 3. but the number of rods with lengths out of specified limits is described by a binomial distribution.

Noting that n = number of engines = 2.P(x + 1 or more) (3-6) To illustrate.0) = (1)(0.14 Practical Statistical Tools for the Reliability Engineer x (n .10) 0 (0.09) = 0. To answer such questions. Let us assume that experience indicates that the frequency of engine failures is 10% per flight. and the probabilities of these events are mutually exclusive. or "x or more" successes. suppose an airplane with two engines could still fly safely with one engine failed. Equation 3-1 becomes: P(0) = 2! 2 x1 (0. we note that we can have from zero to "n" successes in "n" trials.90) 2 = 0.10 and q = 0. NY 13440-6916 • 1-888-RAC-USER .98 1(2 x 1) 1(1) P(1 or less) = (3-7) Reliability Analysis Center (RAC) • 201 Mill Street. For example.81) + 2(0.81 0! (2 . rather than merely "x".x) STAT occurrence equal to p times q . we get: 2! 2! (0. to use Equation 3-1 to determine the probability of an aircraft with two engines getting through a flight with no engine failures. The probability of a successful flight would then be equal to the probability of one or less failures occurring. Rome. it is useful to remember: P(x or less) = 1 .10. We may then want to know the probability of having more than two failures during a mission.90) (2 . Or.0)! 1! (2 .x)! (3-5) As always. in testing one-shot devices. Hence. q = probability the engine will not fail during a flight).90. the probability of "x or less" events is: P(x or less) = ∑ x n! p x q (n − x) 0 x! (n .0) + (0.x)! (3-4) And the probability of "x or more" events is: P(x or more) = ∑ n n! p x q (n − x) x x! (n .0)! 1 (2 x 1) (3-3) Quite often we are interested in the probability of getting "x or less". Therefore.90) = (0. as before. we may have six radios in a cockpit and only need four for a successful mission. we can define p as the probability that a randomly selected engine will fail during a flight (hence.10)(0.10)1 (0. For example.18 = 0.81 + 0.1)! 2 x1 2 x1 = (1)(0. Using Equation 3-4.1) 0! (2 . p = 0.90) (2 .10) 0 (0. multiplying the two expressions gives the total probability of exactly "x" successes in "n" trials. and letting p = probability of engine failure = 0. we may want to know the probability of passing a test of 50 firings with no more than two failures allowed. and x = number of failures we are interested in = 0. such as missiles.90) 2 + (0.90) (2 .

2 The Poisson Distribution The Poisson distribution can be considered an extension of the binomial distribution when "n" is infinite. a recent failure does not make another failure more or less likely)..3. Under these assumptions..1 may be used to test for constancy. the probability of zero failures is: (3-8) When a = P(0) = e -λt (3-9) Equation 3-9 is a basic equation in reliability engineering. When the assumptions of a constant failure rate and independent failures do hold. the methods in Section 6. 3. It calculates the probability of a product completing a mission of time "t" without failure (under the assumptions that is constant with time. etc. Rome. or roses sold in a flower shop. may be more appropriate. per Equation 3-6. Reliability Growth Testing.STAT Section 3: Some Distributions and Their Uses 15 The same result would have been obtained by finding the probability of two engine failures. for an expected number of events "a" the probability of getting exactly "x" events is: f(x) = a x e −a x! t. using Equation 3-5 and subtracting this from one. and that failures are independent events). the trend test in Section 4.g. We can use the Poisson distribution to determine the probability of passing a test run for a fixed time with an allowable number of failures by the expression: a x e −a P(n or less) = ∑ x! 0 n (3-10) where: n = the number of failures allowed The probability of failing the test would be: P(n + 1 or more) = ∑ a x e −a x! n +1 ∞ (3-11) Reliability Analysis Center (RAC) • 201 Mill Street. These events could be product failures.1.g. we can determine an expected number of events in any given interval (e. NY 13440-6916 • 1-888-RAC-USER . the expected number of failures in a mission is the failure rate " " times the mission length "t"). Should a non-constant rate apply. It assumes events occur at a constant average rate. If there is any doubt that events are occurring at a constant rate. The number of events occurring in any interval is independent of the number in any other interval (e. as the reader may wish to verify.

3: 0. as "p" approaches zero and "n" approaches infinity. It may be useful to note that the Poisson and binomial are related in that the expected value "a" in the Poisson distribution can be expressed as the probability of an event "p" times the number of opportunities or trials "n". For example.9963. We assume the failure rate is constant.2222 0.7408 + 0.2222 + 0.3. having 2 or less failures) = 0. which we could also have found directly using Appendix B.3. 3. Table 3-1: Extracts from Appendix A Number of Failures (x): 0 1 2 Probability of (x) Failures When (a) = 0. If we assume ( t) = 0. For any given value of "a". the occurrence of a defective unit in a sample does not increase or decrease the probability that the next unit in the sample will be defective. this assumption is reasonable. where "p" and "n" are terms of the binomial. solutions to Equation 3-8) have been tabulated.3 The Hypergeometric Distribution The binomial and Poisson distributions assume independent events. the probability of passing the test (i.0333 Hence.∑ 0 n a x e −a x! (3-12) It is not necessary to compute Equations 3-8. Tables of Poisson probabilities (probability of "x" events when "a" are expected. Rome.e. The reader may use Equation 3-12 to convert these to solutions of Equation 3-11. Appendix B provides a table with solutions to Equation 3-10. For example. when the population is relatively small compared to the sample size (say no larger than 10 times the sample size) and units sampled are not replaced. Also tabulated are cumulative Poisson probabilities giving the probabilities of "x" or less events when "a" are expected (solutions to Equation 3-10).1. suppose we tested repairable products and would consider the product acceptable if no more than two failures occurred during the test. the expected number of failures for the product would be ( t). assume an office with six employees. i. and (x) = 2. NY 13440-6916 • 1-888-RAC-USER . One such table is presented in Appendix A.. and even cumulative probabilities of "x" or more events (solutions to Equation 3-11).16 Practical Statistical Tools for the Reliability Engineer STAT or more practically: P(n + 1 or more) = 1 . If the product had a failure rate ( ) and was tested for the test time (t). However it is easier to use the tables. what is the probability of the product passing the test? The solution can be found from Equation 3-10.7408 0. As an example. However.. the result of the binomial formula (Equation 3-1) approaches the result of the Poisson formula (Equation 3-8). Where the population being sampled is large in relation to the sample size. the assumption may not be reasonable. 3-10 or 3-11. two of Reliability Analysis Center (RAC) • 201 Mill Street.e. where (a) = 0. or when sample units are replaced after selection and can be drawn again.0333 = 0. Table 3-1 is extracted from Appendix A.

n)! where: N = number of units in the population being sampled D = number of units with the characteristic of interest (event happens) in the population being sampled n = number of units in the sample (sample size) x = number of events in the sample (units with the characteristic of interest) The numerator of Equation 3-13 is the number of ways we can select "x" units from the "D" units in the population that have the characteristic of interest times the number of ways we can select (n .2]! 2 * 1 * 2 *1 P(0) = = = 2! 2! = 6! 6! 6! 6 * 5 * 4 * 3 * 2 *1 2! (6 . From Equation 3-13 the probability of selecting no women for the survey would be: 2! (6 .x)! (n . The probability of "x" events happening in our sample (number of woman selected. The denominator is the number of ways we can select a sample of "n" units from a population of "N" units.x)![(N .STAT Section 3: Some Distributions and Their Uses 17 whom are women.D)! x! (D . the probability of the second also being a woman is one in five. To handle such situations.D) .(n .0)! (2 . If the first selection were a man. For our office example.x) of the (N . If the first selection is a woman.0)![(6 . or one-third. the samples are not independent. (3-13) Reliability Analysis Center (RAC) • 201 Mill Street.2)! 2! 4! 4! 4 * 3 * 2 *1 0! (2 . Suppose we wish to select two at random for a survey of job satisfaction.) is: D! (N . N = six people.0)]! 2! 2![4 .(2 . we can use the hypergeometric distribution. The distribution is essentially a combination of three counting formulas like that of Equation 3-2. etc.x)]! P(x) = N! n! (N . Hence.2) . NY 13440-6916 • 1-888-RAC-USER . the probability of the second being a woman would be two in five or two-fifths. number of green apples in a bag.2)! 2! (4)! 2! 4! 2 *1 * 4 * 3 * 2 *1 24 6 (3-14) = 4 = = 0.D) units without the characteristic of interest. D = two women. and n = two people selected for the survey. The probability that the first selected will be a woman is two in six.4 720 15 48 Four times out of ten we will conduct the survey with no women in the sample. or onefifth. Rome.

Figure 3-2: More Detailed Distribution of Heights As our sample size approaches infinity and our measurement intervals become infinitely small. After enough statistics had been collected. the histogram approaches a smooth curve. NY 13440-6916 • 1-888-RAC-USER .2 Continuous Distributions Suppose we wanted to describe the height of male adults in a community. We could measure the heights of a random sample of adult males (it would not do to measure only basketball players. our histogram will become more like Figure 3-2. and our measurements become more precise. Reliability Analysis Center (RAC) • 201 Mill Street. we would have to organize the data in a meaningful format. as shown in Figure 3-3. Rome. for example). Figure 3-1: Distribution of Heights If our sample size increases. and might elect to use a histogram as shown in Figure 3-1.18 Practical Statistical Tools for the Reliability Engineer STAT 3.

2. we shall be dealing with probability distributions. is the height of the curve at the value "x". the distribution of times to failures. This in itself is of little value. NY 13440-6916 • 1-888-RAC-USER . This is equivalent to the probability that a randomly selected product will operate without failure for time = t. designated by f(x).STAT Section 3: Some Distributions and Their Uses 19 Figure 3-3: Continuous Distribution Curve for Height While many continuous distributions exist. designated by F(x). meaning that the distribution is normalized so that the total area under the curve is equal to one. Then F(t) = proportion of products which will fail before time = t (see Section 3. we shall be concerned with three parameters: • The probability density function. Since the Poisson becomes approximately equal to the binomial when the number of trials (n) is high and the probability of an event (p) is low. For example.1). let f(x) = f(t). can be described by it. for example in considering distributions of times to failure. For symmetrical continuous distributions such as the normal.2. It is simply 1 . the binomial distribution can be approximated by a normal distribution when the number of binomial trials (n) is high (30 or more). are usually not symmetrical. More generally. • • 3. The cumulative density function. the mean and standard deviation (Equation 2-1) of the distribution provide all the information we need. such as the physical dimensions of a product. In either case.2 for an example). designated by R(t). except as the foundation for the other two parameters. The reliability function. In addition. is the proportion of products which have not failed in time = t. and many other parameters. A particularly important characteristic of the normal Reliability Analysis Center (RAC) • 201 Mill Street. is the integral of f(x). Many parameters of interest. The utility of this is that we can then calculate the probability of a measurement being in any given range of values.F(t). This provides the area under the curve from minus infinity to x. usually follow a symmetrical bell-shaped curve called the Normal or Gaussian distribution (to be discussed in Section 3.2. Rome. Other distributions of interest in reliability engineering. such as distributions of times to failure.2. when f(t) is a distribution of failure times. it can also be approximated by the normal distribution under these conditions. the normal distribution is undoubtedly the most useful overall.1 The Normal Distribution While many continuous distributions are of interest for specific applications. the distribution of heights.

The standard deviation of the distribution can be estimated from the standard deviation of the sample (Equation 2-1).1. there can be an infinite number of normal distributions.1. In such cases. However.x) n -1 2 S = (3-15) Since both the mean and the standard distribution can take any value. For example. such as the one presented in Appendix C. The normal distribution is completely defined by its mean and standard deviation. when the number of samples is low. Data is converted from the actual distribution to the standard normal by the formula: z = x- (3-16) where: = = = = the mean of the distribution being converted the standard deviation of the distribution being converted a data point from the distribution being converted the corresponding data point on the standard normal distribution x z Reliability Analysis Center (RAC) • 201 Mill Street. median and mode (see Section 2.2. The percent of the population of interest under any portion of the curve can be determined from the mean and standard deviation. making the fit adequate. the percent of a product which falls outside specified limits. for which the relation of the area under the curve to the distance from the mean (in standard deviations) is available in tables. but the probability of a value being far in the tails of the curve is so small that it may be neglected.2). A normal distribution is symmetrical. analysis is facilitated by converting data of interest to a standard normal curve.1) are identical. this gives a biased estimate.2. and the distribution theoretically extends from minus infinity to plus infinity. Or.1 The Standard Normal Distribution The standard normal distribution is one in which the mean is zero and the standard deviation is 1. they are estimated from sample data. 3. the percent of adult males who are over six feet tall. However. Since the mean and standard distribution of a population are not usually directly measurable.20 Practical Statistical Tools for the Reliability Engineer STAT distribution is that it applies to data from samples. of more interest to reliability engineers. NY 13440-6916 • 1-888-RAC-USER . The latter obviously could not apply to a distribution of heights. the population standard distribution is estimated from the unbiased expression: ∑ (x . The mean of the sampled data is the estimated distribution mean. Rome. even when the population sampled is not normally distributed (see Section 3.0. the mean.

4772).1.0 .∞ and zero (0.0.6826 0.002 inches.1915 0.0 3.0228 Thus.3830 0.0 (0. we can calculate it by subtracting the areas we don’t want from 1. 0. However.0 = 2 0. The solution to our problem is simply the area under a standard normal distribution from z = 2 to ∞ . Table 3-2: Standard Normal Distribution Data z 0.5000 Area Between -z and z 0. NY 13440-6916 • 1-888-RAC-USER .4772 0.5000 + 0.5000 from Table 3-2) and the area from zero to 2.5 1. (3-18) Reliability Analysis Center (RAC) • 201 Mill Street.5 2. Figure 3-4 shows the area between zero and z quantified in column two of Table 3-2.(0.9974 1.001 (3-17) Where z is the point on the standard normal distribution equivalent to 1.002 .0.0 1. and want to find the proportion of rods that are more than 1.002 inches.STAT Section 3: Some Distributions and Their Uses 21 An abbreviated tabulation of data (extracted from Appendix C) for a standard normal curve is given in Table 3-2.4987 0.8664 0.9544 0.4772) = 1 . the total area under the curve.000 0 z Figure 3-4: Standard Normal Distribution Suppose we estimate the mean length of a rod at one inch with a standard deviation of 0. Area wanted = 1. Rome.0228 or 2.3413 0.002 inches. Table 3-2 does not directly list this area.001 inch. From Equation 3-16: z = 1. The areas we don’t want are the area between .4332 0. following a normal distribution.0 ∞ Area Between 0 and z (or Between 0 and -z) 0.9772 = 0.28% of the rods we produce will be longer than 1.

2 Various Other Useful Distributions. fit well to a normal distribution whose mean is the mean of the samples and is equal to the mean of the parent distribution. Table 3-3 is another set of values for the standard normal distribution.2.80 0.005 3. rather than one which solves for areas given values of z. called the central limit theorem.28 1.1 The Lognormal The lognormal distribution is one in which the logarithms of measurements from a population are distributed normally. NY 13440-6916 • 1-888-RAC-USER . the means of the samples will follow a normal distribution with its mean the same value as the mean time to failure of the parent population of rods.33 2.).9544 or 95. but few opportunities to shorten the time significantly.58 Area Between z and -z 0.01 0.10 0.2.2. The lognormal is descriptive of such things as the time it takes to get to work or the time to repair a failure.98 0. Rome. The lognormal is handled by taking the logarithms of the measurements of interest. means the normal distribution can be applied to samples from almost any distribution to provide information on the parent distribution. like Table 3-2 or Appendix C.44% of the rods will meet the specified tolerance.99 Area From z to ∞ 0. This phenomena. Table 3-3: Critical Values of z Critical Value of z 1. the proportion of rods "in spec" would be given by the area under the standard normal from z = -2 to +2.2 The Normal Distribution’s Role in Sampling One of the reasons for the usefulness of the normal distribution is that means of samples from any distribution with a finite mean and variance. The distribution of the measurements themselves will be skewed to the right. and analyzing the resultant normal distribution as discussed in Section 3. Since there is a one-to- Reliability Analysis Center (RAC) • 201 Mill Street. troubleshooting difficulties. From the table. there are often circumstances which lengthen the time (accidents on the road. In Brief 3. We shall use this table in Section 7. but one that solves for z given areas of interest. This table defines "critical values" of (z) marking the ends of some specific areas of the standard normal distribution that are often used in determining confidence intervals in measuring and demonstrating values of parameters that conform to a normal distribution.2.645 1.05 0. 3. if the sample is large enough. The mean time to failure of the rods may be estimated from a sample.2.1. If many samples are taken. the standard deviation of the distribution of sample means is equal to the standard deviation of the parent distribution divided by the square root of the sample size ( σ/ n ).90 0. For example.95 0.025 0. 0. the times to failure of a rod under stress may follow a Weibull distribution (to be discussed later).1. We shall see this in action later. In these cases.002 inches. etc. In addition.96 2.22 Practical Statistical Tools for the Reliability Engineer STAT If we specified that the rods made should be one inch long plus or minus 0.

and is estimated from the sample using the unbiased estimator of Equation 3-15. 3. decreasing and constant failure rates.2).9957.9957 repair times) will be below y = e = e = 20 minutes. or probability of a unit failing by time = t) of: F(t) = 1 .2. one must assume a large sample size and that .F(t).STAT Section 3: Some Distributions and Their Uses 23 one transformation of the points on the lognormal to the points on the normal. 3.e.2. if a measured value from a lognormal distribution of repair times "y" = 20 minutes.1) finds that 90% of the area of the normal is below x = 2. Instead of the standard normal random variable: z = x/ n (3-21) Reliability Analysis Center (RAC) • 201 Mill Street.9957.. then 90% of the area on the lognormal (i.2 The Exponential This distribution describes time to failures of a repairable system. If analysis (which probably involves further transformation of the data to the standard normal and back. because it can accommodate increasing. Rome. When the sample size is small.2.2. For example. as discussed in Section 3. NY 13440-6916 • 1-888-RAC-USER .e -λt -λt (3-20) Since the probability of a unit not failing at time t = R(t) = 1 .2.2. then the corresponding value on a normal distribution "x" = ln 20 = 2. which includes maintainability demonstration tests based on lognormal distributions of repair time. Its probability density function is: f(t) = λe -λt (3-19) resulting in a cumulative density function (percent of units failed at time = t.2.4 The Student t Using the central limit theorem (see Section 3.3 The Weibull When a constant failure rate cannot be assumed. the normal distribution does not really apply. which is the widely used reliability expression encountered earlier in Equation 3-9. is known. we can translate the results back to the original data. when the failure rate is reasonably constant with time.2. A practical application is found in MIL-HDBK-470A.1. the standard deviation of the parent population. Designing and Developing Maintainable Products and Systems.1. We will become more familiar with this distribution in Section 4. Weibull analysis assumes no repair of failed units. R(t) = e . 90% of x 2. 3. the Weibull is often the distribution of choice.

We will interpolate this to z = 1. the value of "t" for 95% goes Reliability Analysis Center (RAC) • 201 Mill Street. However. when z = 1.∞ to "t". However.0. we will compute the values of "z" and "t" which cover 95% of the area under the curves starting from minus infinity (i. Hence.65. Hence.∞ and z. To compare the two distributions. as discussed in the text above the table in Appendix C..e.7. one for each value of "n".65 standard deviations from the mean.0. 95% of the curve is below +1. roughly representing the amount of information available in a sample and equal to the sample size minus one.45 = 0.45 when "z" is between 1. as sample size increases. the values differ with "degrees of freedom". 95% of the area under the curve is between .833.05. 90% of the curve is below +1. the value of "t" approaches the value of "z" for the same area under the curve.6 and 1. This shows that the Student t distribution is wider than the standard normal for relatively small samples. we note that the area from the mean (z = 0) to "z" is 0. Appendix E provides a tabulation of Student t distribution data. Since "z" is measured in standard deviations.833 standard deviations from the mean. As "n" increases.65. only 5% of the area is excluded).65 to ∞ = 1 . directly. NY 13440-6916 • 1-888-RAC-USER . Since "t" is also measured in standard deviations. values of "t" become close to the values of "z". From Appendix C.24 Practical Statistical Tools for the Reliability Engineer STAT where: x = = = = = a sample measurement mean of the sample measurements (and of the parent population) standard deviation of the parent population sample size standard deviation of the sample measurements (the distribution being converted to the standard normal) n / n we have: t = where: S S/ n = the estimated standard deviation of the parent population (per Equation 3-15) = the standard deviation of the sample measurements other terms as above xS/ n (3-22) There is a family of "t" distributions. as the reader can verify by comparing values derived from Appendix C (the Standard Normal distribution) to those given in Appendix E (the Student t distribution). the value of "t" marking the edge of 95% of the area under the curve is 1. the area in the tail of the curve from 1.5 . Rome. The tabulated Student t in Appendix E gives the values of "t" for defined areas from . For our example. For a sample of ten (nine degrees of freedom).

when samples are not replaced Describes many parameters. We will discuss this in Section 8. It is a family of distributions. dependent on the sample sizes. which agrees with our interpolated value of z = 1. for stated risks of error. These are the values of F which cannot be exceeded if we are to conclude there is no difference in the variance of our samples.645 as sample size increases. no repair Replaces standard normal for small samples Testing for significance of differences in the variances of two samples Estimating confidence intervals. It may be used to determine the confidence limits around a measured failure rate or to determine if a failure rate is stationary (i. and is used to test whether or not the samples are from the same population.2.2.e. does not change with time).0. NY 13440-6916 • 1-888-RAC-USER . such as repair times Describes distribution of failures when failure rate is constant Describes distribution of failures for constant or changing failure rate. by type and use in reliability engineering. It is also a family of distributions. including mean values of samples from any distribution with a finite mean and variance All normal distributions can be converted to the standard normal for ease of analysis Describes some parameters of interest.6 The Chi-Square Distribution The Chi-square distribution describes the relation of the true mean of a population to the mean of a sample. 3. When the samples are from the same population. 3.2. Rome. dependent on sample size.STAT Section 3: Some Distributions and Their Uses 25 to a limiting value of 1.2. we can accept or reject the hypothesis that the samples are from the same population within a specified risk of error. 3. We will use the Chi-square distribution in Section 4. By measuring the value of F derived from two samples and determining the probability that the value measured would occur if the samples were from the same population.5 The F Distribution The F distribution describes the ratio of the variances of two independent samples. and testing for a constant failure rate Reliability Analysis Center (RAC) • 201 Mill Street. the value of the F statistic will be distributed about 1. Table 3-4: Summary of Distributions Distribution Binomial Poisson Hypergeometric Normal Standard Normal Lognormal Exponential Weibull Student t F distribution Chi-square Type Discrete Discrete Discrete Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Main Uses in Reliability Engineering Finding probability of "x" events in "n" trials Finding probability of "x" events when "a" are expected Replaces binomial for samples from small populations.65.. Appendix F is a tabulation of "critical values" of the F distribution.3 In Summary Table 3-4 presents a brief summary of the distributions discussed in this section.

26 Practical Statistical Tools for the Reliability Engineer STAT This page intentionally left blank. Reliability Analysis Center (RAC) • 201 Mill Street. NY 13440-6916 • 1-888-RAC-USER . Rome.

mean time between failures)? The answers to these questions determine the statistical tools of interest. or as a frequency of failures (failure rate or its reciprocal. Rome. the integral is the cumulative density function. t1 t2 When t1 = 0. we need more information. defined as the percent of the population failed in the interval ending at t = t2. The curve may be defined by the parameter f(t) which describes its height against time. The integral ∫ f(t) = the percent of all failures occurring between t1 and t2.0 MEASURING RELIABILITY To select the statistical tools useful in measuring reliability. these are rarely sufficient. Reliability Analysis Center (RAC) • 201 Mill Street. It is also the probability that any given unit will fail in a mission of length t2. in this case a failure. Most often.. Note that the integral of f(t) from minus infinity to plus infinity (or from zero to plus infinity) equals unity.1 General Principles We can crudely measure reliability from test data by dividing the number of failures seen by the total hours of operation of our test sample (for a failure rate) or the number of products tested (for a probability of failure).g. we have to be more specific about what we are measuring. as an expected life (mean time to failure). showing the relative probability of a random variable occurring. The area under the curve is unity. f(x) F(x) = ∫0 f(x )dx x Probability Density Function 0 x Time To Failure Figure 4-1: Probability of Failure as Represented by the Area Under the Probability Density Function Figure 4-1 represents a probability density function. and can get it by finding (or assuming) a failure distribution and determining its parameters. decreasing or increasing? Is reliability better expressed as a probability (e. However. F(t). plotted against time. NY 13440-6916 • 1-888-RAC-USER . We shall first consider some general principles. Are we concerned with the reliability of products which are discarded on failure or of products repaired on failure? Do we know if the reliability is constant. of a successful missile launch).STAT Section 4: Measuring Reliability 27 4. 4. From f(t) we can obtain three useful measurements: 1.

it can be shown that as the interval becomes infinitely small it becomes: f(t) f(t) or R(t) 1 .2 The Versatile Weibull Distribution When we are interested in the reliability of a part. we are ready to discuss measuring reliability. an automobile drive chain). R(t). and the denominator represents the proportion surviving at the start of the interval and the length of the interval. force of mortality. The Weibull Probability Density Function is described by the formula: f(t) = where: t -1 e ..28 Practical Statistical Tools for the Reliability Engineer STAT 2. 3. we can often make use of a versatile statistical distribution invented in 1937 by Waloddi Weibull. Rome. or the first failure of a repairable assembly (e. the probability of a unit not failing in a given period of time (t). It is often used to describe the reliability of mechanical items that are subject to a wearout failure mechanism.t1 ) where the numerator represents the proportion of a population failed in the interval.(t/ ) (4-1) t = characteristic life (> 0) = shape parameter (> 0) = time The reliability function (surviving portion of the population at time = t) would be: R = e . This is also occasionally called the hazard rate.g. We shall first consider a useful tool when repair is not a consideration: the Weibull distribution. NY 13440-6916 • 1-888-RAC-USER . With these concepts understood.R(t1 ) R(t1 )(t 2 . Using the relationships between F(t). The failure rate in an interval from t1 to t2 is given by: R(t 2 ) . or ROCOF (Ascher’s acronym for the rate of occurrence of failures for a repairable system). increasing or decreasing failure rates. which can describe constant. The reliability function. 4.(t/ ) (4-2) Reliability Analysis Center (RAC) • 201 Mill Street.F(t).F(t) This defines an instantaneous failure rate at the instant "t". f(t) and R(t). or a non-repairable assembly. is simply 1 .

one needs to plot cumulative percent failures against time on Weibull analysis paper and either use the graphical Reliability Analysis Center (RAC) • 201 Mill Street. and finally by a Weibull with > 1 as wearout mechanisms act to increase failure rate with time. On this paper the X-axis is scaled as the natural logarithm of time. The slope of the plot. When = 1. such as infant mortality due to defects. Equations 4-1 and 4-2 apply to the two-parameter Weibull. infant mortality). or any other parameter appropriate to the failure mechanism of interest. Essentially.F1 ) ∆Y = ln t 2 . NY 13440-6916 • 1-888-RAC-USER . showing a constant failure rate whose reciprocal.STAT Section 4: Measuring Reliability 29 Instead of time. It is possible for the failure distribution of a product to be described by three different Weibull functions at different times in its life: first by a Weibull function with < 1 reflecting improved failure rates as initial quality defects are eliminated. can be calculated by: (4-3) = where: ln ln (1/1 . then by a Weibull function with = 1 reflecting a relatively constant failure rate during the product’s useful life. 63. There is also a three-parameter Weibull. the value of is that point on the time axis marking the point where the plot crosses the 63rd percentile on the Y-axis (actually.e. The Y-axis is scaled as: Y = ln ln (1/1 . Although it has been used to model mixed failure modes. miles. Rome.t0". wearout). This is used in the case where the failure mechanism cannot result in a failure until a certain time "t0" is reached. Test data following a Weibull distribution will plot as a straight line when the cumulative percent failed is plotted against time to failure on special graph paper. obtained by replacing "t" in Equations 4-1 and 4-2 with "t .ln t1 ∆X (4-4) F2 = the percentage failed at time = t2 F1 = the percent failed at t1 To determine the parameters of a Weibull distribution..0.F(t)) where: F(t) = the estimated percent of the population failed Weibull analysis paper is designed to permit the values of and to be found by graphical methods. equals .F2 ) . the Weibull formula reduces to the exponential formula.e. The two-parameter form of the Weibull is more common than the three-parameter form. An example might be cracks in a crystal which do not grow large enough to cause a failure before t0 is reached.ln ln (1/1 .2% cumulative failures). the MTBF. therefore.. and < 1 shows a decreasing failure rate (i. "t" could represent cycles. A value of > 1 indicates an increasing failure rate (i. the Weibull is really a model for a single failure mechanism. .

regardless of the operating time on the other units or when it went on test. with the following results shown in Table 4-1. Mechanical Applications in Reliability Engineering. One is to determine the "median rank" by Bernard’s formula: Median rank (cumulative percent failed) = where: I = rank order of a given failure (I = 1 for the shortest time to failure.7 5. The use of the graph is easier.4 (4-5) The first step would be to rank order the data as shown in Table 4-2. is considered a biased estimator. This.8 5 I . and provides some verification that the data is indeed Weibull.) N = sample size Also.30 Practical Statistical Tools for the Reliability Engineer STAT solutions or the calculations described above.0.3 4. Let us assume we have tested six items to failure and have measured the life to failure in terms of operating cycles. Rome.3 x 100 N + 0. the Weibull reliability function (Equation 4-2) could be used to solve for .2 9. it should be noted that the rank order is determined by the operating times to failure on the individual units. The most obvious way to estimate the cumulative percent failed is to divide the number of failures by the sample size. If the plot is not a straight line. etc. One could even forgo the graph paper by knowing the times corresponding to two percentiles. or any measure of calendar time. the time of the third failure would represent 30% cumulative failures. not the sum of times among the samples. however.) Accordingly. but. From this. there are some subtleties involved in determining the cumulative percent failed. Table 4-1: Life Data Part Number 1 2 3 4 5 6 Life (10 Cycles) 6. If we had ten samples and three failed. To illustrate these points. Reliability Analysis Center (RAC) • 201 Mill Street. one would expect the failure of one sample to better represent the 50th percentile. something is amiss. NY 13440-6916 • 1-888-RAC-USER . With and the known time for any cumulative percent failed. The unit that has accrued the lowest operating time when it fails is the first failure in the rank order. one could calculate . intuitively. (Consider a sample of one: its time of failure would be counted as the 100th percentile. However. we will use an example from RAC Publication NPS.6 1. there are various schemes to determine the cumulative percent failure represented by each failure in a sample.0 2.

45 6. Jeymer Drive. is about 1. etc.3 26.weibullnews. help establish appropriate warranty terms.7 42. Table 4-3: Completed Data Table Rank 1 2 3 4 5 6 Part Number 2 4 3 5 1 6 Median Rank (%) Life (10 Cycles) 10. There are also PC software packages for Weibull analysis. Reliability Analysis Center (RAC) • 201 Mill Street. H. England (Tel: 01-578-6861).8 x 10 . Box 25.55 2. In this case.Weibull.4 6.82 5. On this Weibull paper.2 percentile) to the plot.0 57.91 1. and then down to the corresponding number of cycles.3 2.5. Greenford. For part number 2. determine the expected number of spare parts needed for a set of products. here about 5 5.STAT Section 4: Measuring Reliability 31 Table 4-2: Ordered Data Rank 1 2 3 4 5 6 Part Number 2 4 3 5 1 6 Life (10 Cycles) 1.0 5. The characteristic life. These include SuperSMITH (TM) by Fulton Findings (on the world wide web at http://www.com) and Weibull++ (TM) by Reliasoft (on the web at http://www. NH 03886 (Tel: 603 323-8843).8 5 We now have all the information we need to plot the data on Weibull analysis paper as shown in Figure 4-2.com). calculate the probability of success for a mission. Tamworth.09 9.4 (4-6) After computing the median rank for all six parts.. to estimate the cumulative percent failure it represents. A Weibull reliability function can be used for many purposes. the first in rank: Median rank = 1 .91% 6 + 0.3 0. Two sources are: Team Graph Papers.2 73. a line parallel to the plot is drawn from the circle to the arc at the left edge at the 60% failure line. Rome. NY 13440-6916 • 1-888-RAC-USER . such as to determine an effective burn-in time. This is equal to in the Weibull reliability function. we have the data shown in Table 4-3. Weibull analysis paper is available commercially.W.7 4. and Chartwell Technical Papers. is found by moving horizontally from the circle (across the 63.18 4. The slope of the curve is read from the scale on the arc at the point of intersection.0.2 6. we need to determine the median rank of each failure by Bernard’s formula.6 9.7 x 100 = x 100 = 10.8 5 Next. . Middlesex. Peel & Co.6 89.

% 80.% 95.32 Practical Statistical Tools for the Reliability Engineer STAT 6 .% 40. 0 WEIBULL SLOPE 1.% 2.% 70.% 3.0% 4. Rome.% 20.% 90.6 .% 60.0% 1. for many reasons. Some data are better described by the lognormal.0% 2. First of all.0 1.5 10.8 0 . for example.0. .2.1 Caveats Weibull analysis is not always as easy as the example shown. The data may not plot as a straight line on Weibull paper.0 4. NY 13440-6916 • 1-888-RAC-USER .% 50.% 5.7 .4 1.% PERCENT FAILURE 30.0% 1 2 3 4 5 6 7 8 91 2 3 4 5 6 7 8 91 x 105 x 104 Figure 4-2: Weibull Plot 4. a distribution which is not convertible into a Weibull (it is handled by converting it to a Reliability Analysis Center (RAC) • 201 Mill Street. 2 99. the data may not be distributed according to the Weibull distribution.0 3.

Table 4-4: Failure Data Failure Number 1 2 3 4 5 6 7 Time of Failure in System A 15 42 74 117 168 233 410 Time of Failure in System B 177 242 293 336 368 395 410 Time of Failure in System C 51 94 121 298 313 378 410 The data represents seven failures. by Harold Ascher and Harry Feingold (Marcel Dekker Inc.2. Robert B. NY 13440-6916 • 1-888-RAC-USER . There may be suspensions. 1984). which in system A arrive at intervals increasing with time. because the expected number of failures in a given interval depends on the system age at the start of the interval.3. unchanging with time).. This is referred to as a stationary stochastic process ("stochastic" meaning involving a random variable.. There may be more than one significant failure mode. Rome. 4. We will discuss this further later. It is also called a homogeneous Poisson process (HPP) because the number of failures in an interval follows a Poisson distribution that is "homogeneous" or stationary (i.e.1 Testing for Trends To illustrate the tool we will be discussing. we shall use some contrived data. This assumption is reasonable in that a system is a collection of parts with differing failure mechanisms. The data compares three systems. by Dr. and in system C arrive (pseudo) Reliability Analysis Center (RAC) • 201 Mill Street. Test data may not include all time on the parts (i. in aggregate. 4. but it seems reasonable to start this section with some statistical tools for determining what kind of process we have. in system B arrive in intervals decreasing with time. but rather the times between failures of a system which is repaired after each failure. which. as discussed in Section 3. there are also nonhomogeneous Poisson processes (NHPPs) which describe systems where the number of failures in an interval may follow a Poisson distribution. each having seven failures as shown in Table 4-4. which are parts on test which are removed from test before failing or which fail from mechanisms other than the primary one. but the distribution will change with time.e. There are practical cases where reliability improves with time and also practical cases where reliability degrades with time.1).2. which has many advantages in analysis. Ways to handle these situations are discussed in the New Weibull Handbook. This authoritative reference is available from the Reliability Analysis Center under RAC order code WHDK. and "stationary" meaning that the expected number of failures in a given interval will be the same regardless of the age of the system). time zero is some unknown time before the test starts).. Abernethy. For such a system.3 Measuring Reliability of Repairable Systems Our failure data may not represent the times to failure of a number of parts. can appear to fail at a constant rate. it is often assumed that the failures occur at a reasonably constant rate. However.STAT Section 4: Measuring Reliability 33 normal. borrowed from the text Repairable Systems Reliability.

When our data is failure truncated (i. i. As discussed in Section 2. Rome.. the statistic is calculated by the formula: n -1 ∑ (t i ) /(n . will be zero for perfectly random data. It is obvious that system A has an increasing time between failures. and "tn" is the last failure. But real data is not so obligingly obvious. "Satisfactorily small" can be defined quantitatively as an acceptable risk. the mean of the standard normal distribution. the records end at the occurrence of the last failure).T/2 i =1 U = T 1/[12(n)] where: T = the time at the end of the data (4-8) The statistic approximates the distance (in standard deviations) that the data differs from the mean of a standard normal distribution.t n /2 i =1 U = t n 1/[12(n .. This probability can be directly translated into the distance from the mean of the standard normal distribution. or a decreasing rate of occurrence of failure (ROCOF).e.e. a probability of making a wrong decision that is small enough to satisfy the decision maker.1) . for example. For time truncated data (i. we can reject the hypothesis that there is no trend when the statistic calculates a value whose probability is satisfactorily small. we find useful a statistical measure invented by Laplace.4. Statistical variation results in it taking other values.e. A system with no changes in ROCOF would have an expected value of zero. discussed in Section 2. In system A.34 Practical Statistical Tools for the Reliability Engineer STAT randomly.. and the last occurs 15 hours after the penultimate. This area is the risk of error (in this case the probability that a system with no trend would produce a data set falling in the tail of the curve cut off by the value Reliability Analysis Center (RAC) • 201 Mill Street.1) . the first failure occurs 15 hours after turn-on. the first failure occurs 177 hours after turn-on. NY 13440-6916 • 1-888-RAC-USER . and that system B has an increasing ROCOF. we simply determine the area of the distribution which falls outside a given distance from the mean. positive for increasing intervals between failures (ROCOF decreasing with time) and negative for decreasing intervals (increasing ROCOF). "ti" is the time of the "i " failure in order of occurrence. Applying the principles of hypothesis testing. The result.1)] th (4-7) where "n" is the number of failures. while in system B. with probability decreasing proportionally with the distance from zero. "U". the records end at a given time while the system is in a nonfailed state): n ∑ (t i ) /(n . In each case the seven intervals between failures are identical except for the order of occurrence. as it is in the data given in Table 4-4. and the last occurs 177 hours after the second to last. Hence.

2 1.025 risk). Applying Equation 4-7 to the data in Table 4-4 for system A: 15 + 42 + 74 + 117 + 168 + 233 410 6 2 = .0.0 20.960 1. there is only 0. and a value less than -3. Rome. we can assume a trend for increasing ROCOF (reliability decreasing) with only 0. which indicates a deteriorating reliability to a 97.645 1.09. Table 4-5 relates to Appendix C in that "U" corresponds to "Z" and the probability of error corresponds to the tails of the standard normal distribution (the area under the curve outside of the range -Z to +Z).0 For example.282 Probability of Error (%) 0.2.0 U = 410 1 /(12 x 6) (4-9) The negative result indicates a decreasing ROCOF (improving reliability). Some common critical values are shown in Table 4-5. NY 13440-6916 • 1-888-RAC-USER . we can assume there is a trend with only a 0.1% error.STAT Section 4: Measuring Reliability 35 calculated for the distance from the mean).0 10. and indicates we can accept that reliability is improving with only a 2. except for sign.09.0 U = 410 1 /(12 x 6) (4-10) Which is the same result as Equation 4-9.002 probability of error.5% confidence (1 .2% of the standard normal distribution outside the limits of ±3.326 1. The value of U exceeds the critical value given in Table 4-5 for 5% risk (meaning that there is less than a 5% probability that there is really no trend).0 5.09 permits assuming an improving trend with only a 0.576 2. We can define a "critical value" for the Laplace statistic which represents the value the statistic must exceed for a risk of error satisfactory to us.1% risk. Further.0 2. Doing the analysis for the data of system B: 177 + 242 + 293 + 336 + 368 + 395 410 6 2 = + 2. so if the absolute value of "U" exceeds 3. The refuted hypothesis would be that there is no trend in the direction of our conclusion. Reliability Analysis Center (RAC) • 201 Mill Street. Table 4-5: Critical Values for the Laplace Statistic Critical Value of U (Absolute Value) 3.09.09 2.5% risk. because only one tail would be outside the limit. if the value exceeds +3.

.034 = 0. the area from 0 to z would be (0. as explained in the text of Appendix C.932. Since the area in the upper tail is one minus 0.466 = 0. the greater the confidence provided by a given set of data. A constant ROCOF is indicated. The area in both tails would be twice this or 2 x 0. For example. extending from a confidence limit of 100 hours to infinity). Hence.0398. NY 13440-6916 • 1-888-RAC-USER .0.. we have more confidence in conclusions reached from extensive data than in conclusions reached from skimpy data. the times to failure are distributed in accordance with an Reliability Analysis Center (RAC) • 201 Mill Street.932 probability of error. the area from 0 to z = 0.36 Practical Statistical Tools for the Reliability Engineer STAT Repeating the analysis using the data for system C: 51 + 94 + 121 + 298 + 313 + 378 410 6 2 = + 0. We can come up with an estimate of the risk of error by using Appendix C. we have a 10% risk and a 90% confidence.5 .086.g.282 for a 20% risk. it is equal to 1 . the mean time between failures (MTBF) of a product can be simply determined from the total operating time of all the samples (i.0.e. the sum of the operating times accrued by each unit tested) divided by the total number of failures.2 Confidence Limits when the Failure Rate is Constant If we assume a constant failure rate. For example. The confidence interval can be two-sided (e. between confidence limits of 100 and 200 hours). we need to relate the confidence interval to a portion of a probability distribution curve. for z = 0. for any given set of data. To determine the width of the confidence interval for any given value of confidence. Statistical methods quantify confidence by defining it as the probability of being correct when we state the MTBF of a product is within a given range of values. we have a greater probability of being correct (i. If we have a 10% probability of error.034. Because we have assumed the constant failure rate.1. as shown in Table 4-5). confidence) that a product’s true MTBF has a one-sided confidence limit of 100 hours than we would have in a confidence limit of 200 hours. we would have to be willing to accept a 0..g.000 hours of operating time with 10 failures. Put another way. called the confidence interval.e.466. Rome. The values marking the limits of the interval are called confidence limits. the wider the confidence interval.05 minus the area from 0 to z.3. As might be expected. 4. Risk is defined as the probability that the true MTBF of the product is outside the confidence interval. or one-sided (e. the lowest figure listed. In the Appendix C..086 U = 410 1 /(12 x 6) (4-11) Which does not permit the rejection of the hypothesis that there is no trend even at a 20% risk of error (U = 1. 100 hours of operation during which one failure occurred would give us an MTBF of 100 hours. to reject the hypothesis that there is no trend.0398 = 0. So would 1.086/0.1) x 0. Using a linear extrapolation. for z = 0. Which data set would you prefer as a measure of your product? It should be obvious that the risk in making conclusions based on one failure is greater than the risk in making conclusions from many failures. and is equal to one minus the confidence.

which is a function of the number of failures. To use Table 4-6.599 40. However. we select the chi-square values for the appropriate degrees of freedom based on the number of failures accrued and whether the test was time-truncated or failuretruncated (to be explained later).940 10. called the degrees of freedom (roughly.773 In Table 4-6 and Appendix D. In actuality.236 4.070 18.103 0.865 15. but a dependent distribution describing the relation between the measured and the true MTBF of the product.605 0. This lack of consistency is intended to help the reader understand and deal with statistical tables in many different forms. From the selected values. each member determined by a function of the number of failures recorded. whether for a one-sided limit or for a two-sided interval. we can calculate the confidence.307 31.991 7. Table 4-6 is an abbreviated chi-square table.448 11.493 Chi-Square Value at α 90% 10% 0. and the probability of no failure in any given time is calculated by the familiar reliability function: R = e -(t/MTBF) (4-12) The reliability function is the portion of the failure probability distribution function extending from t to infinity. Table 4-6: Chi-Square Values Degrees of Freedom 2 3 4 5 10 20 22 30 95% 0. α = the selected risk (one minus the specified confidence). when a more consistent format could have been used. A time-truncated test is one which ends with a certain amount of time accrued.851 12. The reader may have noted that we have used different ways of presenting statistical tables.064 7. Rome.924 43. or the area in the tail of the curve from the listed values to infinity.041 30.584 6.711 1. either by reaching a pre-designated limit or by simply ceasing to test.352 0. and represents the portion of a population still operating after time t or the probability that a given unit will not fail between time zero and t. the total test time and the specified risk. since there is no standard format used by all references. the distribution we need for determining confidence intervals is not the distribution of failures itself.779 1. the exact function dependent on circumstances.145 3. NY 13440-6916 • 1-888-RAC-USER . the chi-square is a family of distributions.251 1. Reliability Analysis Center (RAC) • 201 Mill Street. The most common scenario is that of a group of equipment on test for a period determined by the willingness of management to tie up the test units.987 12.256 5% 5. When the exponential distribution of failures holds.813 20.STAT Section 4: Measuring Reliability 37 exponential distribution.412 14.338 18.443 28. the relation between the measured and the true MTBF of the product is described by a distribution called the chi-square.410 33.815 9. degrees of freedom represents the amount of information at hand. a more extensive table is provided in Appendix D.211 4.610 9. as will be explained).

we would have 2(10) + 2 = 22 degrees of freedom. we had 1. If our test data were failure-truncated.813 30. until a certain number of failures occur.000 hours of test time and 10 failures. instead. the risk (α) would be (1 . If. and our 90% confidence limit is found by: MTBF = 2(1. we have 2(1) + 2 = 4 degrees of freedom.desired confidence) 2T (Chi .90) or 0.0. we would be 90% confident that the "true" MTBF is actually 25 hours or more.38 Practical Statistical Tools for the Reliability Engineer STAT A failure-truncated test is one which ends at a given number of failures. The value listed in Table 4-6 for ten percent risk and four degrees of freedom is 7. rather than time-truncated. NY 13440-6916 • 1-888-RAC-USER . The one-sided confidence limit for a set of test data from a time-truncated test is calculated from the formula: MTBF = where: T = total test time n = total failures accrued α = acceptable risk (1 .square value for 2n + 2 degrees of freedom at percentile α) (4-13) (4-14) Hence.779 2T (Chi .000 = = 65 hours 30. Another way is the test of one unit.square value for 2n degrees of freedom at percentile α) (4-16) Reliability Analysis Center (RAC) • 201 Mill Street.000) 2. Plugging this value into the formula we have: MTBF = 2(100) 200 = = 25 hours 7. Rome. if we test for 100 hours and have one failure.813 (4-15) Hence.779 7.desired confidence) For example. repaired after each failure. we would use the following formula to determine the one-sided confidence limit: MTBF = where: T = total test time N = total failures accrued α = acceptable risk (1 .10. For a 90% confidence. the most common manifestation being the test of a number of units until all fail.779. we would be 90% confident that the "true" MTBF is 65 hours or more.

Note that the degrees of freedom remain the same for both time-truncated and failuretruncated tests. we want to lop off 5% of the distribution at each end.940 hours . we would cut off portions of a density function at both ends.605 (4-17) as opposed to 25 hours for the time-truncated test. resulting in a slightly higher MTBF.605 4. for either time-truncated or failure-truncated tests. NY 13440-6916 • 1-888-RAC-USER .STAT Section 4: Measuring Reliability 39 This formula is identical to the previous one.square value for 2n degrees of freedom at percentile = 1 . except that the degrees of freedom are reduced. is determined by the formula: MTBF = 2T (Chi . except that the chi-square value for α/2 is used. roughly. The degrees of freedom is less because we have less information.square value for 2 degrees of freedom at the 95 th percentile 200 = 1. Hence. In our examples. 4. For the one failure in 100 hours scenario: MTBF = 2(100) 200 = = 46 hours at 90% confidence . for a time-truncated test of 100 hours with one failure. The lower end is determined by the same formulas as the one-sided limits. (If we want a 90% confidence in a two-sided confidence interval. because the additional information of time after the last failure does not significantly affect the estimation of the upper confidence limit as it does for the lower confidence limit. Rome. To calculate a two-sided confidence interval.0.103 = (4-20) Reliability Analysis Center (RAC) • 201 Mill Street. rather than 10% of the distribution at one end. there is no operating time after the last failure as there was in the time-truncated case.square value for 4 degrees of freedom at the 5 th percentile 200 = 21 hours 9.α / 2) (4-18) For example. These formulas essentially cut off a portion of a probability density function equal to the desired risk.10/2). we have found the value of MTBF below which there is only 10% probability that the true MTBF would fall.) The upper limit.448 = (4-19) Upper limit = 2(100) Chi . This is. the upper 90% confidence limit would be determined by finding the chisquare value for 2n failures at the 95th percentile (1 . the two-sided confidence limits are: Lower limit = 2(100) Chi .

For example. the more data available. Table 4-7 summarizes the formulas discussed above. MIL-HDBK-189. Even if recoverable and re-usable.1. the analyst must use the appropriate distribution. 2n) Lower Upper . 2n) X (1 . Table 4-7: Confidence Interval Formulas One-sided Confidence interval (MTBF lower limit) Two-sided Confidence intervals For time truncated tests 2T 2 X (α. Naturally. or even 4-2) to determine the probability of these items operating satisfactorily for a particular mission.α/2. This is an exercise in measuring quality from samples and is described in detail in Section 7. or it will fail with no possibility of repair during the mission. Fortunately. provides tables for a distribution (origin not identified) used to determine the confidence intervals of a failure rate that changes according to a nonhomogeneous Poisson process. and a quantitative measure of describing confidence would be quite useful. the better our confidence in the measured probability of success. 4. This type of product is called a "one-shot" device. However. Reliability Analysis Center (RAC) • 201 Mill Street. See Section 6 for more on reliability growth and MIL-HDBK-189. the 90% confidence interval for the data is from 21 to 1. 2n) 2T 2 2 X (α/2. 2T 2T 2 2 X (α/2. and this must be found directly by dividing the number of successful uses by the total number of attempts to use the product. We are 90% confident (have a 10% risk of error) that the true MTBF is between these limits. 2n) Lower Upper . the reliability of the shuttle’s booster rockets would not be appropriately measured in MTBF or any other measure of "life". NY 13440-6916 • 1-888-RAC-USER . 2n + 2) For failure truncated tests 2T 2 X (α.α/2. Either the booster will work properly on demand. we can determine confidence limits on the reliability of one-shot devices by making use of the fact that the successes and failures can be described by the binomial distribution. the rockets are subject to only one short demand per mission.940 hours. Rome. if he can identify or generate one. To calculate other parameters or under other assumptions.40 Practical Statistical Tools for the Reliability Engineer STAT Hence.4 Measuring Reliability of "One-Shot" Products The reliability of the electronics modules in the space shuttle could be measured in terms of mean-time-between-failures (MTBF) and this result used (in Equation 4-12. Reliability Growth Management. Its proper figure of merit is the probability of success. 2n + 2) X (1 . 2T MTBF limit: These formulas apply only to the estimation of an MTBF based on a constant failure rate.

. the zero failure test.(t/ ) where: R = probability of failure free operation t = operating time = characteristic life ( > 0) = shape parameter ( > 0) When several products are tested together. "How sure can I be that the reliability is satisfactory?" Since these questions are different. we would select a characteristic life which represents the minimum acceptable value. 5.1 Zero Failure Tests A zero failure test requires a given number of samples to be tested for a specified time. These are derived from a branch of statistics called hypothesis testing. reliability demonstration seeks to answer. let us consider the simplest reliability demonstration test.0 DEMONSTRATING RELIABILITY Measuring reliability seeks to answer the question. the methods used to answer them are also different. we will consider here the testing of life measures (i.n(t/ ) where: n = number of products tested other terms as above (5-2) (5-1) To demonstrate reliability..e.g. (or perhaps better.4. Leaving the testing of "one-shot" devices to Section 7. For example. the reliability meets a specified number) within acceptable risks. characteristic life and mean time between failures).2. To begin. Rome. R = e . NY 13440-6916 • 1-888-RAC-USER . and now will examine some ways to demonstrate reliability.STAT Section 5: Demonstrating Reliability 41 5. the highest value that we would call "bad") and determine the risk we are willing to take of accepting a product with that characteristic life. We have covered the measurement of reliability. as described in Section 2. the reliability function of a product following a Weibull distribution of failures is: R = e . If no failures occur the product is accepted as meeting reliability requirements. "What is the true reliability of the product?" In contrast. The determination of sample size and test length is accomplished through consideration of the product’s reliability function. a set of methods designed to accept or reject hypotheses (e. That Reliability Analysis Center (RAC) • 201 Mill Street.

e. assuming we have an estimate for and a known sample size: Risk = e .42 Practical Statistical Tools for the Reliability Engineer STAT risk is the probability of the "bad" product having no failures during the test.3 = . We shall also discuss this in Section 5. Rome. For example.10) = .n(t/ ) The equation is solved for "t".15( ). When the exponential distribution of failures is assumed (i. the product is accepted. It should be noted that the only risk considered in our discussion so far has been the "consumer’s risk" (i. are easier to work with than others.e.100 = (t/ ) 2 = 0. know that = 2. all samples are tested for the same operating time.023 (t/ ) = 0.023 = 0. in Section 5. if we test "n" samples for time "t" and have no failures. we will first discuss the derivation of tests in which some failures are allowed. Under the Weibull assumption.10 = e -100(t/ ) (5-3) (5-4) Which gives: ln(0. There was no consideration for the probability of a "good" product failing the test (the "producer’s risk"). Then. If no failures occur.15 t = 0. The zero failure test procedure will work with any distribution for which a reliability function can be defined and all parameters identified. The exponential distribution is the easiest of all.. we can accept the product with the predetermined risk that we may have accepted products with the defined "bad" characteristic life.2. and can test 100 units: 2 0.. though some distributions. Thus. NY 13440-6916 • 1-888-RAC-USER .15( ) (5-5) Thus the samples are tested for a time (t) equal to 0. and products with lower lives will have a lesser probability of passing.100(t/ ) 2 . the failure rate is assumed constant). if we are willing to take a 10% risk of accepting products with a defined "bad" .2. equal test times are not necessary.100(t/ ) 2 . Reliability Analysis Center (RAC) • 201 Mill Street. However.3. the only unknown. Products with higher characteristic lives will have a higher probability of passing the test. as we will discuss in Section 5.2. the probability of a "bad" product passing the test).2.3/ . such as the Weibull.

Rome. It warrants consideration because a test based solely on satisfactory consumer’s risks may unintentionally provide a high risk of rejecting "good" products. yields the probability of getting "x or less" events in "n" trials. A "good" product may be one with an MTBF clearly acceptable for the mission. it does the consumer no good to reject satisfactory products. then the probability of a failure.x)! (5-6) where: p q n = probability of an event happening in one trial = probability of event not happening in one trial = number of trials If we let an event be a failure.1. P(x or less) = ∑ 0 x n! p x q (n . It becomes even more difficult when the producer’s risk is considered. Letting "n" equal the number of units on test.2.2.1 Controlling the Producer’s Risks The producer’s risk is the probability of a test rejecting "good" products. P(x or less) = ∑ (x) n! 1 .1. and the probability of no failures. Reliability Analysis Center (RAC) • 201 Mill Street. It should always be possible to achieve the "good" MTBF with reasonable effort. Equation 5-7 gives the probability of "x" or less failures when all units are operated for "t" time units unless prevented by failure. or equal to the state-of-the-art for the product.1. Equation 3-4. repeated here as Equation 5-6.2 Tests Allowing Failures If we test a number of units for the same test time on each unit. or just something that is an arbitrary ratio higher than the "bad" MTBF that we want to reject. "q". This is obviously not a closed-form solution. and would ordinarily be done by trial and error iteration using a computer.x) x! (n . Though this is called the producer’s risk (for obvious reasons). NY 13440-6916 • 1-888-RAC-USER .e .e − (t/ ) ) . "p". is equal (by definition) to the reliability function (e − (t/ ) ) . as discussed in Section 5.STAT Section 5: Demonstrating Reliability 43 5. and do not allow a failed unit to be repaired and returned to the test. we can derive test plans allowing some failures by using the binomial distribution. As discussed in Section 3. except for failed units. 5.(t/ ) β n−x (5-7) We then set P(x or less) equal to the risk we are willing to take of accepting products with an MTBF equal to . is equal to one minus the reliability function (1 .x)! x e . and solve for "t" given any desired values of "x".(t/ ) β 0 x! (n .

Calculate the probability that a value of life considered good would be rejected (producer’s risk) using the zero-failure test time.x)! n (5-8) To formulate a test with satisfactory producer’s and consumer’s risks. allow another failure.3 Testing Under the Exponential Distribution When the exponential distribution of failures applies (i.2 do not hold. or more) = ∑ x + 1 x! (n . when " " is equal to the "good" MTBF. P(n) = (u) n e -(u) n! (5-10) Reliability Analysis Center (RAC) • 201 Mill Street. Start End If the producer’s risk is still too high. The producer’s risk is calculated from Equation 5-8. (x) n−x n! 1 .1. the assumption of a constant failure rate permits us to derive tests allowing failures by using the Poisson formula. If the producer’s risk is too high. the failure rate is constant). Since the assumptions we used in Section 5.(t/ ) e . Calculate time needed to satisfy the consumer’s risk in a zero-failure test. it does not matter whether or not the test units have equal operating times. one defines the values of both risks. failures can be repaired and the failed unit returned to test.e . However. the binomial distribution cannot be used to derive tests allowing one or more failures. the reliability function is: R = e -(t/ ) (5-9) where: R = probability of zero failures in time (t) = the mean time between failures Since the failure rate does not change with time. In addition. One way is the procedure shown in Figure 5-1.(t/ ) P(x + 1. assume a one-failure test and re-compute the test time for a satisfactory consumer’s risk. recompute test time and recalculate probability until an acceptable producer’s risk is achieved.44 Practical Statistical Tools for the Reliability Engineer STAT Equation 5-7 yields the consumer’s risk when " " is equal to the "bad" MTBF.. Calculate the probability that a value of life considered good would be rejected (producer’s risk) using the one-failure test time. Equation 5-6 can then be used to determine a zero failure test as shown in Section 5. NY 13440-6916 • 1-888-RAC-USER . unlike the tests based on the Weibull distribution.e. Rome. The test time "t" can therefore be set to the sum of the test time among all the units. and solves for values of "n" and "t" that satisfy both risk equations. Figure 5-1: Devising a Reliability Test 5.

the producer’s risk for a test time of "t" with "n" failures allowed. Table 5-1 tabulates test plans for the most reasonable combinations of risk and the ratio of the "good" MTBF to the "bad" MTBF (called the "discrimination ratio"). However. there is no need to go through this routine. Rome. when = the "bad" MTBF) is: P(n) = ∑ n (t/ ) n e -(t/ ) n! 0 (5-12) Thus. is: P(n) = 1 .STAT Section 5: Demonstrating Reliability 45 where: P(n) = the probability of exactly n events u = the expected number of events Since the number of failures expected in time (t) = t/ .e. we can establish a set of tests with different numbers of allowable failures. (Note: in some references 0 designates the "bad" MTBF and 1 is the "good" MTBF. the probability of passing the test (i. 0 is the "good" MTBF and 1. the "bad" MTBF. all of which provide the same consumer’s risk.∑ where: n (t/ ) n e -(t/ ) n! 0 (5-13) = "good" MTBF other terms as defined previously We can use the procedure shown in Figure 5-1 to derive tests with satisfactory producer’s and consumer’s risks. NY 13440-6916 • 1-888-RAC-USER . We shall need this capability in order to formulate tests that have satisfactory producer’s risks as well as satisfactory consumer’s risks. In the table.) Reliability Analysis Center (RAC) • 201 Mill Street. Based on the Poisson formula. the probability of exactly n failures in a test with time equal to t is: (t/ ) n e -(t/ ) P(n) = n! (5-11) If we establish a test where the sum of test time = t and up to n failures are allowed.. the consumers risk.

a reject decision can be made. If the former is satisfactorily higher than the latter. 5. a probability ratio sequential test may be used.0 29. and (2) the probability of occurrence when the test units have the "good" MTBF.5 2. Reliability Analysis Center (RAC) • 201 Mill Street.0 3. A summary of sequential test plans for the most common risks and discrimination ratios is presented in Table 5-2. an accept decision can be made.0 3. the test continues to an arbitrary truncation point (used to assure that the test ends in a reasonable time). Details on other sequential tests are available in MIL-HDBK-781.0 2. Conversely.0 1.0 3.1 Sequential Tests: A Short Cut As Table 5-1 shows.7 1. small values of risk and small discrimination ratios can result in long tests.3 8.0 3. Figure 5-2 shows the form of the test.0. which will permit more rapid decisions than the fixed time test.1 Accept-Reject Criteria Reject (Failures Accept (Failures or More) or Less) 37 36 26 25 18 17 14 13 10 9 6 5 6 5 4 3 3 2 7 6 3 2 1 0 Table 5-1 is taken from MIL-HDBK-781.4 4. Plans. Reliability Test Methods.0 Test Duration (x θ1) 45. which is the most comprehensive reference on reliability testing. NY 13440-6916 • 1-888-RAC-USER .5 1.5 1.8 9.3.46 Practical Statistical Tools for the Reliability Engineer STAT Table 5-1: Fixed Time Reliability Tests Nominal decision Risks Producer’s Consumer’s Risk (%) Risk (%) 10 10 10 20 20 20 10 10 10 20 20 20 10 10 10 20 20 20 30 30 30 30 30 30 Discrimination Ratio (θ0/θ1) 1.8 12..e.5 18. decision times for each failure) will be needed for application.0 2.4 7. when a combination of failures and test time is a predetermined times more likely to have occurred from a test of "good" units than from a test of "bad" units. More details (i. Qualification and Production. when the true MTBF of the test units is much closer to one of the defined "good" or "bad" values than it is to the other. as provided in Table 5-3 for a test with both risks approximately 10% and a discrimination ratio of 2.0 3.5 2.3 5. Rome. Where the ratio of the probabilities is not great enough to make a decision. and Environments for Engineering Development. Where this is unsatisfactory.9 21. The sequential test is based on the ratio of two probabilities: (1) the probability that a combination of failures and test time will occur when the test units actually have the "bad" MTBF.

40 11.40 5.60 N/A Reliability Analysis Center (RAC) • 201 Mill Street.6 9.79 13.70 2.0 3. Expected time for a true MTBF equal to 2.8 6.18 8.67 3.86 6.63 9.79 7.40 2.5 21.9 20.02 10.5 1.10 15.94 17.5 2.72 14.0 0 Time to Accept Decision (Multiples of θ1) Minimum Expected1 Maximum2 6.5 6.0 2.49 16.0 3.94 11.4 10.60 Accept if t ≥ θ1 Times: 4.26 19.60 20.6 4.2 4.72 25.18 14.0 3.15 1.34 20.48 4. Rome.95 11.80 3.08 3.STAT Section 5: Demonstrating Reliability 47 9 8 Reject Cumulative Failures 7 6 5 4 3 2 1 Test Data End Test Accept Continue Test 100 200 300 400 500 Cumulative Test Time.42 5.0 Discrimination Ratio Number of Failures 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Reject if t ≤ θ1 Times: N/A N/A N/A 0.34 12.56 15.1 2.74 10.35 4.65 20.5 2. 2.0 1.75 2. NY 13440-6916 • 1-888-RAC-USER .8 4.60 20.56 9.5 1.88 18.60 20.6 49. 600 700 800 Figure 5-2: Typical Sequential Test Table 5-2: Sequential Tests Nominal Decision Risks Producer’s Consumer’s Risk (%) Risk (%) 10 10 20 20 10 10 20 20 10 10 20 20 30 30 30 30 Notes: Discrimination Ratio (θ0/θ1) 1.19 4. Hrs. Arbitrary truncation point Table 5-3: Sequential Test Plan for 10% Risks.24 7.

we could not make a reject decision until the fourth failure.060 hours (accept). any testing) than the statistics involved. presuming it did not fail a second time.48 Practical Statistical Tools for the Reliability Engineer STAT As an illustration. MIL-HDBK-781. if we had accrued 70 hours (0. and then only if it occurred before we had more than 208 hours. 5. When the failures and times do not permit a decision to accept or to reject. If we had one failure before 440 hours were accumulated. and Production covers them in great detail. shock. we would accept the product. the test continues. we would have to wait until 579 hours to accept the product. Plans and Environments for Engineering Development.4 x 1) among all the units on test. If we had more time than 70 hours accumulated at the third failure. Rome. thermal and power cycling. for that matter. assume we put a number of products on test using Table 5-3 for our accept-reject criteria. Reliability Analysis Center (RAC) • 201 Mill Street. Qualification. NY 13440-6916 • 1-888-RAC-USER . and other test conditions a realistic representation of the environment the product will face in its expected use? Product Configuration and Operation: Is the product on test truly representative of the product of interest? Is it being exercised to the same degree? Monitoring: How often will it be checked for proper functioning. Some other considerations are: • • Definition of Failure: What is a failure? Is a transient event a problem? How much degradation is acceptable? Test Environment: Are the temperature. prevents the test from continuing indefinitely. However. and how? Failure Analysis: recommended? Will failures be analyzed for root cause and corrective action • • • • ESS and Preventive Maintenance: Do the products on test have the same benefits of environmental stress screening (ESS) and preventive maintenance that the production units will have (and no better)? Special Ground Rules: If the number of failures is acceptable. whichever comes first. vibration.70 x 1) or less among the units on test. The first opportunity to reject the product would be when three failures occurred. and 1 = 100 hours. An arbitrary truncation at 16 failures (reject) or 2. The first opportunity to accept the product would happen when we accrue 440 hours (4. If there was no failures at that time. must the cause for the pattern be found and rectified before the product is accepted? • These considerations are beyond the scope of this text. Reliability Test Methods. but a pattern of failures exists.4 Other Test Considerations There is more to reliability testing (or.

the AMSAA/Crow model. placing the straight line becomes rather arbitrary. of course. Duane’s equations for reliability growth are: cum = KT - (6-1) where: cum K T α = = = = cumulative failure rate initial failure rate test time growth rate By finding the rate of change in the number of failures against time. extending the instantaneous failure rate curve until it intersects with the desired failure rate. it is of interest to estimate when the reliability will have grown to a satisfactory value. However. because its solution uses least square regression. who noted that failure rate data plotted as a straight line against operating time on log-log scales. the Duane model. Duane. 6. a method of broad utility. Both of these are described. The other. Reliability Growth Management. given any spread in the data.STAT Section 6: Reliability Growth Testing 49 6. Predicting how much operating time would be required to achieve a desired failure rate was therefore a matter of fitting an equation to the data and solving the equation. because it assumes an underlying distribution to the data. and inst plot as parallel straight lines on log-log graph paper. Various models have been proposed. and reading the corresponding point on the time axis.0 RELIABILITY GROWTH TESTING Reliability growth is the improvement in reliability of a product as design defects are found and eliminated. and the two most popular are based on fitting the data to a straight line on log-log scales. Both methods will be explained. NY 13440-6916 • 1-888-RAC-USER . In either case. with others. This can be done using data from all operational tests of the product and/or by dedicated reliability growth testing. One can. The first. finding time for a given failure rate. Rome. inst = K(1 . in MIL-HDBK-189.1 Duane Growth Analysis The first formal reliability growth model was created by James T. obtain a solution by drawing a straight line through the data points.α) T -α (6-2) where inst = instantaneous failure rate at time T (the failure rate expected if reliability growth stops at time = T). cum Reliability Analysis Center (RAC) • 201 Mill Street. The parameters of the equations are therefore determined by a statistical analysis method called least square regression.

it minimizes the sum of the squares of the distances between the data points and the line. we will perform the least squares regression on the set of data shown in Table 6-1. but showing scatter. Rome.1. Basically.mΣ X i n (6-4) and b = (6-5) Relating this general procedure to Duane’s equations: Y b m X = = = = log of cum log K -α log of cum time As an example.(ΣX i )2 /n ΣYi .(ΣX i ΣYi )/n 2 ΣX i .50 Practical Statistical Tools for the Reliability Engineer STAT 6. The basic equation for any straight line plot is: Y = mX + b (6-3) To determine the equation from a set of paired data points using the method of least squares: m = ΣX i Yi .1 Least Square Regression Least square regression is a way of finding a straight line fitting a set of data believed to be linear. NY 13440-6916 • 1-888-RAC-USER . Reliability Analysis Center (RAC) • 201 Mill Street.

9914 -2.011 -7.0782 6.375 0.0962 0.6234 -1.69897 -0.0284 0.777 -2.6075 7.0462 -2.0535 5.5467 -1.674 -5.0683 4.0166 0.00807 Logλ(Yi) 0 -0.90309 1.51145 -0.60206 0.62 (6-6) b = log(K) = [(-30.5797 4. NY 13440-6916 • 1-888-RAC-USER .9805 3.0809 0.3579 2. Test Time 1 4 8 13 20 30 42 57 78 104 136 177 228 292 372 473 599 757 956 1205 1518 1879 2262 2668 3099 Log Cum.070 -5.1 .1335 2.2480 2.680 -7.7833 -1.8834 9.2739 3. and the instantaneous failure rate is plotted by another straight line parallel to the cumulative failure rate.36248 0.97 Using the values from Table 6-1: m = -α = 80.0931 XiYi 0 -0.3197 -1. or α = 0.7140 8.77728 -0.(55.62 = 1. Failures 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Cum.499 -1. Numerically.45(3.0 0.30102 -0.4711 -1.0117 0. the cumulative failure rate.167 0.0068 = 0.7559 1.39)/25 145.051 -2.1688 -1.8791 2.45 (6-7) A typical Duane plot is shown in Figure 6-1.935 -4.0199 0.62.39 ΣX i Yi = -80.0.262 -1.140 0.0921 -1.330 -2.STAT Section 6: Reliability Growth Testing 51 Table 6-1: Growth Data Cum.718 11.0102 0.16.16 = 1.250 0.6348 3.200 0.3947 -1.099) -0.739 12.00986 (6-8) Reliability Analysis Center (RAC) • 201 Mill Street.308 0.484 -5.42597 -0.60206 -0.296 -4.0832 3.032 -1.00899 0.4912 2 Xi 0 0.500 0.8601 -1.0338 0.58)]/25 = 0.325 -6.1551 7.121 10.0570 0.3010 1.4771 1.5597 6.115 0. calculated from the data using Equation 6-1.253 11.(-0.5705 2.2892 8.4926 10.3545 3.0810 3.933 -3.0479 0.85387 -0.0170 2. The cumulative failure rate is plotted as a straight line on the log-log scale.253 -3.627 -2.1813 3.4654 2.62)(55.9318 -1.4262 3.1819 2.6232 1.97 .81557 1.2441 -1.3847 -0.1 λ 1.1139 1. Rome.39) .8921 2.7774 2.918 -6.58 ΣYi = -30.58) 2 /25 = .93930 -1.6749 2.(55. is: cum = KT -α = 1. or K = 10 0.5518 5. Time (Xi) 0 0.5697 -0.7799 -1.0138 0.0403 0.307 ΣX i = 55.0678 0.6926 2.45 x 0.58)(-30.0168 -1.2408 1.188 2 ΣX i = 145.585 -3. at the time of the last failure.0238 0.7011 -1.1812 -0.

devised another approach to reliability growth analysis based on the assumption that reliability growth is a non-homogenous Poisson process with a Weibull intensity function (Translation: a process in which the number of failures follows a Poisson distribution. using Equation 6-2. but the Poisson distribution changes with time in such a manner that the instantaneous failure rate resembles a Weibull function). Under this assumption: cum = inst = T ( -1) βT ( -1) (6-10) (6-11) where: = = T = growth rate initial failure rate test time Reliability Analysis Center (RAC) • 201 Mill Street.099) -0. NY 13440-6916 • 1-888-RAC-USER . Larry Crow.62 = 0. is: inst = K(1 .52 Practical Statistical Tools for the Reliability Engineer STAT Failure Rate (Failures/ Hours) Cumulative Failure Rate Instantaneous Failure Rate Cumulative Test Time Figure 6-1: Typical Duane Plot The instantaneous failure rate.62)(3.α) T -α = 1.0037 (6-9) 6.45 (1 . Rome.2 AMSAA Growth Analysis Dr. then at the Army Material Systems Analysis Agency (AMSAA).0.

786 54.3944 3.6376 4. NY 13440-6916 • 1-888-RAC-USER .0415 1.3700 1.787 17.30 73.8627 2.731 29.4095 1.94460 0.6525 5.50034 0.38 154.6493 1.8797 1.1615 1.6821 3.5518 5.1761 0.31483 0.2416 2.6095 2.798 22.71369 0.0938 3. Rome.368 39.0388 6.38 238.99 i =1 Xn/Xi 3099 774.613 8.3012 3.0000 ln(Xn/Xi) 8.3306 6.4739 5.STAT Section 6: Reliability Growth Testing 53 The parameters of the AMSAA/Crow model are determined by what is called a maximum likelihood method: = N i =1 ∑ ln (T/X i ) N (6-12) where: N = T = Xi = and: = N T Using the AMSAA model on the same data that we applied the Duane model yields the results shown in Table 6-2.508 13.9594 5.1199 1.1262 2.592 10. of recorded failures total test time (= X n when test ends at a failure) time when an individual failure occurs (6-13) Reliability Analysis Center (RAC) • 201 Mill Street.3621 2.14975 0 no.0431 4.9958 3.95 103.5718 2.1736 4.6436 1.75 387. Table 6-2: Growth Data Revisited Cumulative Failure Cumulative Test Count Time at Failure (Xi) 1 1 2 4 3 8 4 13 5 20 6 30 7 42 8 57 9 78 10 104 11 136 12 177 13 228 14 292 15 372 16 473 17 599 18 757 19 956 20 1205 21 1518 22 1879 23 2262 24 2668 25 3099 24 ∑ ln (X n /X i ) = 72.

34 .34 = 1.34)(1.1) = (1. using Equations 6-10 and 6-11: cum = inst = T ( . Managing Reliability Growth.54 Practical Statistical Tools for the Reliability Engineer STAT = 25 = 0. Reliability Analysis Center (RAC) • 201 Mill Street.1) = 0.) Table 6-3: Comparison of Estimates Parameter estimated cum inst Duane Estimate at 3099 hrs.3. However.99 (6-14) = 25 (3. Rome.099) 0.0037 failures/hour 0. because they are based on less data. NY 13440-6916 • 1-888-RAC-USER .099) (0. 0. a comparison of Equations 6-8 and 6-9 to 6-16 and 6-17 shows that the Duane and AMSAA models can lead to different answers.099) (0. Table 6-3 compares the estimates at the end of the test (3. for example). and in growth data the early points are not as significant as the later points.34 .1) = 0. However.34 72.625 (6-15) At the end of test.0027 failures/hour One reason for the differences shown in Table 6-3 might be that the least squares solution treats all the data points equally. confidence intervals can be drawn around the plot by finding the limits of specified areas of a distribution (as we did in Section 4.099 hours.0027 failures/hour (6-16) (6-17) The AMSAA/Crow equations will yield the same type of straight line plots on log-log paper as the Duane equations. the assumption of a non-homogenous Poisson process with a Weibull intensity function may not quite fit. though we used the same data for both.008066 failures/hour T ( .625)(3.00986 failures/hour 0.1) = (0. Or. there are some advantages to the AMSAA/Crow model simply because a distribution is assumed.008066 failures/hour 0. AMSAA Estimate at 3099 hrs.2. Tables of the appropriate distribution for analyzing data from a non-homogeneous Poisson process are found in MIL-HDBK-189.625)(3. For example. There is no preferred solution.

In our discussion of measuring reliability. where "p" is the proportion defective and "n" is the sample size.p) n (7-2) Reliability Analysis Center (RAC) • 201 Mill Street.) p(1 . the same methods of solution apply. 7. we used confidence limits to better understand the trust we could put into a measured value of life.2.0 SAMPLING (POLLING) AND STATISTICAL QUALITY CONTROL Almost all statistical tools are designed to analyze the characteristics of a population of something from data on a sample taken from that population. "n". An analogous problem is to determine the proportion of voters in favor of some proposed government action. We also often want to measure quality. the normal applies well only when p or (1 . and solved by the same methods.) Because we are sampling a parameter distributed binomially: = where: = the standard deviation of the population = the proportion of interest (defects. yes votes. A key assumption in these applications is that the attribute of interest follows the binomial distribution (a product is defective or not. A problem analogous to this.4. and the measurements will be distributed normally. In our discussion of demonstrations.p) (7-1) p The standard deviation of the measurements. S = p(1 . NY 13440-6916 • 1-888-RAC-USER .1 Measuring Quality from Samples The central limit theorem (discussed in Section 3. We also often want to demonstrate the acceptability of a product. is a function of the standard deviation of the parent population.STAT Section 7: Sampling (Polling) and Statistical Quality Control 55 7. we used tests of hypotheses to decide how confident we were that an accept or reject decision was correct.2) tells us that when we measure a parameter such as percent defective in an infinite number of samples. is the demonstration of reliability for a "one-shot" product. and. for measures of life. etc.p) > 0. in terms of an attribute such as the proportion of products in a distribution containing defects.1 and np > 5.). from sample data. in terms of the proportion defective. Rome. "S". the mean of these measurements will be equal to the percent defective in the population from which we took the samples. we shall see. as mentioned in Section 4.1. a voter is for or against. (Caveat: for proportions. and the size of the sample. " ". failure rate or mean-time-between-failures. etc.

96 and + 1. we need the critical value of "z" marking the limits to 95% of the area under the normal curve around the mean. ˆ ˆ Put another way.96 and + 1.10 0.96 where: ˆ p is the measured proportion in the sample.80 0.98 0. people polled.95 = probability that ( p . if we desire a 95% confidence.01 0. when "z" equals the critical value.96S and p + 1. For example. Rome.645 1.58 Area Between z and -z 0. NY 13440-6916 • 1-888-RAC-USER . in the standard normal distribution 95% of all data fall between -1.33 2. we are 95% sure the true proportion is between p .96S.99 Area From z to Infinity 0.95 0. Reliability Analysis Center (RAC) • 201 Mill Street.025 0. Table 7-1 (a copy of Table 3-3) gives these values for some common areas of interest.90 0. of parts checked.56 Practical Statistical Tools for the Reliability Engineer STAT where: S = standard deviation of the distribution of sample measurements taken n = sample size (no. or: ˆ 0. as discussed in Section 3: z = ˆ p-p S (7-3) where: ˆ p = a measurement taken on one sample (of size n) p = the mean of the measurements on all samples (= the population mean) S = standard deviation of the measurements Our next step is to choose a critical value of "z" for the confidence we want.96. etc.005 As Table 7-1 shows.1.05 0.p)/S is between -1.) Since the measurements follow a normal distribution. we can convert it to a standard normal. The confidence is equal to the area under the standard normal defined as between "z" and "-z". Table 7-1: Critical Values of z Critical Value of z 1.28 1.96 2.

STAT

Section 7: Sampling (Polling) and Statistical Quality Control

57

However, since we don’t know p, we also don’t know S. Fortunately, we can approximate it ˆ by substituting p for p, yielding an expression called the standard error:

ˆ SE(p) =

ˆ ˆ p(1 - p) n

(7-4)

To illustrate this, let us discuss a political poll. Suppose someone polled 1,000 people on ˆ some question and found 500 yes votes. This means p = 500/1000 = 0.50. Our 95% confidence limits are therefore:

**ˆ Lower limit = p - 1.96S ˆ Upper limit = p + 1.96S
**

Using the Standard Error in place of S:

ˆ Lower limit = p 1.96 ˆ ˆ [p(1 - p)] n 1.96 ˆ ˆ [p(1 - p)] n = 0.5 1.96

(7-5) (7-6)

[0.5(1 - 0.5)]

1,000

= 0.469

(7-7)

ˆ Upper limit = p +

= 0.5 +

1.96

[0.5(1 - 0.5)]

1,000

= 0.5309

(7-8)

Thus we are 95% sure that the true value (p) is between 0.469 and 0.5309, which is the ˆ measured value ( p ) of 50%, plus or minus about 3%. Note that the latter figure is not 3% of the ˆ measured value of p (50%), but 3% of the total number of votes. In polling, this 3% is called the margin of error. The margin of error will change with the desired confidence. As an example, for a 99% confidence interval, the area under the standard normal curve (see Table 7-1) is between ±2.58 standard deviations. Using 2.58 in place of 1.96 in Equations 7-7 and 7-8 yield a margin of error of 4%, for 99% confidence. However, the selection of sample size can have a much more significant effect. We will leave it to the reader to verify, if he (or she) wishes, that a poll of 100 people with the same measured value (0.5), for a 95% confidence interval, would have about a 10% margin of error (0.098 to be more precise), and a poll of 10,000 people with the same result would have a margin of error of about 1%. The expense of polling 10,000 people instead of 1,000 or so is seldom considered worth the greater precision.

ˆ ˆ The margin of error is also dependent on the value of p , but is highest when p = 0.5. For ˆ example, when p = 0.1 (or = 0.9), the margin of error for 1,000 samples at 95% confidence is ˆ 0.0185 or less than 2%, as compared to 3% at p = 0.5.

Reliability Analysis Center (RAC) • 201 Mill Street, Rome, NY 13440-6916 • 1-888-RAC-USER

58

Practical Statistical Tools for the Reliability Engineer

STAT

7.1.1 Caveats These conclusions assume that the sample is truly representative of the population of interest. Fortunately for us, making a random selection of products to measure defect rate is a lot easier than assuring that people polled (and their responses) really represent the population of interest. This method presumes a large sample. If small samples are used (less than 30 or so), the Student t distribution (see Section 3.2.2) should be used instead of the standard normal. This is done by using Student t tables (see Appendix E) in lieu of Standard Normal tables in finding critical values for the desired confidence. For example, we have determined from Table 7-1 that the critical value for 95% confidence using the standard normal is 1.96. Using Appendix E, we would find a value of "t" under the column for 0.975 (when 0.975 of the area under the curve is between -∞ to "t", 1 - 0.975 = 0.025 is in one tail of the curve. Hence, both tails contain 2 x 0.025 = 0.05, or the area between "t" and "-t" is 0.95). For a sample size of 4 (three degrees of freedom), this value (from Appendix E) would be 2.353. Using 2.353 instead of 1.96 in Equations 7-7 and 7-8 yields wider confidence limits, reflecting the greater uncertainty caused by small samples. As sample size increases, the differences between Table 7-1 and Appendix E vanish. Neither the normal nor the Student t distribution apply well when p < 0.1 or np < 5. It is possible to estimate confidence limits using the Poisson distribution when p < 0.1 and the binomial when p < 0.1 and np < 5, but since these are discrete distributions, they are awkward to handle. The reliability engineer may not have enough interest to pursue this further, but quality engineers have done so. The usual practice is to approximate the distributions by a smooth curve (the higher the sample size, the better this works) and use graphical methods devised for finding the confidence limits. Some simple examples of this are presented in Statistical Methods in Engineering and Manufacturing, by John E. Brown, Quality Press, Milwaukee, 1990. 7.2 Demonstrating Acceptability Through Sampling In Section 7.1, we were concerned with measuring an attribute of a population through a sample. Quite often, we are more concerned with assuring that the value of an attribute is acceptable than knowing the value itself. For this purpose, we are not concerned with measuring to some confidence, but rather testing a hypothesis (the product is acceptable defined as having better than a stated defect rate, or the product is not acceptable, defined as having worse than a given defect rate) to some confidence (or equivalently, with some acceptable risk of error). Or, we may be concerned with the reliability of a "one-shot" device, and seek to verify that its probability of success when called upon is satisfactory to some desired confidence. The same methods will apply. A sampling test is defined by sample size and an allowable number of samples with the undesired trait. For example, take a sample of 5 units and allow no defects. The probability of passing this test will depend on the true defect rate of the product. Intuitively, we would expect a product with a defect rate of 1% to pass this test most of the time and a product with a defect rate of 50% to fail most of the time. Statistical analysis is needed to determine what "most of the time" is in terms of probability of passing, especially for products with less extreme defect rates.

Reliability Analysis Center (RAC) • 201 Mill Street, Rome, NY 13440-6916 • 1-888-RAC-USER

STAT

Section 7: Sampling (Polling) and Statistical Quality Control

59

Plots of the probability of acceptance versus the actual defect rate of a product are called operating characteristic (OC) curves. An ideal OC curve would look like Figure 7-1, where all products below a specified value would fail and all above would pass. Unfortunately, the real world is not so obliging and the typical shape of an operating characteristic curve is as shown in Figure 7-2.

1

Probability of Acceptance P(θ)

0

0%

θ θ (% Defective)

(High Qual)

100% (Low Qual)

Figure 7-1: Ideal O-C Curve

P(θ) Probability of Acceptance of a Batch of Quality θ

}α

0 0% (High Qual) Process Average θ1 θ2 θ (% Defective)

}

β

100% (Low Qual)

**α = risk of rejecting product with defect rate =
**

= risk of accepting product with defect rate =

1 2

Figure 7-2: Practical O-C Curve Figure 7-2 illustrates two types of risks involved in sampling. A customer who considers to an acceptably low 2 the worst defect rate that he can tolerate would like tests that keep value. For this reason is called the consumer’s risk. A manufacturer who considers 1 the best defect rate he can practically achieve would like tests that keep α to an acceptably low level. Hence, α is called the producer’s risk. These terms have been discussed in Section 5, Reliability

Reliability Analysis Center (RAC) • 201 Mill Street, Rome, NY 13440-6916 • 1-888-RAC-USER

Typically. however.1. rather than solve Equation 7-10. AQL stands for Acceptable Quality Level and AQL plans are designed to accept the designated AQL most of the time. Equation 3-1 in Section 3. NY 13440-6916 • 1-888-RAC-USER . and generate a set of test plans in a similar manner. In practice. and will use this fact to give an example.r)! r (7-10) The consumer can set P(accept) = for p = worst acceptable defect rate and generate a set of plans by setting values of "r" and solving the equation to determine n for each value of "r". P(r) = where: p n r P(r) = = = = the proportion defective sample size number defective probability of getting exactly r defective units in a sample of n units n! p r (1 .r 0 r!(n . These plans are referred to as LTPD Plans.r)! (7-9) If r or less failures are allowed in a test of n units. When "r" = 0. LTPD stands for Lot Tolerance Percent Defective. Rome.p) n . the probability of passing the test is: P(accept) = ∑ n! p r (1 . when "r" = 0 we can solve Equation 7-10 in closed form. These plans are referred to as AQL plans. This iterative process is an ideal job for a computer. sampling plans consider only one risk.0 = (1 .0)! (7-11) Taking logarithms of both side.p) n 0! (n . Equation 7-10 simplifies to Equation 7-11: P(accept) = n! p 0 (1 . but quite tedious when done manually. while a supplier checking his quality control will use plans based on α . a value of "p" is selected and P(accept) determined for various values of "n" and "r" until a plan giving an acceptable value of P(accept) for a reasonable sample size is produced.r r!(n .60 Practical Statistical Tools for the Reliability Engineer STAT Tests.p) (7-12) Reliability Analysis Center (RAC) • 201 Mill Street. A customer doing incoming inspections of suppliers products will use test plans based on . Equation 7-11 becomes: log P(accept) = n log(1 .α for p = the defect rate he considers realistic for his product.p) n .1. In either case the risks are computed from the binomial formula. However. The producer can set P(accept) = 1 . reprinted here as Equation 7-9. LTPD plans are designed to reject the designated LTPD value most of the time. where we derived tests that considered both risks.p) n .

Sampling Procedures and Tables for Inspection by Attributes.p) (7-13) Let us assume a consumer wants to be 90% confident that products he receives have an LTPD of 0. 7. Integrated Circuits (Microcircuits) Manufacturing.90) . a bad lot of parts.05 log (0. available from the American Society for Quality (ASQ). Sampling Procedures. Rome.02228 (7-15) Hence.4-1993. is no longer centered about the target value.0.95) . and (1 .STAT Section 7: Sampling (Polling) and Statistical Quality Control 61 or: n = log P(accept) log (1 . In this case.85.0.04576 = = 2. using a test with no failures allowed. LTPD plans are tabulated in MIL-PRF-19500K. in Appendix C. There are published tabulations of test plans for use by quality professionals. in Appendix D. The ANSI/ASQC standard replaces MIL-STD-105. P(accept) would equal one minus the producers risk or the desired confidence (0.0.p) = (1 . Special causes (e. and also in MIL-PRF-38535E. To help maintain stable processes during production. Let us suppose the producer of the same product wanted to be 90% confident that the products made had an AQL of 0. NY 13440-6916 • 1-888-RAC-USER . an inadequately trained process operator.15 = 0.0.16 log (0. tool wear..3 Statistical Quality Control A product is created by some process. When the process is satisfactory.0. the variability will consistently be between acceptable limits about a target value.g. P(accept) would be the consumers risk or one minus the confidence or (1 . General Specification For.90) = 0. Semiconductor Devices. The reader should note that these publications are not user-friendly to the neophyte and take a procedural approach which obscures the correlation of the table entries to a confidence limit. Using Equation 7-13: n = log (0.05. American National Standard.05) = 0.90) and (1 .1. Performance Specification. etc. General Specification For.) can change the variability among products in production so that it is no longer within satisfactory limits.10) .p) = 1 .95. using a test with no failures allowed. There is always some variability among products due to inherent variability in the process. Again using Equation 7-13: n = log (0.10. which has been cancelled.0. or both. Performance Specification. the producer would test 2 units and allow no failures.15.0 = = 14. In this section we will Reliability Analysis Center (RAC) • 201 Mill Street.0706 (7-14) The consumer would therefore test 14 units and accept the product if no failures occurred.85) . AQL plans are found in ANSI/ASQC Z1. For this case. the discipline of Statistical Quality Control was established.

7% of the data points should be within plus or minus three standard deviations from the mean (from Appendix C. The basic assumption is that as long as sample measurements taken periodically during production vary randomly within an expected variance. NY. we need to establish an expected variance for the data.4987.. the process is in control and needs no adjustment. by E. Reliability Analysis Center (RAC) • 201 Mill Street. Non-random measurements or indications of variance outside the expected distribution show some special influence at work which can be found and corrected to restore the expected performance. Grant and R. Rome. NY 13440-6916 • 1-888-RAC-USER . the results can be presented on a run chart.L. at z = 3. Note that in the automotive industry. Should the data plot exceed these limits.62 Practical Statistical Tools for the Reliability Engineer STAT discuss these statistical tools and also some statistical tools that can help define an acceptable process. is also known as Statistical Process Control (SPC).9974 = 99. control limits are often set higher than three standard deviations in automotive plants.4987 = 0. the area from 0 to z is 0. Hence. We can therefore establish control limits at three standard deviations from the mean with little chance that samples from a stable process will exceed them.1 Control Charts If samples of a process are taken over time. An authoritative reference for SQC is Statistical Quality Control. With the limits in place. 1 2 3 4 Data Samples 5 6 Figure 7-3: Run Chart To determine the significance of the data in Figure 7-3.7%). or the area between "z" and "-z" is 2 x 0. Most other applications are satisfied using three standard deviations. and a parameter of interest measured. and that 99.0. Taking advantage of the central limit theorem. such as Figure 7-3. Leavenworth. 7. McGrawHill Book Co. The sample data are plotted and analyzed on control charts. Statistical Quality Control (SQC).S. using control limits encompassing even 99.7% of expected variation would result in too many false alarms.3. we assume that the data will be normally distributed about the process mean. the run chart has become a control chart. where many thousands of control charts are used. 1989 (6th Edition). we can feel we have good evidence that the process has gone out of control (see Figure 7-4).

NY 13440-6916 • 1-888-RAC-USER .3S 1 2 3 4 Data Samples 5 6 Figure 7-4: Control Chart The control limits are determined from the standard deviation of the distribution of the sample mean measurements.2 Control Charts for Variables Controlling variables. The process target can also be used as the centerline.STAT Section 7: Sampling (Polling) and Statistical Quality Control 63 Indicates Process Out of Control Upper Control Limit (UCL) = Mean + 3S Centerline = Process Mean Lower Control Limit (LCL) = Mean . for example a desired part diameter. This is obtained by dividing the standard deviation of the process by the square root of the size of one sample. can be done with the aid of a control chart called the X chart. The standard deviation of the process. The upper and lower control limits would be: UCL = X + 3S′ n 3S′ n (7-16) LCL = X − where: (7-17) X = the centerline value n = the number of units in a sample S′ = the process standard deviation Reliability Analysis Center (RAC) • 201 Mill Street. such as part diameter. 7.. tensile strength. in turn. Rome. or the mean of the means of many equal sized samples. The center line is the process mean determined by the average measurement of a (hopefully large) sample.3. The procedure for doing this will vary dependent on the distribution of the parameter of interest. This looks like Figure 7-4. can be estimated from the sample data. which ordinarily follow a normal distribution. etc.

as such.64 Practical Statistical Tools for the Reliability Engineer STAT The process standard distribution.X ) n S’ = where: Xi = = = i =1 n -1 (7-18) mean of one sample mean of sample means number of samples taken X n 7. UCL Mean Variation X LCL UCL R LCL Range Variation Figure 7-5: X and R Chart Combination Range is simply the difference between the highest and lowest measurements of the units in the sample.3. NY 13440-6916 • 1-888-RAC-USER . It is necessary to determine the average range. as before. as shown in Figure 7-5. we will also usually be interested in the variation about the mean. If the mean of a sample is on target. we do not have a good situation. This value becomes the centerline of the R chart. However the control limits for both charts are determined from the following formulas: UCL(X) = X + A 2 R (7-19) Reliability Analysis Center (RAC) • 201 Mill Street. The centerline of the X chart is determined by the mean of many sample means. S’.3 Range Charts If we are interested in the mean of a variable such as product diameter. preferably from the mean of many samples. and. It is a measure of variation. One way to do this is to use the formula: 2 ∑ (X i . For this reason. but the variation is too great. is determined from sample data. can be used instead of the process standard deviation in setting control limits for both the X and R charts. the X chart is often accompanied by a Range chart. Rome.

000 + 1.37 0.72 1.22 0.000 .57 2.00 1.000.18 0.00514 LCL(R) = D3 R = 0(0.000 + 0.58 0.27 2.88 1. Rome.18 D3 0 0 0 0 0 0. and are presented in Table 7-2.002) = 0.65 1.08 0. Table 7-2: Statistical Constants Sample Size 2 3 4 5 6 7 8 9 10 12 15 20 A2 1.00204 LCL(X) = X .0.28 0.86 1.99796 UCL(R) = D4 R = 2.57(0.42 0.34 0. NY 13440-6916 • 1-888-RAC-USER . then X = 1.002.59 For example.002 inches.1.78 1.14 0. if we are producing rods with a mean length of one inch and find our mean range in many samples is 0.02 0.11 2. assuming the processes measured follow a normal distribution.002) = 1.31 0.48 0.02(0.00204 = 0. D3 = constants (see Table 7-2) The constants in the control limit formulas have been worked out by statisticians.41 D4 3.27 0.A2 R = 1.73 0.002) = 1.82 1.28 2. R = 0.002) = 0 (7-23) (7-24) (7-25) (7-26) Reliability Analysis Center (RAC) • 201 Mill Street.35 0. For a sample size of three: UCL(X) = X + A2 R = 1.00204 = 1.02(0.92 1.000 .STAT Section 7: Sampling (Polling) and Statistical Quality Control 65 LCL(X) = X − A 2 R UCL(R) = D 4 R LCL(R) = D 3 R (7-20) (7-21) (7-22) where: UCL(X) LCL(X) UCL(R) LCL(R) = = = = upper control limit for X chart lower control limit for X chart upper control limit for Range chart lower control limit for Range chart X = process mean = mean of sample means A2.22 0. D4.

if the plot does not look random. For the former.e. we could be interested in the proportion of defective product or the defect rate. rather than has in some range of value. This was discussed in Section 7. The only way to improve the results would be to change the process to a better one. even if the control limits are not exceeded. etc. and often provides clues about the problem.3. where we were concerned with measuring the attribute of interest. going from close to the Upper Control Limit to Close to the Lower Control Limit without any points in between. One example could be a process done on either of two machines that are adjusted differently.1.2.66 Practical Statistical Tools for the Reliability Engineer STAT 7. defects per cubic foot. the end result would be to double the natural variance of the process. A ball is red or it is not. Of more interest.4 Interpreting Control Charts A plot running outside the control limits is not the only sign of unexpected variation. for example. and every time a sample measures above the center line an adjustment is made to produce shorter rods. where we were concerned with demonstrating the attribute of interest.5 Controlling Attributes Variables such as dimensions are not always the parameter of interest. for example. For example. Hence. while for the latter we are interested in such ratios as defects per product. a plot hugging the control limits. We may be concerned with an attribute.3. if the plot is random and between control limits. you have grounds for concern. If. NY 13440-6916 • 1-888-RAC-USER . 7. However. If defects are our concern. which is something that a product either has or lacks. taking an AQL approach. probably the misrecording of data because of a fear of reporting measurements off target. we would not care whether or not a product had more than one defect. a machine is making rods and every time a sample measures below the center line an adjustment is made to produce longer rods.. the process should not be adjusted. do not tinker with it. Rome.2) and the defect rate by a Poisson distribution. we use different Reliability Analysis Center (RAC) • 201 Mill Street. The proportion of defective product is described by a binomial distribution (as discussed in Section 7. In summary. Statistical Quality Control for attributes is most akin to the latter. Some possible signs of non-random variation (i. and in Section 7. a plot hugging the centerline (all points falling within one standard deviation of the centerline) shows some abnormal condition. If the process is in control. trouble) are: • • • • • • One or more points outside the control limits 10 consecutive points above the center line or below the center line 7 consecutive points in a steadily increasing or steadily decreasing pattern 14 points alternating up and down 15 consecutive points within plus and minus one standard deviation of the center line 2 consecutive points in the band between plus two and plus three standard deviations or in the band between minus two and minus three standard deviations. shows some factor switching the distribution from high to low. Any plot that does not appear to be random is evidence of loss of control. a product is defective or it is not. In contrast.

This serves as the centerline for the p chart. NY 13440-6916 • 1-888-RAC-USER .p). we have: UCL = p + 3 p(1 − p)/n LCL = p − 3 p(1 − p)/n (7-28) (7-29) It is possible for the calculated LCL to be a negative number.5.STAT Section 7: Sampling (Polling) and Statistical Quality Control 67 formulas in determining the control limits. The process standard deviation would be the square root of p(1 .1 Proportions The control chart for proportions (called a "p" chart) is derived from the value of the proportion of interest. It also makes a difference whether or not a constant sample size is used. so the LCL in such cases is set to zero. as shown in Figure 7-6. Rome. Effect of Smaller Sample UCL Effect of Larger Sample Centerline LCL (0) Min LCL = 0 Figure 7-6: "p chart" for Different Sample Sizes Reliability Analysis Center (RAC) • 201 Mill Street. the control limits would be different for every sample. 7. If the sample size is not constant. and the standard deviation of samples containing "n" units would be: S = p(1 − p) / n (7-27) Again setting control limits at plus and minus three standard deviations (of the distribution of sample measurements) from the center line. estimated by the number of defective parts divided by the number of parts tested in a large sample. This has no meaning.3.

SQC may be used to monitor its stability. but not necessarily satisfactory. the average defect rate.5. when the product can be described by a normally distributed parameter. To compute Cp.6 Caveat: "In Control" May Not Be "In-Spec" It is important to note that a process that is "in control" is not necessarily one meeting specified limits. Hence: UCL = c + 3 c LCL = c − 3 c (7-32) (7-33) These formulas assume a Poisson distribution for the process. we have: UCL = u + 3 (u / n ) LCL = u − 3 (u / n ) (7-30) (7-31) For constant sample sizes. it is necessary to determine the standard deviation of the parameter of Reliability Analysis Center (RAC) • 201 Mill Street. estimated from a large sample. NY 13440-6916 • 1-888-RAC-USER . the average number of defects. and. The standard distribution of both the process and the sample’s number of defects is the square root of c. the standard deviation of the process would be the square root of the average rate. 7.68 Practical Statistical Tools for the Reliability Engineer STAT 7. and are most accurate when the sample size is ≤ 10% of the size of the population sampled. "u".3. is the centerline.3. Rome. An extension to this procedure is the calculation of figures of merit describing the relationship between the distribution of the parameter of interest and the designated acceptable range for its values. short of changing the process itself. Hence. can be used for the centerline. We will now discuss some measures that can be used to determine the capability of a process to create products within specified limits. the standard deviation can be estimated from a sample and then used to calculate the proportion of the product outside any given limits. setting the control limits as usual. and the sample standard deviation would be "u" divided by the square root of the sample size.1 Measuring Process Capability Process Capability (Cp) is one measure of the ability of a process to produce acceptable products. 7.2 Rates If we are interested in the defect rate or defect density. and no amount of tinkering will improve the product. "c". described in Section 8. once a satisfactorily capable process is installed. it is doing as well as can be expected. These measures assume the process is stable. to assure that the probability of finding a defect remains reasonably constant as samples are selected and removed from the population. For rates. Necessary changes can be determined by the statistical design of experiments.6.3. As discussed in Section 3. If it is in control.

In current conditions.STAT Section 7: Sampling (Polling) and Statistical Quality Control 69 interest. It is not until Cp equals or exceeds 1. This means that all the product from the target value to plus and minus 3 are within the specified limits.7% of the product is in the range defined as acceptable. as shown in Figure 7-7. Cp = 1. Therefore.x ) n = where: = xi = x = n = i =1 n -1 (7-34) standard deviation ith measurement mean of all measurements number of measurements Cp is then calculated by: Cp = USL . 99. a Cp of 1. The "6 " quality program initiated by Motorola aims for a full 6 Reliability Analysis Center (RAC) • 201 Mill Street.LSL 6 (7-35) where: USL = Upper Specification Limit (highest value considered acceptable) LSL = Lower Specification Limit (lowest value considered acceptable) = Standard Deviation Where the difference between the USL and LSL is 6 .3. this is considered marginal. Rome. LSL Target USL Figure 7-7: Process Capability (Cp) Chart As mentioned in Section 7.1. This is done by measuring the parameter in a sample of the product and using the formula: 2 ∑ (x i . NY 13440-6916 • 1-888-RAC-USER .7% of a product is distributed between plus and minus 3 .0 means that 99.3 that a process is considered good.

For this reason. Most industrial and commercial practices are. their process would be equivalent to a "6 " process (one having a Cpk of 1.).4 time in a million queries.2 Measuring Process Performance Process Performance (Cpk) measures the capability of a process when the mean value of the parameter of interest is not necessarily on target. which equates to 3.LSL)} 3 (7-36) where: Min = smaller of the two values = process mean other terms as defined above Motorola’s "6 " goal is a Cp of 2. Such cases can be related to the others through the error rate. the shortest distance from the process mean to either specification limit is 4.4 parts in a million "out of spec".6. Rome. described in Section 7.3.0 and a Cpk of 1. 7.0.200 per million opportunities. Cp cannot be translated into percent "in-spec". taxpayer questions to the IRS are answered either correctly or not. as shown in Figure 7-8.5). However. The "6 " philosophy can also be applied to processes parameters not described by a normal distribution. For example.2. having error rates around 6. a binomial process.6. If the IRS answered incorrectly only 3. Lower Spec Limit Target Value Process Mean Upper Spec Limit Figure 7-8: Process Performance (Cpk) Chart Cpk is calculated by the formula: C pk = Min {(USL . NY 13440-6916 • 1-888-RAC-USER .3. about "4 " processes. since the mean of a process is also subject to variation. Motorola also utilizes a process performance figure of merit. a Cp of 2. Reliability Analysis Center (RAC) • 201 Mill Street.5.5 off target. This means that when the process mean is 1. ( .70 Practical Statistical Tools for the Reliability Engineer STAT between the target value and either of the specification limits (a total spread of 12 between the upper and lower specification limits). or are equivalent to. Motorola does not assume the target value will be centered. Where the process mean is off target.5 .

. The key to test efficiency is the use of "orthogonal arrays".1 Designing Experiments The first consideration in designing an experiment is the identification of the factors to be tested. like the use of flux. if necessary. but this is inefficient and would not show any interactions between factors (e. We will do so by setting process parameters to values determined by the outcome of the experiment. one well suited for a range of use environments). We can also evaluate the significance of our results using tools from a family of methods called the analysis of variance (ANOVA). we can employ the statistical design of experiments (DOE).0 USING STATISTICS TO IMPROVE PROCESSES If a process is in control (see Section 7). the parameters of the process are changed. This could be solder defects per card for the wave solder example. For example. For parameters that are either present or absent. These are the process parameters that we can control and that we believe will affect the product. the height of the solder wave. and far enough apart so any affects are noticeable. A value of zero would correspond to a setting of 100 degrees which is not a test setting. the values are coded. the only way to improve it is to change it. We will need a high and low setting for each factor. normalize or maximize the output. but much simpler to use only two. more often. and each low setting given the value minus one ("-" on the test matrixes). perhaps a higher solder wave works better with colder solder than with hotter solder). and the other setting will represent the absence of the factor. To most efficiently find desired improvements. NY 13440-6916 • 1-888-RAC-USER . or. For analysis purposes. A new process may be created. and to examine interactions of factors. each high setting given the value plus one (indicated on test matrixes by "+").e. Notice we may want to minimize. rod length for a rod manufacturing process. a setting of 120 degrees would be valued as plus one and one of eighty degrees valued as minus one. Reliability Analysis Center (RAC) • 201 Mill Street. but may be a solution after analysis (to be discussed later). We can use DOE to find the optimum process parameters for a defined use environment. It is possible to use more than two levels of a factor. or add flux to the mixture. We also need a measure of the process output. We could test the effects of each possible change individually. "high" or "low" will represent the presence of the factor. we might change the temperature of the solder. An example for a two parameter test is given by Table 8-1. to improve a wave solder process. or miles per gallon for an automobile. The next step is to set factor levels to be used as test settings. wave height and use of flux in a wave solder process.. Hence. 8. "High" and "low" are arbitrary terms and it does not matter if the "low" value (of temperature. for example) is actually greater than the "high" value. or adapt it to find a robust design (i. one setting. These are matrices describing test settings that allow the effects of each factor to be separated from the others.g.STAT Section 8: Using Statistics to Improve Processes 71 8. Examples given above were the temperature. Rome. if the high value of temperature were 120 degrees and the low value were 80 degrees. The two settings should be close enough together to assure that any difference in the outcome caused by the factors is reasonably linear between the high and low test settings. It is a good idea to have the factors selected by a team of people involved with the process.

g.72 Practical Statistical Tools for the Reliability Engineer STAT Table 8-1: Two Factor Orthogonal Array Run 1 2 3 4 A + + B + + In Table 8-1. The actual running of the tests should be done to minimize the probability of any bias from factors that are not being tested. Reliability Analysis Center (RAC) • 201 Mill Street. such as temperature and wave height. processing 100 cards through the wave solder machine) with the settings as shown in the matrix. Each test run is a repetition of the established test (e. the interaction is "low". NY 13440-6916 • 1-888-RAC-USER . The most ideal situation would use several iterations of each run performed in a random order. if there were some concern that the workers involved may make a difference. Environmental effects such as humidity could be compensated for by repeating the same test run on different days and averaging the outcomes. Table 8-2: Expanded Test Matrix Run 1 2 3 4 Factors (test settings) A B + + + + Interactions (by-products) A*B + + Results (measured outcomes) Setting both factors A and B to "high" or both to "low" is equivalent to setting their interaction to "high". The orthogonal array of Table 8-1 is a full factorial array in that all combinations of high and low settings for all factors are tested. The general form of this is shown in Equation (8-1) for the array shown in Table 8-3. Expanding Table 8-1 to include the interaction of the two factors. For other combinations. Once the outcome for each run is determined. we get the matrix shown in Table 8-2. A plus in the matrix under a factor means the high setting is used during the corresponding test run and a minus means that the low test setting is used. a linear regression is performed to determine the optimum settings for the parameters tested.. Such an array will also permit the analysis of all possible interactions between factors as a by-product. Rome. The analysis will not differentiate between factors that are set in the experiment and factors that are defined as byproducts. all the tests would be run with the same workers. A and B represent test factors. For example. and providing a column for results.

it would have no impact changing the output from its average value ( Y ). Rome.g. These values are then translated to settings (e. it reduces the output by ( ∆ A)/2. low setting = -1) ∆B . and when A is +1.STAT Section 8: Using Statistics to Improve Processes 73 Table 8-3: Orthogonal Array Run 1 2 3 4 AVGAVG+ A + + Factors (test settings) B + + Interactions (by-products) A*B + + Results (measured) y1 y2 y3 y4 ∆ (y1 + y2)/2 (y2 + y3)/2 (y1 + y3)/2 (y3 + y4)/2 (y1 + y4)/2 (y2 + y4)/2 (AVG+) . Factors B and A*B have similar effects. minimum or nominal). as shown in Table 8-4. NY 13440-6916 • 1-888-RAC-USER . These values of " ∆ " are used in Equation 8-1 multiplied by the coded values of the factors. ∆ B and ∆ (A*B).= Average outcome when the factor in a column is at its low setting Equation 8-1 merely quantifies an assumption that the output will change from its mean value linearly with changes in the factors. ∆C . A AVG+ = Average outcome when the factor in a column is at a high setting AVG. If A were zero (the mid value between -1 and +1). let us use some hypothetical values. To illustrate.. B. based on the measured differences in the output. When A is -1. Y Y ∆A A = = = = Reliability Analysis Center (RAC) • 201 Mill Street.(AVG-) values from column A in matrix coded value of A (high setting = +1. Hence. The difference between "AVG+" and "AVG-" ( ∆ ) for each factor is the change in Y as the factor varies from -1 to +1 (its low and high values). temperature) to be used in the new process. The regression equation is then used to find values of the factors (between plus and minus one) which give the desired output (maximum. as the factors vary from low to high. Equation 8-1 predicts the outcome of a process for any values of the factors between the high and low values tested.(AVG-) for each column Y = Y + ∆A ∆B ∆(A * B) A + B + (A * B) 2 2 2 (8-1) where: expected output average output = (y1 + y2 + y3 + y4)/4 (AVG+) . it increases the output by ( ∆ A)/2. similar to ∆A . C.

Hence. If it is not. Then we would set the vales of A and B between plus and minus one so that Y = 5. If B is a parameter which can only be present or absent. so Y = 7 . A three factor. such as miles per gallon. A and B would be set to their high values. The optimal setting should also consider the costs involved.0. This may not be possible. the high settings are represented by plus one.74 Practical Statistical Tools for the Reliability Engineer STAT Table 8-4: Sample Test Results Run 1 2 3 4 AVGAVG+ ∆ A + + 9 5 -4 B + + 8 6 -2 A*B + + 7 7 0 Y 10 6 8 4 Y=7 Y = 7 . If the outcome were something we want to maximize. We would expect the average defect rate to change from 7 to 4 by using the high settings of both factors.g.1 = 4. It may.5 (a setting three quarters of the difference between the low setting and the high setting) for Y = 7 . The simple array of our example can be expanded to handle any number of factors. and B would be set at a value mid way between the high value and the low value.(-2) .B Using the equations given in Table 8-3.2 . it is good practice to perform a test at those settings to confirm that the expected new output is indeed achieved.1 . This should be identified and made a factor in a new set of experiments.. NY 13440-6916 • 1-888-RAC-USER . In that case. Hence.(-1) = 7 + 2 + 1 = 10. it can only take the values plus and minus one. be less costly to change A than to change B.2A .5) . Suppose the outcomes were a measurement that we wanted to be 5. Reliability Analysis Center (RAC) • 201 Mill Street. it indicates there is another factor at work that was not tested. One way would be to set A at plus one and B at zero. the outcomes listed in Table 8-4 result in the regression equation Y = 7 . There will always be 2n test runs. We would raise the average gas mileage from 7 to 10 mpg by setting factors A and B to their low values.B + 0(A*B) = 7 . B could be set to plus one and A set to 0. Y = 7 .0 = 5.(2 x 0. Rome. A would be set at the high value represented by plus one. we would want to minimize it.1 = 5.2 . we would maximize the regression equation by setting both A and B to their low settings (-1). and hence the above solution could not apply. full-factorial matrix would be as shown in Table 8-5. Once the optimal settings are derived. Y = 7 . In the equation. new experiments can be devised to see if settings outside the range tested can produce more improvements. for example.B.1 = 7 . a defect rate).2A . If Y represented some undesirable quantity (e. where n = the number of factors. Once the improvements have been verified and initiated.2A . In that case.

NY 13440-6916 • 1-888-RAC-USER .1. Saturated arrays are also called Taguchi arrays after Genechi Taguchi. When it is reasonably safe to assume there will be no interactions. the interaction columns are used to determine test settings for additional factors. strength. For example. 8. of course.STAT Section 8: Using Statistics to Improve Processes 75 Table 8-5: Three Factor Full-Factorial Array Run 1 2 3 4 5 6 7 8 Factors (test settings) A B C + + + + + + + + + + + + Interactions (by-products) A*B B*C A*C A*B*C + + + + + + + + + + + + + + + + Outcomes 8. defects) in order to observe the effects of the factors on all important considerations.1. triple interactions are quite rare. Table 8-6: Saturated Array (Table 8-2 Modified) Run 1 2 3 4 A + + Factors (test settings) B C (replaces A*B) + + + + Results (outcomes) Thus. three factors can be tested in the same number of tests used for two factors in a fullfactorial array. Rome. a leading proponent of their use. such as atmospheric pressure. It is also possible to use hybrid arrays in which some of the columns of expected interactions are not used for additional factors. a great economy can be achieved by using saturated arrays.2 Testing for Robustness We have considered only one outcome per experiment. but other columns are. but Risky In our numerical example (Table 8-4). The test will then show all the interactions of the other factors and the effects of the new factor without any additional runs. Another variation is to perform the experiment under differing values of an uncontrollable factor. as shown in Table 8-6. so the column in Table 8-5 for A*B*C can often be commandeered for a new factor. The risk is. It is also possible to use multiple outcomes (dimensions. providing the new factor is not likely to interact with any of the others. The best solution would then be the one that provided the best overall results. that some interaction is significant and its effects will be confounded with the new factor using the same settings. the interaction of A and B had no effect. The matrix of Table 8-5 would permit the testing of seven factors in lieu of the three it is designed for. in the array of Table 8-2.1 Saturated Arrays: Economical. Desired "settings" of uncontrolled factors Reliability Analysis Center (RAC) • 201 Mill Street. a factor "C" can be tested using the settings in the column representing the by-product A*B. This is often the case. For example. In these arrays.

76

Practical Statistical Tools for the Reliability Engineer

STAT

can be obtained either by waiting for the factor to assume a desired test value or by using special test equipment, such as environmental chambers, to simulate the factor. Again, the preferred solution is a set of settings for the controllable factors which give the best overall results. For example, Table 8-7 shows outcomes for three controlled and two uncontrolled variables. If the desired result was the lowest output (it might represent defect rate, for example) and the uncontrolled factors were equally likely to be at their low and high values, the settings for test run 6 would be preferred, even though under some conditions other test settings produced better outcomes. Table 8-7: Testing for Robustness

Controlled Factors Settings Test A B C Run 1 2 + 3 + 4 + + 5 + 6 + + 7 + + 8 + + + Uncontrolled D Factors E Uncontrolled Factors "Settings" + + + + 9.0 1.8 1.6 7.8 2.6 2.0 2.3 4.8 2.7 2.0 1.5 3.4 2.2 1.5 1.7 1.9 2.4 1.6 1.5 2.9 1.9 1.7 1.7 1.8 3.3 1.6 1.6 3.3 2.6 1.8 1.6 1.9

Finally, Taguchi uses outcomes expressed as signal to noise measures, which consider both the mean and the variation in the output. Interested readers are invited to pursue these avenues on their own. A comprehensive basic text on the subject is Understanding Industrial Designed Experiments, by S.R. Schmidt and R.G. Launsby, Air Academy Press, Colorado Springs, CO, 1989. 8.2 Is There Really a Difference? Recognizing the existence of variance in all data, one should always consider that a difference in two measurements of a parameter, such as the measured outcome of an experiment, might be due to chance and not to other factors. For example, consider the data shown in Table 8-8. Table 8-8: Defect Data

Day of the Week Monday Tuesday Wednesday Thursday Friday Total Average Defects per 100 Units Produced Day Shift Night Shift 1 3 2 7 6 7 4 6 7 2 20 25 4 5

There seems to be a difference between the defect rate of the day shift and that of the night shift. However, there is a wide range in the data for each shift, and we should question the validity of assuming there is really a difference between the shifts. To resolve the issue, we will use a statistical technique from a family of techniques known as ANOVA.

Reliability Analysis Center (RAC) • 201 Mill Street, Rome, NY 13440-6916 • 1-888-RAC-USER

STAT

Section 8: Using Statistics to Improve Processes

77

ANOVA stands for Analysis of Variance, and is intended to provide means to separate the influences of many different factors on a parameter of interest. The specific ANOVA application which can determine the significance of the difference between two sets of data requires the following assumptions: 1. 2. 3. 4. The data points follow a normal distribution The data points are independent from each other Variability is about the same in each set of data The data sets we are comparing have the same number of data points

The basic premise we shall use is that if the data sets were really different, there would be a wider variation between the data sets than within the sets. The variance within the data could be estimated from the variance in either data set, but under our assumptions, we can use both data sets and calculate a quantity called the Mean Square Error (MSE) from the formula:

2 ∑ (n - 1) Si k

MSE = 1 where: k n

k(n - 1)

(8-2)

= number of data sets = number of measurements in each set

2 Si = Variance of one data set

and:

2 Si

=∑

1

n

( y i - y i )2

n -1

(8-3)

where:

yi = yi =

value of one data point in a data set mean value of data in the set

The term k(n - 1) is called the degrees of freedom provided by the data. From the data in Table 8-8, the degrees of freedom = 2(5 - 1) = 8.

Reliability Analysis Center (RAC) • 201 Mill Street, Rome, NY 13440-6916 • 1-888-RAC-USER

78

Practical Statistical Tools for the Reliability Engineer

STAT

The variance between groups, called the mean square error between groups (MSB) is computed by:

MSB =

where:

n k 2 ∑ (y i - y ) k -1 1

(8-4)

y = mean of all data sets, all other terms as defined above k - 1 = the associated degrees of freedom for MSB, = (2 - 1) = 1

If there is a real difference between data sets, the MSB should be greater than the MSE; otherwise the ratio should be close to one. Under the assumptions listed, the ratio of MSB to MSE follows an F distribution. Since this is so, we can use tables of the F distribution, such as the one in Appendix F, to determine whether or not a ratio calculated from the measured data is compatible with a ratio of 1.0 (i.e., the hypothesis that there is no real difference), within a specified risk.

MSB = F MSE

(8-5)

**Using the data in Table 8-8,
**

MSB = 5 [(4 - 4.5) 2 + (5 - 4.5) 2 ] = 2.5 2 -1

2 (5 - 1)Sday + (5 - 1)S 2 night

(8-6)

MSE =

2(5 - 1)

5 2 ∑ y i - y day 1

(8-7)

2 S day

=

(

)

5 -1

=

(1 - 4)2 + (2 - 4)2 + (6 - 4)2 + (4 − 4)2 + (7 - 4)2

4

=

(- 3)2 + (- 2)2 + (2)2 + (0)2 + (3)2

4

5 2 ∑ y i - y night 1

=

9+4+4+0+9 26 = = 6.5 4 4

(8-8)

S2 night

=

(

)

5 -1

=

(3 - 5)2 + (7 - 5)2 + (7 - 5)2 + (6 - 5)2 + (2 - 5)2

4

=

(- 2)2 + (2)2 + (2)2 + (1)2 + (- 3)2

4

=

4 + 4 + 4 +1+ 9 22 = = 5.5 4 4

(8-9)

Reliability Analysis Center (RAC) • 201 Mill Street, Rome, NY 13440-6916 • 1-888-RAC-USER

2 4 7.94 6. it marks the border of an area of the distribution equal to our acceptable risk (i. A simple way is to plot paired measurements on a scatter diagram. If our calculated F statistic exceeds the value in the table for the degrees of freedom provided by the data.2 19. we use the 0. if we want to analyze the relationship between the defect rate of a wave Reliability Analysis Center (RAC) • 201 Mill Street.32 and our calculated F statistic is 0.. we must accept the hypothesis that there is no real difference between the data sets.e. Excerpts from such a table are given in Table 8-9. Hence.STAT Section 8: Using Statistics to Improve Processes 79 Thus: MSE = 4(6. the probability of being wrong). which does not have data points for eight degrees of freedom. This procedure can be used to test the significance of differences in outcomes of statistical experiments to avoid making expensive process changes when the differences are caused by statistical variation rather than the factors tested.5) 26 + 22 48 = = = 6 8 8 8 (8-10) and: F = MSB 2. If we are willing to be wrong no more than 5% of the time. For example. and the percentile is equivalent to the risk we take (i.417 MSE 6 Tables of the F distribution are organized by degrees of freedom and percentiles (see Appendix F). such as the outcome of an experiment and one of the factors.26 3.8 degrees of freedom is 5. the probability of the statistic being in the area cut off is equal to our defined risk when the hypothesis that there is no difference in the data is true).59 6. we can reject the hypothesis that there is no real difference with no more than a risk equal to the percentile (5% using the above table). 8.69 The table lists what are called critical values.71 6.0 19. we extracted Table 8-9 from a more extensive table than the one in Appendix F.05 Significance Critical values of F for 0. (To avoid extrapolation.417.05 percentile table. In this case.5 = = 0.32* 4.05 risk Degrees of Degrees of Freedom for MSB Freedom for 1 2 3 4 MSE 1 161 200 216 225 2 18. The degrees of freedom are a function of the data. the table value for 1.84 * Critical value for data used in the example 5 230 19.5 19.3 6.07 3. Rome.3 How Strong is the Correlation? Often it is useful to know the relationship between two variables.e. NY 13440-6916 • 1-888-RAC-USER . As discussed many times earlier in this text..) Table 8-9: Critical Values for F at 0.39 8 5.5) + 4(5.46 4.

Rather. SOLDER TEMPERATURE Figure 8-1: Scattergram In interpreting the scatter diagram. the better the correlation. A figure of zero indicates no relationship between the variables. Although an eyeball analysis may be sufficient for many uses. the slope of a scattergram is not an indicator of correlation. unlike most charts. let us use the following data shown in Table 8-10. A negative result shows an inverse correlation (as one factor increases. Hence. Rome. o o o o o o o o o o o o o o o o DEFECT RATE . Reliability Analysis Center (RAC) • 201 Mill Street. we could measure defect rate at various temperatures and plot the results on a chart with temperature as one axis and defect rate as the other. NY 13440-6916 • 1-888-RAC-USER . a quantitative evaluation of correlation is often quite useful. the other decreases) and a minus one is a perfect inverse correlation. the width of the cloud is the indicator. To illustrate.80 Practical Statistical Tools for the Reliability Engineer STAT solder process and the solder temperature. The narrower the cloud. This is easily. r = (∑ Dx )(∑ Dy ) 2 2 ∑ Dx Dy (8-11) where: Dx = the difference between a value of x and the mean of the values of x = ( x i . accomplished using the correlation coefficient. it is important to note that the slope of a line drawn through the cloud of data points is an artifact of the scales of the axes.y) Perfect correlation would be shown by a correlation coefficient equal to one. Figure 8-1 shows such a plot.x) Dy = the difference between a value of y and the mean of the values of y = ( y i . if somewhat tediously.

but the data cloud is wide.010 0. we will compute the correlation coefficient.030 Defect Rate . Computing the terms we need to solve for the correlation coefficient is made easier by using a data table such as the following shown in Table 8-11.040 . and the correlation may or may not be significant. Reliability Analysis Center (RAC) • 201 Mill Street.020 .030 0.010 0.035 0.020 0. NY 13440-6916 • 1-888-RAC-USER . Rome.010 250 255 260 270 Temperature 280 290 295 Figure 8-2: Scattergram of Data in Table 8-10 This appears to show some correlation.040 0.STAT Section 8: Using Statistics to Improve Processes 81 Table 8-10: Paired Data Data Pair 1 2 3 4 5 6 7 8 9 10 x (Temperature) 250 255 260 265 270 275 280 285 290 295 y (Defect Rate) 0. To resolve the issue.025 0. .020 0.030 0.015 The scattergram for this data is given in Figure 8-2.

25 506.14625 -0.0135 +0.23625 -0.030 0. Reliability Analysis Center (RAC) • 201 Mill Street.0235 Solving Equation 8-11: r = .5 2.5 Dx 2 Dy (y .00000225 0. isolation of the truly critical factors is required.0115 +0. However.25 56.25 306.5 17.5 7.020 0. but it is possible that both are changing in response to some other factor.0009528) = .02625 -0.5 12.5 -12.x ) 22. changes in defect rate may be caused by changes in temperature.19125 x 0 506.040 0.00004225 0.00018225 0.25 156.20625 -0.0085 Dy2 0.00001225 0.0065 +0.25 6.82 Practical Statistical Tools for the Reliability Engineer STAT Table 8-11: Data Analysis Data Pair 1 2 3 4 5 6 7 8 9 10 x 250 255 260 265 270 275 280 285 290 295 y 0.5 -17.0.0035 +0.5 0.015 y Dx (x . Caveat: Note that correlation does not necessarily mean causation. As one goes down the other goes up.5 0.00027255 0.8375 (2062.0035 +0.5 -7. The moral is that to improve the process.8375 2062.5 -2.5974 (8-12) The result shows a fair negative correlation between temperature and defect rate.00018225 0.25 306.5)(0.025 0.25 2 ∑ Dx 0 ∑ DxDy 272. Rome.25 56.0165 -0. NY 13440-6916 • 1-888-RAC-USER .00875 -0.035 0.00001225 0. Also.00013225 0. one can imagine scenarios where the defect rate could cause temperature changes (suppose the operator reacted to defects by adjusting the machine in a way that changes the operating temperature).04875 +0. there is a lot of noise in the data and predicting one factor from the other can be done only roughly.010 0.0.0065 -0.020 0.00007225 2 ∑ Dy DxDy -0.0135 -0.010 0.0015 -0.25 6.5 -22.03375 +0. For example. the speed of the solder flow might be the ultimate cause of both temperature changes and defects.04375 -0.0009528 -0.y ) -0.25 156.030 0.08625 -0. In the example.00004225 0.

NY 13440-6916 • 1-888-RAC-USER . and a useful reference for its practical application in reliability engineering.STAT Section 9: Closing Comments 83 9. the two disciplines intersect quite often in the problems encountered by the reliability engineer. Reliability Analysis Center (RAC) • 201 Mill Street. The Reliability Analysis Center hopes the reader has found this book a relatively painless introduction to the world of statistics. Rome.0 CLOSING COMMENTS While both statistics and reliability engineering encompass far more than this book attempts to cover.

NY 13440-6916 • 1-888-RAC-USER . Rome. Reliability Analysis Center (RAC) • 201 Mill Street.84 Practical Statistical Tools for the Reliability Engineer STAT This page intentionally left blank.

NY 13440-6916 • 1-888-RAC-USER .STAT Appendix A: Poisson Probabilities 85 Appendix A: Poisson Probabilities Reliability Analysis Center (RAC) • 201 Mill Street. Rome.

Rome. NY 13440-6916 • 1-888-RAC-USER . Reliability Analysis Center (RAC) • 201 Mill Street.86 Practical Statistical Tools for the Reliability Engineer STAT This page intentionally left blank.

Rome.3595 0.2 0.0011 0.4966 0.0988 0.7 0.5 0.7408 0.0000 0.2222 0.0006 0.4 0.5488 0.0126 0.8187 0.3 0.2681 0.0072 0.0002 0.0111 0.0000 0.0002 0.0077 0.0003 Reliability Analysis Center (RAC) • 201 Mill Street.0012 0.0020 0.0033 0.0284 0.1217 0.0536 0.STAT Appendix A: Poisson Probabilities 87 Appendix A: Poisson Probabilities The probability of "x" events occurring.1637 0.0000 0.0001 0.3293 0.4493 0.0050 0.0758 0.4066 0.9048 0. Values of "x" 0 1 2 3 4 5 6 Values of "a" 0.9 0.1647 0.0333 0.0030 0.6703 0.0045 0.0003 0.6065 0.1 0.0007 0.0198 0.0002 0.0000 0.0164 0.0905 0. NY 13440-6916 • 1-888-RAC-USER .3033 0.0001 0.0383 0.3659 0.0494 0.0000 0.0016 0. when "a" are expected.0000 0.0000 0.3476 0.0000 0.0000 0.8 0.0004 0.0000 0.1438 0.6 0.0001 0.

0000 0.0000 0.0054 0.0 5.0663 0.0013 0.0254 0.0829 0.0012 0.0595 0.1755 0.3679 0.0000 15.0120 0.2707 0.0029 0.0000 0.0418 0. Rome.0960 0.0053 0.0020 0.0048 0.0000 0.0058 0.0076 0.0001 0.0217 0.0337 0.0001 0.0760 0.0016 0.0844 0.0181 0.0001 0.0027 0.0271 0.0005 0.0324 0.0557 0.0019 0.2707 0.0298 0.0361 0.0029 0.0009 0.0002 0.0000 2.0387 0.0001 0.0005 0.0001 0.0000 0.0948 0.0516 0.0001 0.0363 0.0299 0.0002 0.0034 0.0083 0.2240 0.0000 0.0106 0.0000 0.0181 0.1462 0.0001 0.1563 0.0001 0.0000 0.0204 0.0001 0.0002 0.0000 0.0071 0.1954 0.0846 0.0 0.0 0.0446 0.0000 Values of "a" 4.0557 0.0486 0.0005 0.0842 0.0216 0.0669 0.1008 0.0631 0.0498 0.0153 0.0003 0.0004 0.0000 10.0002 0.0733 0.0 0.1404 0.1044 0.1954 0.0002 0.0023 0.1839 0.0104 0.1353 0.0000 20.0128 0.1137 0.0000 0.0 0.0004 0.0956 0.0194 0.0000 0.0347 0.1024 0.88 Practical Statistical Tools for the Reliability Engineer STAT Values of "x" 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 1.1126 0. NY 13440-6916 • 1-888-RAC-USER .0000 0.0000 0.1465 0.0083 0.0901 0.0 0.0000 0.0000 0.0081 0.0000 3.0888 0.0013 0.0902 0.0125 0.0378 0.0009 0.1042 0.1251 0.0001 0.0646 0.0847 0.0000 0.1680 0.0001 0.0034 0.0009 0.0008 0.0189 0.0176 0.0001 Reliability Analysis Center (RAC) • 201 Mill Street.3679 0.1755 0.0082 0.0729 0.0521 0.0613 0.0504 0.0183 0.0037 0.0002 0.0034 0.0006 0.0005 0.0000 0.1494 0.0007 0.0006 0.0 0.0003 0.0067 0.0019 0.1804 0.0000 0.0031 0.0000 0.0706 0.0050 0.0132 0.0000 0.0769 0.1251 0.0133 0.2240 0.0000 0.0019 0.0 0.0888 0.1024 0.0002 0.0004 0.0343 0.0653 0.

STAT Appendix B: Cumulative Poisson Probabilities 89 Appendix B: Cumulative Poisson Probabilities Reliability Analysis Center (RAC) • 201 Mill Street. NY 13440-6916 • 1-888-RAC-USER . Rome.

90 Practical Statistical Tools for the Reliability Engineer STAT This page intentionally left blank. Rome. Reliability Analysis Center (RAC) • 201 Mill Street. NY 13440-6916 • 1-888-RAC-USER .

0000 1.0000 1.6703 0.6 0.9963 0.4066 0.9999 1.5 0.9988 0.0000 0.8 0.0000 1.9 0.9384 0.9967 0.9824 0.9769 0.9098 0.9630 0.9953 0.9920 0.8088 0.9982 0.0000 Reliability Analysis Center (RAC) • 201 Mill Street.3 0.0000 1.9999 1.1 0.7 0. Values of "x" 0 1 2 3 4 5 6 Values of "a" 0.0000 0.0000 0.9993 0.0000 1.9372 0.4493 0.9943 0.9998 1.8187 0.9999 0.5488 0.0000 1.9986 0.0000 0.9856 0.9990 1.9998 0.STAT Appendix B: Cumulative Poisson Probabilities 91 Appendix B: Cumulative Poisson Probabilities The probability of "x or less" events occurring.4966 0.8781 0.9084 0.7725 0.9977 0.9659 0.0000 1.9909 0.9997 1.0000 0.0000 1. Rome.9526 0.9966 1.0000 1.0000 0.7408 0.6065 0.0000 1. NY 13440-6916 • 1-888-RAC-USER .9992 0.0000 1.9998 1.9866 0.4 0. when "a" are expected.0000 1.2 0.8442 0.

9834 0.6767 0.000 15.9929 0.0003 0.9991 0.9999 1.5590 0.9880 0.0 0.7205 0.9992 1.4580 0.0000 1.0075 0.9994 0.000 2.9730 0.0000 0.9863 0.9997 0.1183 0.8894 0.9997 0.9997 0.9994 0.9998 0.7486 0.0373 0.5831 0.2381 0.9994 0.4335 0.8571 0.0000 3.0 0.9160 0.1992 0.9319 0.9951 0.000 1.9979 0.92 Practical Statistical Tools for the Reliability Engineer STAT Values of "x" 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 1.7358 0.1246 0.0000 0.000 10.7852 0.0027 0.6160 0.0000 0.0001 0.9998 0.3329 0.9997 0.000 1.9803 0.0661 0.4702 0.9474 0.000 1. NY 13440-6916 • 1-888-RAC-USER .9197 0.0000 1.2210 0.0008 0.000 0.0104 0.9917 0.9994 0.0002 0.9655 0.2203 0.2970 0.9167 0.0000 20.9996 0.9997 1.0 0.9780 0.9473 0.0000 0.2675 0.6968 0.8666 0.000 0.8749 0.0183 0.000 Reliability Analysis Center (RAC) • 201 Mill Street.0 5.4405 0.5679 0.0000 0.8645 0.6289 0.0404 0.000 1.0067 0.9999 1.3814 0.6472 0.9787 0.9489 0.0179 0.9810 0.3679 0.0000 0.000 Values of "a" 4.9999 1.0390 0.8877 0.7874 0.0000 1.1353 0.1302 0.4655 0.9990 0.9999 1.9966 0.0000 1.0671 0.9513 0.0008 0.2650 0.0021 0.0005 0.9990 0.8431 0.9985 0.0 0.0000 0.9999 1.9858 0.0498 0.7916 0.6436 0.9166 0.0 0.9981 0.9971 0.6639 0.9999 0.0108 0.1048 0.9954 0.8192 0.0 0.9965 0.1846 0.0 0.0000 0.4232 0.9886 0.9682 0.3631 0.9220 0.0697 0.8152 0.9664 0.0028 0.4060 0.9919 0.7622 0.0214 0.0000 0.9988 0.9988 0.9963 0. Rome.9999 1.9670 0.9863 0.0293 0.9961 0.9983 0.000 1.1564 0.0916 0.0050 0.9466 0.9936 0.9945 0.9972 0.9999 1.

STAT Appendix C: The Standard Normal Distribution 93 Appendix C: The Standard Normal Distribution Reliability Analysis Center (RAC) • 201 Mill Street. NY 13440-6916 • 1-888-RAC-USER . Rome.

Reliability Analysis Center (RAC) • 201 Mill Street.94 Practical Statistical Tools for the Reliability Engineer STAT This page intentionally left blank. Rome. NY 13440-6916 • 1-888-RAC-USER .

5 2.0793 0.9 Area 0 to Z 0.4772 0.6 2.5 0.8 2.4452 0.4987 0.1179 0.0 2.4993 0. The same formula computes the area under the tail of the curve between -Z and minus infinity. The areas in the tails from Z to plus infinity are one (the total area under the curve) minus 0.4554 0.STAT Appendix C: The Standard Normal Distribution 95 Appendix C: The Standard Normal Distribution 0 z Figures are the areas under the curve between 0 (the mean value of Z) and Z.4032 0.4981 0. The area from -Z to Z is simply twice the values shown.1554 0.3 3.3849 0.4997 ∞ 0.6 1.3159 0.2257 0.4953 0.1 0.3 0.4861 0. multiply the figures given in the table (the areas from 0 to Z) by two (yielding the areas from -Z to Z). To find the area in both tails outside of the range -Z to Z.4995 0.2 2.0 3.4893 0.7 0.5 (the area from minus infinity to 0) minus the figures shown in the table (the areas between 0 and Z).4938 0.4332 0.4 1.3 2.8 1.5000 Reliability Analysis Center (RAC) • 201 Mill Street.8 0.9 1.2 3.3 1.1 3.4 0.0 1.4974 0.2 1.1915 0.5 1.4918 0.1 2.4821 0. NY 13440-6916 • 1-888-RAC-USER . Rome.4 Area 0 to Z 0.4990 0.7 1.3413 0.2881 0.2580 0.4641 0.1 1.4613 Z 2.3643 0.6 0.4965 0.9192 0.0398 0.7 2.9 3.2 0.4 2. Note that the same figures apply to the areas from 0 to -Z. and subtract these from one (the total area under the curve). Z 0.

Rome. Reliability Analysis Center (RAC) • 201 Mill Street. NY 13440-6916 • 1-888-RAC-USER .96 Practical Statistical Tools for the Reliability Engineer STAT This page intentionally left blank.

Rome. NY 13440-6916 • 1-888-RAC-USER .STAT Appendix D: The Chi-Square Distribution 97 Appendix D: The Chi-Square Distribution Reliability Analysis Center (RAC) • 201 Mill Street.

Rome. Reliability Analysis Center (RAC) • 201 Mill Street. NY 13440-6916 • 1-888-RAC-USER .98 Practical Statistical Tools for the Reliability Engineer STAT This page intentionally left blank.

114 20.672 10.314 18.953 16.070 3.446 0.70 0.848 17.152 12.624 15.574 17.034 11.352 18.059 3.340 4.088 2.336 14.198 13.833 3.940 20.614 7.000157 0.571 7.594 5.857 14.255 8.565 14.178 5.807 9.336 12.822 4.339 5.588 23.50 0.98 0.196 11.393 8.021 22.336 14.000628 0.897 9.292 19.879 14.337 11.847 16.179 7.455 0.351 0.002 13.409 15.719 26.341 3.440 17.145 1.697 14.091 14.307 11.985 7.182 20.564 2.659 18.660 5.348 1.390 10.0158 0.527 7.357 0.716 15.812 6.338 8.134 1.962 9.915 11.0201 0.95 0.584 1.325 4.195 3.490 4.107 4.792 25.346 1.599 23.865 6.256 15.940 4.364 25.0404 0.926 12.765 5.342 3.352 0.085 12.649 2.567 10.267 9.168 5.380 6.571 4.610 2.306 18.473 18.339 5.211 0.338 6.226 6.647 27.STAT Appendix D: The Chi-Square Distribution 99 Appendix D: The Chi-Square Distribution Degrees of Freedom 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Chi-square value when the area under the curve from the value to infinity is: 0.524 12.591 13.000 4.015 7.821 13.229 5.992 13.204 3.467 10.508 29.611 16.148 10.344 2.872 1.297 0.336 Reliability Analysis Center (RAC) • 201 Mill Street.125 16.493 20.053 3.340 4. NY 13440-6916 • 1-888-RAC-USER .943 23.379 17.790 9.867 24.939 21.575 5.609 4.151 18.633 8.343 2.856 11.240 15.646 2.600 12.577 28.101 21.343 3.260 9.848 15.828 5.865 12.752 1.578 16.851 12.721 14.336 13.733 3.547 10.115 0.237 10.554 0.187 19.768 22.338 7.703 22.005 1.337 10.711 1.239 1.366 0.337 10.167 2.892 7.041 16.042 8.293 13.386 0. Rome.032 2.185 0.634 9.337 12.062 19.90 0.266 19.368 6.99 0.80 0.103 0.713 1.651 13.531 16.0642 0.989 8.148 0.338 14.443 14.338 7.304 7.928 18.445 17.542 10.408 7.532 3.906 9.312 11.558 3.064 1.475 24.671 6.708 19.00393 0.635 2.337 9.578 6.429 0.117 11.261 8.337 8.820 21.424 2.

631 17.587 30.064 7.939 27.307 21.916 41.140 46.311 22.429 32.769 27.618 24.605 5.074 1.210 3 3.693 49.419 48.475 8 9.113 44.054 26.566 44.033 16.000 17 19.885 42.412 31.856 45.688 14 16.507 18.250 40.144 33.963 28 31.741 40.813 33.362 15.196 36.684 16.773 47.151 21.256 43.665 4.100 Practical Statistical Tools for the Reliability Engineer STAT Degrees of Chi-square value when the area under the curve from the value to infinity is: Freedom 0.919 19.067 16.899 14.592 15.01 1 1.642 6.161 23.530 36.601 22.642 27 30.10 0.278 29 32.461 35.02 0.557 46.270 42.296 29.837 11.209 11 12.635 2 2.242 14.472 27.985 19.410 35.289 9.962 50.812 7 8.524 11.511 21.622 18.892 Reliability Analysis Center (RAC) • 201 Mill Street.382 37.086 6 7.064 23.191 20 22.679 21.017 14.995 33.987 18.668 13.566 21 23.388 15.542 26.563 38.011 15.251 7.246 31.991 7.383 9.980 25 28.645 12.781 13.322 19.108 28.168 20.615 32.171 29.869 32.319 32.968 41.858 26.314 26 29.689 23.217 13 15.038 28.259 30.553 33.760 25.578 16 18.236 11.465 23.812 22.706 3.900 27.924 37.172 38.775 25.408 3.656 12.119 16.027 37.007 35.090 9 10.824 9. NY 13440-6916 • 1-888-RAC-USER .409 18 20.725 12 14.231 8.638 24 27.337 45.391 34.172 30.20 0.204 30.805 19 21.687 36.030 13.30 0. Rome.633 32.666 10 11.588 30 33.087 42.275 19.070 13.343 39.222 18.549 21.912 36.675 22.418 20.219 4.141 15 17.415 40.341 4 4.841 5.277 5 6.812 18.558 10.289 23 26.671 36.989 7.139 39.346 34.307 24.685 26.412 6.652 41.803 12.488 11.873 29.026 24.996 28.815 9.795 35.020 37.675 34.301 30.096 29.615 24.659 40.779 9.362 25.878 5.932 22 24.989 28.442 15.05 0.642 2.

STAT Appendix E: The Student t Distribution 101 Appendix E: The Student t Distribution Reliability Analysis Center (RAC) • 201 Mill Street. Rome. NY 13440-6916 • 1-888-RAC-USER .

NY 13440-6916 • 1-888-RAC-USER . Rome.102 Practical Statistical Tools for the Reliability Engineer STAT This page intentionally left blank. Reliability Analysis Center (RAC) • 201 Mill Street.

845 2.055 3.725 2.753 2.310 1.781 4.250 3.440 1.552 1.717 2.833 2.333 1.101 2.688 0.699 2.408 5.708 2.689 0.947 2.363 1.869 5.365 2.688 0.819 2.350 1.975 0.365 1.691 0.706 0. NY 13440-6916 • 1-888-RAC-USER .000 2.816 0.325 1.684 0.528 1.457 1.646 3.725 3.727 0.703 0.106 3.922 3.583 1.160 2.721 2.476 1.373 3.776 3.541 2.073 4.587 4.415 1.761 2.291 Reliability Analysis Center (RAC) • 201 Mill Street.695 0.473 1.303 6.048 2.683 0.707 3.015 2.677 0.169 3.602 1.604 4.STAT Appendix E: The Student t Distribution 103 Appendix E: The Student t Distribution Degrees of Freedom = Sample Size .078 1.571 3.683 0.684 0.745 3.850 3.706 31.690 3.131 2.500 1.687 0.898 2.086 2.262 2.060 2.821 2.714 2.021 2.812 2.508 1.447 3.423 1.690 0.000 0.460 3.883 3.819 3.821 1.692 0.779 2.746 2.718 1.064 2.750 2.683 0.353 3.729 2.660 2.756 2.711 2.492 1.069 2.624 1.323 1.960 2.697 2.679 0.012 2.∞ to t is: 0.686 0.619 31.771 2.95 0.701 2.686 0.782 2.462 1.345 1.90 3.787 2.792 3.861 2.314 1.042 2.201 2.355 3.657 9.056 2.041 4.074 2.706 2.674 0.289 1.807 2.659 3.599 12.318 1.610 6.093 2.740 2.311 1.479 1.120 2.700 0.977 2.681 0.671 2.895 2.341 1.638 1.485 1.896 1.797 2. Rome.306 2.296 1.860 2.707 3.182 4.681 1.703 2.718 0.337 1.878 2.228 2.998 1.319 1.768 3.734 2.356 1.980 2.533 1.358 1.576 0.316 1.179 2.658 1. Value of t when area from .959 5.886 1.132 2.330 1.943 2.685 0.437 4.920 4.145 2.674 3.390 1.995 63.282 0.831 2.764 1.015 3.965 3.467 1.704 2.314 12.765 0.143 1.9995 636.499 3.328 1.318 4.539 1.796 2.321 1.567 1.711 0.697 0.383 1.326 0.771 2.551 3.685 0.684 2.747 2.75 1.763 2.221 4.518 1.694 0.645 1.741 0.925 5.841 4.617 2.315 1.110 2.924 8.650 1.1 Degrees of Freedom 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 40 60 120 Inf.99 6.310 1.313 1.052 2.372 1.965 2.921 2.045 2.080 2.140 4.684 0.397 1.032 3.

Reliability Analysis Center (RAC) • 201 Mill Street. NY 13440-6916 • 1-888-RAC-USER .104 Practical Statistical Tools for the Reliability Engineer STAT This page intentionally left blank. Rome.

Rome.STAT Appendix F: Critical Values of the F Distribution for Tests of Significance 105 Appendix F: Critical Values of the F Distribution for Tests of Significance Reliability Analysis Center (RAC) • 201 Mill Street. NY 13440-6916 • 1-888-RAC-USER .

Reliability Analysis Center (RAC) • 201 Mill Street. Rome.106 Practical Statistical Tools for the Reliability Engineer STAT This page intentionally left blank. NY 13440-6916 • 1-888-RAC-USER .

80 1.32 20 6210 99.80 4.92 2.56 5.16 1.59 2.96 4.0 30.39 6.54 1.54 4.98 2.10 2.05 3.10 2.41 3.80 3.76 1.31 3. 1 161 18.57 40 251 19.00 for 10% risk d. 1 39.84 2 200 19.4 13.31 6.94 4.76 3.94 5.o.23 2.84 2.2 16.33 2.26 5.63 4.51 1.66 5. 63.4 11.2 99.7 9.97 2.34 5.30 inf.20 1.61 2.o.51 2. 254 19.21 1.4 8.74 3.80 3.45 2.17 2.4 26.64 4.94 2.84 1.0 13.94 1.28 2.45 3.28 6. MSE 1 2 3 4 5 10 20 40 inf.39 4.96 5.16 2.38 2.52 2.51 1.55 6.19 3.79 4.0 9.31 5.16 3.02 2.02 3.7 14.3 10.72 4. 6370 99.62 2.85 1.59 inf.2 29.0 9.87 2.30 2.48 3.46 2.24 9.06 3.6 9.46 4.7 12.2 9. 1 4052 98.85 4.39 5.37 3.3 19.49 3.71 2.7 9.5 34.79 6.39 inf.42 40 62.73 2.84 3.69 2.00 3 216 19.25 2.32 2.71 3.1 6.4 9.5 26. MSE 1 2 3 4 5 10 20 40 inf.61 1.99 5. MSE 1 2 3 4 5 10 20 40 inf.19 5.1 5.88 40 6290 99.53 5.69 1.5 8.f.53 5.10 3.10 3.01 8.49 5.37 1.00 5.32 3.23 3.41 2.21 2.1 7.08 2.5 16.11 4.08 Degrees of Freedom for MSB 4 5 10 55.71 6.18 4.37 2.10 7.71 1.32 3.0 10.12 9.47 5.00 for 5% risk d.16 5.5 10. Rome.56 2.7 28.9 8.11 1. NY 13440-6916 • 1-888-RAC-USER .06 1.18 3.00 Reliability Analysis Center (RAC) • 201 Mill Street.71 2 49.13 1.29 4.94 2.61 3 5403 99.5 14.5 9.1 13.30 3 53.91 2.77 2.1 21.83 3.43 4.63 2 5000 99.61 1.4 28.2 9.61 2.f.f.0 8.61 4.60 Degrees of Freedom for MSB 4 5 10 225 230 242 19.35 4.59 5.78 Degrees of Freedom for MSB 4 5 10 5625 5764 6056 99.09 2.35 2.59 5.85 5.00 1.54 4.o.3 99.3 7.60 20 61.44 2.8 18.52 3.44 5.92 3.42 1.23 4.3 9.83 20 248 19.2 60.5 26.2 16.5 9.55 4.2 27.36 2.13 3.99 1.66 1.08 3.05 4.0 15.78 2.5 11.55 4.12 1.8 57.2 19.84 1.STAT Appendix F: Critical Values of the F Distribution for Tests of Significance 107 Appendix F: Critical Values of the F Distribution for Tests of Significance for 1% risk d.84 2.5 9.29 9.5 8.79 1.38 1.

108 Practical Statistical Tools for the Reliability Engineer STAT This page intentionally left blank. Reliability Analysis Center (RAC) • 201 Mill Street. NY 13440-6916 • 1-888-RAC-USER . Rome.

Sign up to vote on this title

UsefulNot useful- Rome Laboratory Reliability Engineer’s Toolkit, April 1993by goldpanr8222
- Little, R.E. - Mechanical Reliability Improvement - Probability and Statistics for Experimental Testing [Marcel Dekker 2001]by htosun_1
- Introduction to Engineering Reliability engineering Reliability Conceptsby haithamelramlawi7503
- Life Cycle Reliability Engineeringby Wagner De Souza Lima

- Rome Laboratory Reliability Engineer’s Toolkit, April 1993
- Little, R.E. - Mechanical Reliability Improvement - Probability and Statistics for Experimental Testing [Marcel Dekker 2001]
- Introduction to Engineering Reliability engineering Reliability Concepts
- Life Cycle Reliability Engineering
- 30828739 Reliability Engineering
- Introduction to Weibull Analysis Ver4
- Abernethy - The New Weibull Handbook
- Vejbul knjiga
- Dimitri Kececioglu - Reliability Engineering Handbook, Vol. 1
- Mathematical Models for Systems Reliability~Tqw~_darksiderg
- Accelerated_Testing_and_Validation.pdf
- advances in reliability and quality.pdf
- Engineering Reliability
- Reliability for Begineers
- Reliability 3
- Reliability Engineering
- Applied Life Data Analysis
- Modern Statistical and Mathematical Methods in Reliability (
- Applied Reliability, Third Edition - Paul A. & Tobias.pdf
- probability reliability
- r2A Risk and Reliability 5th_Edition
- Life Cycle Costing and Reliability Engg
- Statistics for Experimenters
- book on bussiness statisitics!!!
- Practical System Reliability
- Introduction to Reliability Engineering 2nd Ed - e e Lewis
- Life Cycle Cost and Reliability for Process Equipment
- Software Reliability Engineering More Reliable Software Faster and Cheaper 2nd Edition
- RealScan-10_MTBF
- Asset Information for FMEA-Based Maintenance
- 24992753 Practical Statistical Tools for the Reliability Engineer