You are on page 1of 94

This standard is for EDUCATIONAL USE ONLY.

Copyright by ASTM Int'l (all rights reserved); Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. No further reproductions au

FftOFERTY

OF

university of Michigan
Liabraries
1017
AftTES 8C1ENTIA VBRITAS

The

Copyright by ASTM Int'l (all rights reserved); Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. No further reproduc

A GUIDE FOR FATIGUE TESTING AND THE STATISTICAL ANALYSIS OF FATIGUE DATA

Prepar'ed by COMMITTEE £-9 ON FATIGUE AMERICAN SOCIETY FOR TESTING AND MATERIALS

1963

Reg. U. S. Pat. Off.

ASTM Special Technical Publication No. 91-A* (Second Edition} Price: $5.00; to Members: $4.00

Copyright by ASTM Int'l (all rights reserved); Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by Published by the This standard is for EDUCATIONAL USE ONLY. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. No further reproduc

AMERICAN SOCIETY FOR TESTING AND MATERIALS 1916 Race Street, Philadelphia 3, Pa.

Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. 1964 . Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. No further reprodu Printed in Baltimore. February.©BY AMERICAN SOCIETY FOR TESTING AND MATERIALS 1963 Library of Congress Catalog Card Number: 63-16331 Copyright by ASTM Int'l (all rights reserved). Md.

Others who assisted in the preparation of the Second Edition were W. R. Torrey. Ruley. J. B. W. Toolin. B. D. Upon the formation of Subcommittee VI on the Statistical Aspects of Fatigue. N. Wilk contributed to the discussions at various conferences. Lankford. A. H. T. W. N. They include: (1) revisions in the definitions (Section II) and their separate publication as ASTM Tentative Definitions E 206. W. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. W. F. R. E. R. B. The coordination of contributions and discussions was done by H. H. Shaffer.FOREWORD The First Edition of this Guide was the composite work of many people who contributed a great deal of time to the discussion and writing of the text under the guidance of Task Group Leader. H. Chairman of Subcommittee VI. Schuette. F. Hyler. 1 Definitions of Terms Relating to Fatigue Testing and the Statistical Analysis of Fatigue Data (E 206). Ransom. respectively.1 (2) an expansion of Section IV on the number of test specimens. In addition to the above. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. F. but also edited and arranged for the printing of the advance copies of the text. W. C. Frankel. P. N. Cummings. B. Murphy. In 1956. R. T. J. Ellis. Kao. As a result of this study. extensive revisions have been made in various sections as printed in this Second Edition. Appreciable contributions to the statistical parts of the Guide were also made by D. H. R. on the use of the Weibull distribution function for fatigue Me. and M. N. S. and E. F. A major portion of the statistical section was written by Miss Mary N. Dodge. (3) changes in Section V on tests of significance. Stulen became Leader of the Task Group and the Guide was completed under his direction. and C. B. Ward. Gaver. Torrey. No further reproduction iii . Gohn not only contributed to the discussion and planning. R. and a first rough draft was prepared in 1954 and revised in 1955. E. Part 3. Copyright by ASTM Int'l (all rights reserved). M. Heller. Marco. Miss M. Peterson. Schulte. and (4) the preparation of a new section. Other contributors to these drafts were E. Hooke. McClintock. The original Task Group was organized under the leadership of J. Stulen. Peterson. Lankford. B. 1962 Supplement to Book of ASTM Standards. H. This work was carried out by four Task Groups headed by S. A. H. George R. R. Stulen. P. Moyer. E. Cummings. and G. this subcommittee was asked to review the First Edition and to make any revisions necessary to bring the Guide up to date. Gohn. Findley. T. Appendix IV. F. A. K. E.

Copyright by ASTM Int'l (all rights reserved). Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.—The Society is not responsible. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. No furthe . for the statements and opinions advanced in this publication. as a body.NOTE.

Step Tests of 42 Specimens 50 This standard is for EDUCATIONAL USE ONLY.—Fatigue Test Data 31 14.—R. V.—Confidence Intervals for Percentages 28 12. s 36 18.—Computations for Significance Tests 46 21.—Median Percentage of Survivors for the Population 24 9.—Fatigue Test Data 29 13.—Minimum Number of Specimens Needed for Determining 95 Per Cent Confidence Intervals of Stated Width for a Population Standard Deviation.—Minimum Number of Specimens Needed to Detect a Stated Difference Between the Means of Two Populations 21 8.—-"Probit" Test Data 33 16. a 19 4.—Minimum Number of Specimens Needed for Determining 95 Per Cent Confidence Intervals of Stated Width for a Population Mean.—Computations for Fitting a Response Curve by Method of Least Squares 35 17.—Minimum Number of Specimens Needed to Detect a Stated Difference Between a Mean and a Fixed Value 21 7. No further Values 38 reproduct 20.—Prot Test Computations 52 v TABLE PAGE . Tue Jul 9 10:11:02 EDT 2013 Values 37 Downloaded/printed by 95 Per Cent Confidence Limits for Fatigue Strength 19. III.—Minimum Number of Specimens Needed in Each Sample to Detect if a Standard Deviation of One Population Is a Stated Multiple of the Standard Deviation of Another Population 20 6.—Computation of Standard Deviation. 22. and Abbreviations Test Procedures Minimum Number of Test Specimens and Their Selection Analysis of Fatigue Data Appendices Miscellaneous Reference Tables Additional Technique for Distribution Shape Not Assumed Analysis of Correlation Between Two Variables The Weibull Distribution Function for Fatigue Life References Index PAGE 1 2 8 16 22 55 68 69 71 78 81 LIST OF T A B L E S 1.—Minimum Number of Specimens Needed to Detect if the Standard "Deviation of a Population Is a Stated Percentage of a Fixed Value 20 5.—Method of Computing Cent Confidence Limits for Per Cent Survival Copyright by ASTM Int'l95 (allPer rights reserved). Purposes of Fatigue Testing Definitions. R. Moore Rotating Beam. II.—Confidence Intervals for the Median 26 10.CONTENTS I. Symbols.—Method of Computing Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.—Percentages Surviving 108 Cycles 32 15.—Allocation of Test Specimens for "Probit" Method of Test 11 2.—Approximate Confidence Intervals for the Mean 27 11.—Analysis of Data in Table 21 51 23. IV. p 19 3.

—Typical Fatigue Test Data.—Per Cent of Specimens Having at Least the Indicated Fatigue Strength at 107 Cycles 9. No further reproductions authorize .—Probability-Stress-Cycle (P-S-N) Curve for Phosphor-Bronze Strip 2. 26.—Graphical Illustration of Prot Data 6.—Prot Test: Stress as Linear Function of Stress Cycles 10. with Runouts 53 56 58 60 61 62 63 64 65 67 69 72 74 75 75 77 LIST OF F I G U R E S FIGURE 1.—Per Cent Failed at Weibull Mean 16. 15.—Prot Test Computations 25.—Typical Fatigue Test Data.—Mo.—Estimation of Weibull Distribution Function Parameters for Data in Table 38.—Unpaired Rank Test 27.—Mean-Rank Estimates of the Per Cent Population Failed Corresponding to Failure Order in Sample 37. 10 11 12 14 15 22 34 51 53 54 70 72 73 75 76 77 RELATED ASTM PUBLICATIONS Abstracts of Articles on Fatigue (STP 9) Fatigue Manual (STP 91) (1949) Statistical Aspects of Fatigue (STP 121) (1951) Fatigue.yis for Runs Among Elements in Samples of Sizes Ni and Nz 32.—Working Significance Levels for Quadrant Sum 35.vi TABLE CONTENTS PAGE 24. Without Runouts 39..—k Factors for S-N Curves (Normal Distribution Assumed) 34.—Percentiles of the x2/d-f. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.—Response Curves for a Particular Type of Steel 8.—F Distribution 33.25 and uo.—Ordinate Locations Corresponding to Per Cent Failed Values 36. Tue Jul 9 10:11:02 EDT 2013 Acoustical Fatigue (STPInt'l 284) Fatigue of Aircraft Structures (STP 338) (1963) Downloaded/printed by This standard is for EDUCATIONAL USE ONLY.—Illustration of Staircase Method 4.—Representation of "Step" Testing of Single Specimen 5.—Typical Fatigue Test Data 38.—Minimum Per Cent of Population Exceeding Median of Low Ranking Points.Distribution 31.—Percentiles of the x2 Distribution 28.—"Normal" or Gaussian Distribution Curve 7..—Scatter Diagram 12.—Log-Log Plot of Prot Data 11.—Estimation of Weibull Distribution Function Parameters for Data in Table 39.—Areas of the "Normal" Curve 29.—Values of t 30.—Typical Weibull Distribution Curves 13.—Construction of Weibull Probability Paper from Log-Log Paper 14. with Emphasis on Statistical Approach (STP 137) (1952) Papers on Metals (STP 196) (1956) Fatigue of Aircraft Structures (STP 203) (1956) Large Fatigue Testing Machines and Their Results (STP 216) (1957) Basic Mechanisms of Fatigue (STP 237) (1958) Fatigue of Aircraft Structures (STP 274) (1959) Copyright by ASTM (all (1960) rights reserved).—Response or Survival Tests 3.

3. In order to specify the reliability of these estimates. 2. This standard is for EDUCATIONAL USE ONLY.. deflection-) amplitude and cycle life-to-failure for a given material or component. a number of new techniques have been developed for evaluating the fatigue properties of materials. 1949. A group of test specimens at each stress level. Modified staircase method. Soc. C. 1964 GUIDE FOR FATIGUE TESTING AND STATISTICAL ANALYSIS About 15 years ago. ASTM STP 91.(load-.STP91A-EB/Feb. Increasing amplitude tests. 1. the application of statistical methods to the analysis of the test results of samples offers a means for estimating the characteristics of the population from which the samples were taken. described the principal types of testing machines then in use.astm. "Probit" method. The principal acceptable procedures discussed in this guide are: A. Furthermore. obtained by testing a sample of fatigue specimens in accordance with one of the previCopyright by ASTM Int'l(2) (all to rights reserved). this guide has been prepared. presented detailed instructions for the preparation of test specimens. Staircase method. they must be based on the results of testing a sample of fatigue specimens which have been drawn at random from a population of possible fatigue specimens and tested in accordance with acceptable testing procedures. The term "standard" test. Step method.1 That Manual attempted to standardize the symbols and nomenclature used in fatigue testing. PURPOSES OF FATIGUE TESTING The purposes of fatigue testing are (1) to estimate the relationship between stress. Am. outlined test procedures and techniques. Prot method. 2. ASTM Committee E-9 on Fatigue prepared a Manual on Fatigue Testing. The primary purposes of the statistical analysis of fatigue data are: (1) to estimate certain fatigue properties of a material or a component (together with measures of the reliability) from a given set of fatigue data. "Standard" tests (constant amplitude or classical Wohler method). 1. 1 Copyright© 1964 by ASTM International www. Response tests (constant amplitude). Testing Mats. as used here. strain-. and (2) to compare the fatigue properties of two or more materials or components.org . Statistical theory provides information (a) reproduc 1 2 Fatigue Manual. Single test specimen at each stress level.2 1. does not imply an ASTM standard. Since the Manual was first prepared. and gave some suggestions for the presentation and interpretation of fatigue data. 2. Tue Jul 9 procedures 10:11:02 EDTfor 2013 ous test procedures. To take cognizance of these developments. on No further from similar populations. and provide objective comparing Downloaded/printed by two or more sets of fatigue data to determine whether or not the data come Escuela Politcnica del Ejercito en Ecuador ESPEalso pursuant to License Agreement. I. B.

pursuant to License Agreement. This standard is for EDUCATIONAL USE ONLY.—The term fatigue in the materials testing field. Part 3. Even with some basic training. NOTE 1.—The process of progressive localized permanent structural change occurring in a material subjected to conditions which produce fluctuating stresses and strains at some point or points and which may culminate in cracks or complete fracture after a sufficient number of fluctuations (Note 2). Tue Jul 9 10:11:02 EDT 2013 ..4 1. 3. II." 2. without taking into account the effect on the stress produced by geometric discontinuities such as holes. 4 Escuela Politcnica delAm.ESPE 1948.3—1948. hi general. N. DEFINITIONS. Test procedures are discussed hi Section III while techniques for analyzing the data obtained in these tests are given in Section V and the Appendices. has—in at least one case glass technology—been used for static tests of considerable duration. apply to those cases where the conditions imposed upon a specimen result or are assumed to result in uniaxial principal stresses or strains which fluctuate in magnitude. 1962 Supplement Downloaded/printed by to 1961 Book of ASTM Standards (E 6). sequential loading.3 The symbols used are. and random loading require more rigorous definitions which are.-^-The smallest segment of the stress-time function which is repeated periodically. Ejercito en Ecuador ASA No. AND ABBREVIATIONS Relating to Fatigue Tests and Test Methods: To encourage uniformity of terminology. the terms dealing primarily with fatigue testing and test methods are also published in ASTM Definitions E 6. beyond the scope of this section. Multiaxial stress. NOTE 2. Definitions of certain statistical terms are included. Fatigue Life.—The stress at a point calculated on the net cross-section by simple elastic theory. 4.—The number of cycles of stress or strain of a specified character that a given specimen sustains before failure of a specified nature occurs. Nominal Stress. theory is left to the references. Stress Cycle. Definitions 3 to 19. The purpose of this guide is to describe some statistical treatments that are suitable for the analysis of fatigue data obtained in any one of the foregoing test methods and to present these statistical treatments in a form useful to the test engineer. No further Copyright by ASTM Int'l (all rights reserved). grooves. etc. inclusive. those recommended in the American Standard Letter Symbols for Mechanics of Solid Bodies. but only enough of the basic concepts of statistics are included to make the methods understandable.2 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA the most efficient use of a limited number of test specimens and (b) the number of test specimens required to give a specified degree of confidence in the test results. SYMBOLS. it is difficult to locate the techniques particularly useful in fatigue testing in the statistical literature.—Fluctuations may occur both in stress and with time (frequency). a type of test generally designated as stress-rupture. as in the case of "random vibration.'fillets. Z10. S. Standards Assn. at present. Fatigue (Note 1). 3 Definitions of Terms Relating to Methods of Mechanical Testing.

Sr. —A plot of stress against the number of cycles to failure. at which 50 per cent of the specimens of a given sample could survive N stress cycles in which Sm = 0. SYMBOLS. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant License Agreement. 6. S-N Diagram. —A hypothetical value of stress for failure at exactly N cycles as determined from an S-N diagram. Sa-—One half the range of stress. Smin. This is also known as the median fatigue strength at N cycles (see definition 47). as well as in others that follow. Tue 10:11:02 EDT 2013 becomes very large. A or R. S/.The diagram indicates the S-N relationship for a specified value of Sm. Mean Stress (or Steady Component of Stress). or R and a specified probability of survival. NOTE. The value of SN thus determined is subject to the same conditions as those which apply to the S-N diagram. Stress Cycles Endured. but a log scale is sometimes used. —The limiting value ofJul the9 median fatigue strength as N by ASTM Int'l (all rights reserved).—The number of cycles of a specified character (that produce fluctuating stress and strain) which a specimen has endured at any time in its stress history. that is. that is 9. -Smt^ror Sa .->-Certain materials and" environments preclude the to attainment of a fatigue . A. that is. N. The stress can be Smax. 14. For S a linear scale is used most often. tensile stress being considered positive and compressive stress negative. For N a log scale is almost always used. Stress Amplitude (or Variable Component of Stress). Range of Stress.—The stress having the highest algebraic value in the stress cycle. —The stress having the lowest algebraic value in the cycle. tensile stress being considered positive and compressive stress negative. and the ratio of the minimum stress to the maximum stress. 12. —The value of SN which is commonly found in the literature is the hypothetical value of Smai ./ Minimum Stress. Stress Ratio. 11. Two commonly used stress ratios are: The ratio of the stress amplitude to the mean stress. 8. further reproduct NpTE. SN. Fatigue Strength at TV Cycles. Sm. Nolimit. Copyright Fatigue Limit.n. the nominal stress is used most commonly. AND ABBREVIATIONS 3 5/ Maximum Stress. In this definition. that is. ?. or Sa.DEFINITIONS. 13. Sm. —The algebraic average of the maximum and minimum stresses in one cycle.—The algebraic difference between the maximum and minimum stresses in one cycle. Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. that is 10. —The algebraic ratio of two specified stress values in a stress cycle. Smax.

of a specified character to the hypothetical fatigue life. 9 10:11:02 EDT 25. each of which represents a different stress ratio. Parameter. g. to the corresponding nominal stress. obtained from the S-N diagram. Fatigue Notch Sensitivity. 23. C—The ratio of the number of stress cycles. —The method of selecting the sample determines the population about which statistical inference or generalization can be made. 24. Estimation. N. A ASTM summary calculated fromTue theJul observed values in a2013 Downloaded/printed by sample. Escuela Politcnica del Ejercito Ecuadora ESPE pursuant to License Agreement.—The specimens tested at one time. n. Smax and/or Smin to the mean stress Sm. NOTE. —The ratio of the fatigue strength of a specimen with no stress concentration to the fatigue strength at the same number of cycles with stress concentration for the same conditions.—A measure of the degree of agreement between Kf and Kt for a particular specimen of a given size and material containing a stress concentrator of a given size and shape. K/. Group. —In specifying Kj it is necessary to specify the geometry and the values of Smax .—The specimens selected from the population for test purposes. The constant life fatigue diagram is generally derived from a family of S-N curves. and N for which it is computed. for a 50 per cent probability of survival. NOTE. such as a population median or a population standard deviation. 21.—A constant (usually unknown) denning some property of the frequency distribution of a population. each of which is for a single fatigue life. 22. Constant Life Fatigue Diagram. K 19. Sm . NOTE. 15. Relating to Statistical Analysis: 20. —A procedure forenmaking statistical inference about the .—The hypothetical collection of all possible test specimens that could be prepared in the specified way from the material under consideration. Kt-—The ratio of the greatest stress in the region of a notch or other stress concentrator. A group may comprise one or more specimens. Fatigue Notch Factor. No f 26. —A common definition of fatigue notch sensitivity is q = (Kf — i)/(Kt — 1\ in which q may vary between zero (where Kf = 1) and one (where Kf = t). Copyright — by Int'lvalue (all rights reserved). This standard is for EDUCATIONAL USE ONLY. as determined by the theory of elasticity (or by experimental procedures that give equivalent values). —The theory of plasticity should not be used to determine Kt . at one stress level. Theoretical Stress Concentration Factor (or Stress Concentration Factor). or consecutively.—A plot (usually on rectangular coordinates) of a family of curves. NOTE. Statistic. Cycle Ratio.4 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA Values tabulated as fatigue limitsun the literature are frequently (but not always) values 6f SN for 50~per cent survival at N cycles of stress in which Sm = 0. for stress cycles of the same character.—The way in which the frequencies of occurrence of members of a population or sample are distributed according to the values of the variable under consideration. Sample. Frequency Distribution. 17. 18. N. Population (or Universe). that is. relating Sa. A or R.

No further re .—This value of 5 provides a statistic that is used in computing interval estimates and several test statistics (see definitions 34 and 42). 28. Am.—The sum of the squares of the differences between each observed value and the sample average divided by the sample size minus one. or values. Sample Standard Deviation. 30. Tolerance Level. Escuela Politcnica del Soc. It is a point estimate of the population mean. in a stated proportion of the times such statements are made.—The two statistics that define a tolerance interval.—The percentage of observed values between two stated values of the variable under consideration. Estimate." but this statistic is not as useful. An effect or difference between popu6 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. Some texts define s2 as "the sum of the squares of the differences between each observed value and the sample average divided by the sample size.—The sum of all the observed values in a sample divided by the sample size. SYMBOLS. Sample Median.DEFINITIONS.—The estimate of a parameter given by a single statistic. Confidence Level (or Coefficient). For small sample sizes. it is the average of the two middlemost values. Tolerance Interval. ESPE 1957. s. (One value may be — « or + °°. AND ABBREVIATIONS 5 numerical values of one or more unknown population parameters from the observed values in a sample. Ejercito en Ecuador ASTM STP 15-C. pursuant to License Agreement.—The two statistics that define a confidence interval. It is a point estimate of the population median. Testing Mats. Point Estimate.—The estimate of a parameter given by two statistics. Tue Jul 9 10:11:02 EDT 2013 41. Sample Variance. It is a point estimate of the percentage of the population between the same two stated values. 27. Interval Estimate. 37.—Statistically significant.—This value of s* provides both an unbiased point estimate of the population variance and a statistic that is used in computing interval estimates and several test statistics (see definitions 34 and 42). 36. It is a point estimate of the population standard deviation. NOTE. (See the ASTM Manual on Quality Control of Materials5 or texts on statistics for an unbiased estimate of the standard deviation of a Normal population.—An interval computed so that it will include at least a stated percentage of the population with a stated probability." will be true. "the population parameter lies in this interval.—-The middle value when all observed values in a sample are arranged in order of magnitude if an odd number of specimens are tested.. Significant.—The stated proportion of the times the confidence interval is expected to include the population parameter.) 34. a measure of the "spread" of the frequency distribution of a population. on the average.—The particular value. of a parameter computed by an estimation procedure for a given sample. Sample Average (Arithmetic Mean). or 50 per cent point. denning the end points of an interval. (One stated value may be — °° or + °°.) 40. Confidence Interval. It is a point estimate of the population variance. It is not the same as a confidence level but the term confidence level is frequently associated with tolerance intervals. s2. 35.—The square root of the sample variance. — 32. Sample Percentage. Tolerance Limits.—An interval estimate of a population parameter computed so that the statement. Copyright by ASTM Int'l (all rights reserved). 29. If the sample size is even. 38.—The stated probability that the tolerance interval includes at least the stated percentage of the population.) 33. 31. 39. j underestimates the population standard deviation. Confidence Limits. NOTE.

6 FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA lations is said to be present if the value of a test-statistic is significant. Fatigue life for p per cent survival values. 90. the average). such as 95. p may be any number. may also be estimated from the individual fatigue life values. 43.—The use of the sample median. that is. when applied to a collection of data without further qualification the term "fatigue life" is ambiguous.—The stated probability (risk) that a given test of significance will reject the hypothesis that a specified effect is absent when the hypothesis is true. S-N Curve for 50 Per Cent Survival. 47.—An effect which is statistically significant may or may not have engineering significance.—An estimate of the stress level at which 50 per cent of the population would survive N cycles.—This is a special case of the more general definition 48. NOTE.—A curve fitted to the median values of Copyright ASTM Int'l (all rights reserved). lies outside of predetermined limits. 42. Escuelawould Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. etc. 90. Fatigue Strength for p Per Cent Survival at N Cycles. since there is no test procedure by which a frequency distribution of fatigue strengths at N cycles can be directly observed.—A function of the observed values in a sample that is used in a test of significance. NOTE. NOTE 1. 9 estimate 10:11:02 EDT fatigue life at by each of several stress levels. such as 95. The observed value of the median fatigue life estimates the fatigue life for 50 per cent survival.. Fatigue Life for p Per Cent Survival.—A test which. 46. by use of a test-statistic. NOTE 2. etc. NOTE 2. However. etc. Median Fatigue Life.—An estimate of the fatigue life that p per cent of the population would attain or exceed at a given stress level. 90. 44.—The middlemost of the observed fatigue life values.—The limiting value of fatigue strength for p per cent survival as N becomes very large. . definition 14. NOTE 1. instead of the arithmetic mean (that is. purports to provide a test of the hypothesis that the effect is absent.—The rejection of the hypothesis indicates that the effect is present. 49. where p is any number.—The estimates of the fatigue strengths for p per cent survival values are derived from particular points of the fatigue life distribution. No further reproductions authorize population survive.—In the literature. the abbreviated term "fatigue life" usually has meant the median fatigue life df the group. since there is no test procedure by which a frequency distribution of fatigue strengths at N cycles can be directly observed. Test-Statistic. NOTE. Fatigue Limit for^ Per Cent Survival. such as 95. (See Note. arranged in order of magnitude. p may be any number. Significance Level. Median Fatigue Strength at N Cycles.—The estimate of the median fatigue strength is derived from a particular point of the fatigue life distribution.—An estimate of the stress level at which p per cent of the population would survive N cycles.Tue It Jul is an of 2013 the relationship between Downloaded/printed applied stress by and the number of cycles-to-failure that 50 per cent of the This standard is for EDUCATIONAL USE ONLY. 48. is usually preferred. Relating to Statistical Analysis of Fatigue Data: 45. In the case where an even number of specimens are tested it is the average of the two middlemost values. of the individual specimens in a group tested under identical conditions. Test of Significance.) 50.

it should be used in technical papers only when adequately described.-lb Unit of work / Polar moment of inertia Int'l (all rights of reserved). NOTE. for mathematical analysis the use of Greek symbols is generally preferred. It is an estimate of the relationship between applied stress and the percentage of the population that would survive N cycles. AND ABBREVIATIONS 7 NOTE 1. Symbol Term A Area of cross-section. 107. No further reproductions authorized. NOTE 2.—This is a special case of the more general definition 51. Kf Fatigue notch factor . In general. For stress. 52.—Caution should be used in drawing conclusions from extrapolated portions of the S-N curves. 90. such as 95. the S-N curves should not be extrapolated beyond observed life values. p may be any number. Stress ratio C Cycle ratio c Distance from centroid to outermost fiber D or d Diameter E Modulus of elasticity in tension or compression e (epsilon) Strain f t-lb Unit of work G Modulus of elasticity in shear 7 (gamma) Shear strain / Moment of inertia i Subscript denoting ith term in.DEFINITIONS. Response Curve for N Cycles. S-N Curve for p Per Cent Survival. etc. LIST OF SYMBOLS AND ABBREVIATIONS The following terms are frequently used in lieu of or along with the terms covered by the preceding definitions. the curves should not be extrapolated to other values of p.—Caution should be used in drawing conclusions from extrapolated portions of the response curves. SYMBOLS. 51.—Values of the median fatigue strength at N cycles and the fatigue strength for p per cent survival at N cycles may be derived from the response curve for N cycles. It is an estimate of the relationship between applied stress and the number of cycles-to-failure that p per cent of the population would survive.—In the literature.—A curve fitted to the fatigue life for p per cent survival values at each of several stress levels.—A curve fitted to observed values of percentage survival at N cycles for several stress levels. NOTE 1. NOTE 2. the use of S with appropriate lower case subscripts is preferred for general purposes. In general the symbols are those recommended in the American Standard Letter Symbols for Mechanics of Solid Bodies (see footnote 4). Tue Jul 9 10:11:02 EDT 2013 ksi Copyright by ASTM Thousands pounds per square inch Downloaded/printedor by kips per square inch This standard is for EDUCATIONAL USE ONLY. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. the abbreviated term "S-N Curve" usually has meant either the S-N curve drawn through the means (averages) or the medians (50 per cent values) for the fatigue life values. In general. Since the term "S-N Curve" is ambiguous. if p falls within the range of the per cent survival values actually observed. etc. where N is a preassigned number such as 106.

TEST PROCEDURES Copyright by ASTM Int'l (all rights reserved). Stress . there was only oneThis accepted method of conducting laboraDownloaded/printed by standard is for EDUCATIONAL USE ONLY. tory fatigue tests on aEjercito material or component. or v (nu) (generally used in applied mechanics) N n P psi P III. M n (mu) preferred.A (7 Estimate of standard deviation (T Variance Standard deviation of x <rx Variance of x <rx2 T Torque. Tue Jul 9 10:11:02 EDT 2013 Until recently. is described in Section V of the . This "standard" test. Minimum stress LJrmn SN Fatigue strength at N cycles Sr Range of stress Ss or T (tau) Shear stress Tensile stress St Stv Tensile yield strength Tensile strength Su (7 (sigma) Standard deviation. Temperature t Time T (tau) or Ss Shear stress 2 Symbol Kt Ib-ft Ib-in.8 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA Term Theoretical stress concentration factor Unit of torque Unit of torque Bending moment Poisson's ratio Fatigue life. Sample size Load Pounds per square inch Probability of failure. number of cycles JNumber of stress cycles endured. Probability of survival. Per cent failure. Per cent survival Fatigue notch sensitivity 9 Stress ratio R s Sample standard deviation s2 Sample variance S Nominal stress S or <r (sigma) Normal stress Stress amplitude Sa Compressive stress Sc Fatigue limit Sff Compressive yield strength »Jcj/ Mean stress »SJ>» Maximum stress S^ max C . No further reproduct single specimens at several stress levels. using Escuela Politcnica del en Ecuador ESPE pursuant to License Agreement.

each fatigue specimen is cycled at a different constant stress (or strain) amplitude until fracture occurs. some specimens must be run at stress levels high enough to produce failures in order to have data from which the fatigue limit may be estimated. the investigator usually has some preconception of this value for the material or component to be tested. The latter methods also are used for comparing the long-life fatigue properties of different materials or different methods of processing. or assemblies are being tested.x TEST PROCEDURES 9 Manual on Fatigue Testing (STP91). see Section V. A. using only one specimen at each stress level. each group should consist of at least four specimens in order to estimate the variCopyright by ASTM Int'l (all rights reserved). No more different stress levels must be investigated for the determination of the . a number of new methods for performing more meaningful fatigue tests have been introduced. Near the fatigue limit. succeeding specimens are then tested at stress levels either above or below this value. In this procedure. however. The stress levels are usually selected to cover a series of stresses ranging from high values. When the objective is to determine an S-N curve. to low values at which no failure will occur (runouts) or at which failure will occur only after an extremely large number of cycles. Therefore. full size sections. at which failure will occur within a limited number of cycles. In this case. that this test method did not give adequate information for many of the purposes for which fatigue data are needed. each method having certain advantages. or (3) machine parts. Experience showed. Depending upon the results of the first test. Ten or more specimens are preferable to obtain some Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. the "standard" tests (Sections I Al and A2) are generally the most suitable. Three orfurther repr Escuela del Ejercito en distribution Ecuador ESPE of pursuant to License Agreement. B2. Group of Specimens Tested at Each Stress Level: Since the "standard" test. indication asPolitcnica to the shape of the fatigue life values. The choice of test method depends upon the objective of the test and the number of available test specimens. it is more satisfactory to test several specimens at each of a number of different stress levels. All seven of these experimental fatigue testing techniques are described in the following paragraphs. This method of test is used when the investigator has available only a relatively small number of specimens for test. "STANDARD" TESTS (CONSTANT AMPLITUDE) 1. To determine the long-life fatigue strength or the fatigue limit. 2. If the primary interest is in the longlife end of the S-N relationship (often called the fatigue limit). (2) the supply of material is limited. Such is generally the situation when (1) the fatigue specimens are expensive. and B3) or increasing amplitude tests (Sections I Cl and C2) are recommended. within the last ten years. Tue Jul 9 10:11:02 EDT 2013 ability of the data. until a stress level is reached at which the specimen does not fail within the prescribed number of cycles. response tests (Sections I Bl. the first stress level is selected somewhat above the estimated fatigue limit. gives very little information concerning the variability of the material or component and test procedure. For analysis of the data. Single Test Specimen at Each Stress Level: In the "standard" test method described in STP 91.

FIG. The test is not limited to this application. provided that the specimens are properly allocated to the various stress levels. it is just as valuable for estimating the fatigue strength or the fatigue limit at any other percentages of survival. ASTM Int'l (all rights reserved). a number of groups also should be tested at stress levels in the vicinity of the fatigue limit. at least four or five stress levels are used in a test of this nature. A fifth stress level producing approximately 50 per cent survivors is Copyright by essential. Furthermore. . Such curves are sometimes referred to as probability-stress-cycle (P-S-N) curves. When used to estimate the fatigue limit at 50 per cent survival. 1. B. more specimens should be tested in the long-life than in the short-life range. at least two stress levels should be selected so that the percentage of specimens surviving N cycles will be less than 50 and two more stress levels selected at which the percentage of survivors will be more than 50.—Probability-Stress-Cycle (P-S-N) Curve for Phosphor-Bronze Strip.10 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA S-N curves for p per cent survival (see Fig. I). one or more groups of specimens are tested for a fixed number of cycles at four or five different stress levels distributed about the stress of interest. and 50 per cent will survive. "Probit" Method: In the "Probit" method.6 Generally. To determine the fatigue limit of the material. the stress at which 50 per cent of the test specimens will fail prior to. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. to obtain approximately an equal degree of precision throughout the range of the S-N curve. Tue Jul 9 10:11:02 EDT 2013 desirable but not 6 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. that is. RESPONSE TESTS (CONSTANT AMPLITUDE) 1. This test has been used primarily for estimating thefatigue limit of a material. the preassigned cycle life. This would increase the total number required to at least seven groups. N.

In "Probit" tests a group should consist of not less than five specimens and the total tested at all stress levels should be at least 50. FIG. 2.5 2 3 5 a The group size is the number of specimens included in a test at one stress level. ble 1).95 2. As an alternative to the use of the relative group sizes (TaThis standard is for EDUCATIONAL USE ONLY.—ALLOCATION OF TEST SPECIMENS FOR "PROBIT" METHOD OF TEST. 98 1 1. A properly designed "Probit" test will give more useful fatigue data than any of the other response or increasing amplitude tests. whatever group size is chosen for testing at stress levels for which the expected per cent survival is between 25 and 75. Escuela Ejercito en used Ecuador ESPE pursuant to License Agreement. the sizes of other groups must be increased by the factor in the second column to obtain the number of test specimens required for testing at stress levels for which the per cent survival is expected to be larger. Expected Per Cent Survival Relative Group Size" 11 25 to 75 15 to 20. provided that No fu . 80 to 85 10. or smaller. If the stress levels are chosen successively. This allocation is suggested so that the observed percentage survival values will have approximately equal weight. the group size required for the other levels will be determined more easily. starting with levels requiring the smallest group size. Previous data for the same material or similar materials should be used as a guide for choosing the stress levels. a condition necessary for fitting the response curve by the usual method of least squares.90 5. if similar precision is to be obtained in the test results. whenever they are available. groupsPolitcnica of other del sizes can be at each stress level.—Response or Survival Tests. The relative group sizes for different stress levels are shown in Table 1. otherwise a preliminary test such as that described under Sections IIIA1 or B2 may be required. The distribution of the total number of available test specimens will depend upon the purpose of the test. Tue Jul 9 limits 10:11:02 by sponseDownloaded/printed curves.TEST PROCEDURES TABLE 1. This allocation also by facilitates the (all computation of confidence onEDT the 2013 reCopyright ASTM Int'l rights reserved). Thus.

NOTE—Specimens numbered in chronological order. In the staircase method the specimens are tested sequentially. Escuela Politcnica del Ejercito enis Ecuador ESPE pursuant to License Agreement. 2. and so forth. If the first specimen does not fail. The data are recorded as shown in Fig. The chart shows at a glance the stress level that should be used for the next test. the actual number of cycles-to-failure should be recorded for each specimen that fails before 107 cycles. The specimens that did not fail are designated by the o's and those that failed as #'s.—Illustration of Staircase Method. 3. The first specimen is tested at a stress level equal to the estimated value of mean fatigue strength for the prescribed number of cycles or until it fails.7 Figure 2 presents data that might be obtained in a "Probit" test of the type described if the preassigned number of cycles were 107.a variation of the "Probit" method. 7 The boldface numbers in parentheses refer to the list of9 references appended Copyright by ASTM Int'l (all rights reserved). This standard is for EDUCATIONAL USE ONLY. Number of cycles for each test is constant unless failure occurs beforehand. The selection of the proper increment of stress level is very important. in reducing the number of specimens tested may be offset by an increase in the time required to conduct the test. N* The advantage gained FIG. If the specimen fails. The Staircase Method: The staircase (or "up-and-down") method of testing is. one at a time. the next specimen is tested at a stress level that is one increment below the first stress level.12 FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA weighting factors are employed and the analysis conducted as indicated in references (1-3) . It may require fewer specimens than the latter but is likely to be useful only when the primary interest is in the mean fatigue strength corresponding to a preassigned cycle life. the second specimen is tested at a stress level that is one increment above the first stress level. No further re not the median fatigue strength which used elsewhere in this guide. if it fails before that number of cycles. . such as the plotting of P-S-N Curves. Although not required for the "Probit" analysis. 3. 8 Downloaded/printed by The staircase method as described in the literature estimates the mean fatigue strength. so that the data may be available for other types of analysis. Tue Jul 10:11:02 EDT 2013 to this guide.

TEST PROCEDURES 13 Ideally. Further. EscuelaWhen Politcnica del Ejercito Ecuador ESPE pursuant to may License No fu men occurs. so chosen that about 50 per cent of the test specimens survive at the middle stress level. which gives results for a wider range of stress values. response curvesen are not available. the staircase method is not an efficient experimental procedure. When only a few parts are available for determining the fatigue limit. if the main interest lies in estimating the response curve—rather than the mean strength—at N cycles. Since the testing is concentrated at stress levels near the mean fatigue strength value. say 107. In the modified staircase method. a decrease inJul the 9 probability of surDownloaded/printed byand the test is repeated until failure of the specivival of about 5 per cent This standard is for EDUCATIONAL ONLY. the USE test be Agreement. the data may be combined for statistical analysis. 3. Step Method: In many cases the "Probit" or staircase methods of test require more specimens than are available. Previous data for the same or similar materials are needed in order to choose the stress levels efficiently. at most. the number of specimens tested may be less than for the "Probit" method. at least 30 specimens should be tested because. T. If the results are not significantly different. data for tests 1. Such data are discarded up to the first pair of data giving opposite results. some preliminary testing may be required. only half of the test results are actually used in the computation of the mean fatigue strength. In general. If data obtained by the staircase method are analyzed by response curve methods. and 3 should be discarded. so that rn = T. In the modified procedure. about 70 per cent survive at the lower stress level. the applied stress level is inCopyright by ASTM Int'l (all rights Tue 10:11:02 EDT 2013 creased by an amount corresponding toreserved). a natural desire is to test each part until it actually fails instead of just counting the number of runouts. the total number of specimens. 3. This treatment is known as the modified staircase method. Modified Staircase Method: The time required to complete a test by the staircase method can be reduced by dividing the one long staircase program into several shorter. as in any other test in which specimens in a group are tested on more than one machine. 2. is divided into r groups of n each. If none are available. If typical response curves for the material are available. Each group is tested as a separate staircase program. the results may be statistically biased because of the sequential nature of the staircase method. Thus several machines may be used simultaneously. most of the tests should be made at three stress levels. INCREASING AMPLITUDE TESTS 1. independent staircases and conducting these several tests simultaneously. for example. in Fig. a check should be made to determine whether the machines give significantly different results. the test may be started at a stress level corresponding to a percentage survival of approximately 90 per cent. For each successfully completed run. with a separate chart for each group. When testing a limited number of specimens sometimes the practice is to run each specimen at several stress levels for a large number of cycles. started . and about 30 per cent survive at the higher stress level. C.

the effect of coaxing or under-stressing the material must be known.14 FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA at a stress level equal to about 70 per cent of the estimated fatigue limit. FIG. Escuela PolitcnicadelEjercitoenEcuadorESPEpursuant toLicenseAgreement. Before the step technique of fatigue testing can be safely used. 2013 greater precision by in the estimates of the median and the variability of the Downloaded/printed This standard is for EDUCATIONAL USE ONLY. the fatigue strength corresponding to a preassigned value of N for each specimen or part may be estimated. fatigue strength. TueJul 910:11:02 EDTstrength. As a result. will have their fatigue limits artificially raised by coaxing or under-stressing at low stress levels. This method is illustrated graphically in Fig. and the stress increments should be approximately 5 per cent of the estimated fatigue limit. 4. In this manner.Nofurtherreproductionsauthorized." However. Certain steels. For those materials with which neither appreciable coaxing nor damage occurs. Although step tests have been made with a single specimen. sensitive to strain-aging. it is possible to estimate the fatigue strength of each specimen or part by stressing it at consecutively higher levels until the specimen fails. in recent years it has been observed that understressing does not greatly affect the true fatigue limit of some alloys. a number of stress levels of runouts are usually necessary before failure of the specimen occurs.—Representation of "Step" Testing of Single Specimen. whereas the fatigue strengths of other materials may be decreased by damage due to "under-stressing. 4. such as many of the alloy steels and a few of the nonferrous materials (4). this method has not been considered an acceptable technique because the fatigue strengths of some materials will be increased or "coaxed" by stressing them at stress levels below their fatigue limits. The main disadvantage of the procedure is that thespecimens are run initially at a sufficiently low stress level so that failure will not occur. In other cases it is thought that coaxing or under-stressing may damage the material artificially and cause premature failures. four or more are needed to estimate the median fatigue A larger sample gives Copyrightby ASTM Int'l(allrights reserved). . In the past.

it has been found that. that the long-life fatigue strengths of nonferrous alloys can always be evaluated by this method. by the use of the Prot procedure. at small and Escuela Politcnica del Ejercito en Ecuador pursuant Agreement. however. No further rep . In contrast to the step method. arrange Tue for small weights. Copyright by ASTM Int'l way (all rights Jul 9 10:11:02 EDT 2013 to be poured into a container at a constant rate. Another is toreserved). one attoaLicense time. A number of specimens is tested at the same rate of loading until each specimen fails. It is not certain. 5. of the estimate found from constant amplitude methods. «2 . One of the simplest methods for obtaining a constant rate of loading in a fatigue test is to use a stream of water flowing at a constant rate into the loading container. indicate different loading rates in psi per cycle. «s.TEST PROCEDURES 15 2. which is required in the Prot analysis. To date. in France. In the Prot method. The Prot Method (5-8): In 1945 Marcel Prot. 5. as discussed in Section III Cl. At least three rates of loading are used to establish and check the linear relationship between stress and the power of the loading rate. it is suggested that at least 20 test specimens be used to obtain the data needed for the Prot analysis because of the wide scatter in fracture stress usually found in Prot fatigue data. the fatigue limits of many alloy steels may be obtained within a few per cent FIG. By using the Prot method. devised a rapid method for estimating the fatigue limit of a material.—Graphical Illustration of Prot Data. The use of this technique is restricted not only to those materials which are not sensitive to coaxing effects. such as shot. but also to materials that apparently have a fatigue limit. a good estimate of the fatigue limit may be obtained in a fraction of the time required by other methods but at the expense of more uncertainty than is present in most of the other test methods. been obtained by adding small weights byESPE hand. The lowest rate should be as small as practicable and the highest rate should be low enough so that the specimen does not fail by yielding before fracture. «!. Fairly good results have Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. The type of data observed is shown in Fig. the test on a specimen is first started at an alternating stress of about 60 to 70 per cent of its estimated fatigue limit and the stress is raised at a constant rate.

5. more perfect testing machines than those that are tested later when wear of the machines has modified their characteristics. NOTE. pp. if one bar of a batch of bar stock is tested. Tue Jul 9 how 10:11:02 2013 sb_n^1H he rqm'eH r^it by A common misconception is that randomization can_ Downloaded/printed This standard is for EDUCATIONAL USE ONLY. all the points obtained at a given rate should fall on the same straight line. No further be accomplished by del rough-and-ready procedures such as reaching blindly repr . IV. For example. Any device that increases the stress at practically a continuous. Escuela Politcnica Ejercito en Ecuador ESPE pursuant to License Agreement. Onp job of thp and is to decide the EDT randomization Copyright bypnginppr ASTM Int'l (allstatistician rights reserved). for example.—For constant rates of loading. constant rate can be used. Specimens tested at the beginning of a program may have the advantage of being tested on newer. 366-370) the influence of all variability inherent in the material and testing procedures is given a fair chance of being reflected in the test data. By random selection and allocation of the test specimens.16 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA equal cycle increments. Small variations in the rate of loading or variations in the testing speed may cause scatter such as that shown in Fig. it is often tacitly assumed that the remaining bars are the same as the one tested. There are innumerable stages in the testing program in which any one specimen or any one group of specimens may be affected differently from others from the same population. For each batch the furnace settings are slightly different. These are but a few examples of* the many factors that may produce significant biases in the results unless controlled by appropriate randomization. Within each heat-treatment batch. such blanks are heat treated in batches. The following are some of the factors for which randomization might be considered: Position of specimen within the whole batch of material Heat-treatment batch Position of specimen in heat-treating furnace Order of quenching Order of polishing Assignment to testing conditions (stress level and so forth) Order of testing Assignment to testing machine Machine operator This list will suggest other variables of importance in particular programs. those specimens near the walls of the furnace are under slightly different conditions from those in the center. Implicit in these tests is the assumption that the sample tested is "representative" of the population. Specimens prepared at the start of the day are machined with sharper tools than those that succeed them. MINIMUM NUMBER OF TEST SPECIMENS AND THEIR SELECTION The practicability of fatigue tests is based upon the assumption that establishing the fatigue characteristics of a given material by studying the performance of a random sample selected from a larger body of possible specimens (the population or universe) is possible. using a table of random numbers (9. Usually they are not because.

For example. The techniques of experimental design are too involveoL however. see Appendix pp. In most cases the statistician will be able to plan the experiments to measure not only the effects of the main variables under study but also the effects of the more important secondary variables as well. Copyright by ASTM Int'l (all rights reserved). LITE DISTRIBUTION SHAPE NOT ASSUMED 1. ' "Some indication of the minimum number of specimens needed for a given degree of confidence in the results obtained when using the different test procedures has been given in Section III. to be included in this Guide. see references (10-12). one group of five specimens at each stress level is needed for an 87 per cent survival curve corresponding to a 50 per cent confidence level. atrained^tatisti^ cian shoiild_b£_£OJiau]lejJ_whenever possible. Part 3. including 50 per cent. No f For TablesPolitcnica 25 to 33. Tests conducted in accordance with such a plan can be analyzed to give an estimate of the importance of each of the known variables that contribute to the scatter in the test results. and do this without requiring many. A.NUMBER or SPECIMENS AND THEIR SELECTION 17 into the box of specimensjQr^the^iextjto^b^tested. For example. S-N Curves: The minimum number of fatigue test specimens needed at each stress level depends on: (1) which per cent survival curve is desired and (2) what confidence level is desired. 55-67. 10 Escuela del Ejercito I. The best procedureJs_tQ_ae±-iiD the program on the basis of random numbers as previously suggested (9"). For a 50 per cent confidence level and one group tested at each stress level. additional specimens. To obtain fatigue data that can be used most efficiently. 9 The following sections discuss the minimum number of specimens needed for each type of analysis given in Section V when the sample size is fixed before testing. if any. from Table 25. For supplementary references on this subject. Table 8 in Section VA1 shows the number of specimens needed for several values of per cent survival. 1961 Book of ASTM Standards. en Ecuador pursuant to License Agreement. a 95 per cent survival curve requires at least 13 specimens at each stress level. When several S-N curves are to be drawn from the same data. Tue Jul 9 10:11:02 EDT 2013 9 See also Recommended Practice for Choice of Sample Size to Estimate the Average Downloaded/printed by Quality of a Lot or Process (E 122). This standard is for EDUCATIONAL USE ONLY. at least five groups of 10 specimens at each stress level are needed for an approximately equivalent S-N curve. The sample can be "biased" by unconscious and unrecognized trends of human behavior as well as by unknown patterns of arrangement. in jjlanning the experiments and specimen selection. Table 2510 provides similar information for one or more groups tested at each stress level and seven values of confidence level. All samples are assumed to be randomly selected samples from the population under consideration. Table 25 should be studied carefully to find the best combination of number of groups and group size. but it is summing how chance procedures. For a 90 per cent confidence level. ESPE .

the confidence interval for any given confidence level becomes narrower and the difference between the observed value and the universe value becomes smaller. 4. For differences of two or more percentages (other than 50 per cent) no precise estimate of the minimum number of specimens needed is possible unless prior estimates of the percentages are available. N . confidence limits are equal to the observed minimum and maximum values up to a sample size of nine. Tests of Significance: The minimum number of specimens needed depends upon the desired magnitude of the difference that should be detected and the size of the risks that can be tolerated. In general. If a good estimate of p = per cent survival/100 is available. In general. (See Table 9 on page 26. See footnote 15. For the median at a confidence level of 0. Comparing the widths of the confidence intervals gives some idea of the size of sample needed. page 29. or 5. 4. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by A discussion of the minimum number of specimens and their allocation This standard is for EDUCATIONAL USE ONLY. At least five specimens should be included in each group. Table 11 (Section V A2) gives values of 95 per cent confidence limits for four sample sizes. for equations for setting up tables similar to Table 9 (in Section V A2) for other percentage points. as the sample size increases. it is difficult to relate the desired values and the criteria for the significance test given in Table 26 for two groups and in Table 27 for more than two groups.18 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA 2.) If ranges for prior samples from the same population are known. when the width of the confidence interval becomes less than the observed range. When the rank test is used to test the differences of group medians. If the number of specimens are only 3. Escuela Politcnica Ejercito III en Ecuador ESPE pursuant to License Agreement. It is more difficult to determine the minimum number of specimens needed for a confidence interval of a given width for fatigue life corresponding to a stated value of per cent survival other than 50 per cent. to stress levels is given del in Section Bl. At least 15 specimens should be included in each group. the sample sizes would be larger than for medians. Table 10 (see Section V A2) gives procedures for computing confidence intervals for the mean. Response Curves: Copyright by ASTM Int'l (all rights reserved).95. a sample size can be chosen so that the interval will have approximately the desired width. the minimum sample size is approximately: where E = one half the desired width of a 95 per cent confidence interval (see ASTM Recommended Practice E 122)9. 3. For per cent survival values. Single Stress Level: The minimum number of specimens needed depends upon the desired width of the confidence interval for each parameter. Estimates of Parameters.

and for n = 3 to 25. TABLE 2.5 ± 0. a.0 385 190 84 47 30 21 16 13 10 8 .further thereproductions sample sizes should be .2 1. p. a. and 99.0 10 1.NUMBER or SPECIMENS AND THEIR SELECTION B.9 per cent survival curves for four values of confidence level. 0. Confidence Intervals: For the Mean. including 50 per cent. n 08 2.6 0. Width of Interval Number of Specimens. 99.14<7 0.3 ± 0.) The number of specimens tested at each stress level can be smaller than the group sizes needed when the life distribution is not assumed (Section V A). Table 2reserved).4 0.4 1.8 ± 0. Each sample is assumed to be drawn at random from its population. The minimum number of specimens should increase as the per cent survival increases.8 ± 0. Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. /*..6 1. is available. If the Escuela sample is estimate ato as well as No /*. Some Estimate of <r Available. Width of Interval 0 2<r 0.9 1.9 ± 1. gives the minimum Copyright by ASTM Int'l (all rights Tue Jul 9 10:11:02 EDT 2013 number of specimens needed for confidence intervals of stated width for the mean. 3. Politcnica delused Ejercito ento Ecuador ESPE pursuant License Agreement.« increases. as well as in Section V B that the fatigue data can be transformed so that they will be approximately Normally distributed. LITE DISTRIBUTION SHAPE ASSUMED 19 1.2 =t 0. TABLE 3.2 0.6 ± 0.3 0. authorized. 90. * The values of re given in Table 30. of the population. _ n = Where: E = width of interval 2 0 Based on Fig.—MINIMUM NUMBER OF SPECIMENS0 NEEDED FOR DETERMINING 95 PER CENT CONFIDENCE INTERVALS OF STATED WIDTH FOR A POPULATION MEAN.7 0. (Note that the rate of decrease is less as.—If a good estimate of the population standard deviation. S-N Curves: Table 33 gives k factors for computing points on 75.0 384 96 43 24 15 11 8 6 5 4 0.7 ±0. » Confidence Limits X ± 0. x = sample mean. Normal Distribution: It is assumed here..4 ± 0. but there is no definite criterion for choosing a particular group size except for the relative magnitudes of the k values. 95. Standard Deviation. 1 of Greenwood and Sandomire (13). Assumed Known. "ASTM Designation E 122 (see footnote 9).1<7 Number of Specimens.—MINIMUM NUMBER OF SPECIMENS" NEEDED FOR DETERMINING 95 PER CENT CONFIDENCE INTERVALS OF STATED WIDTH FOR A POPULATION STANDARD DEVIATION.8 0. 2.5 0.4 06 . A Normal distribution is assumed in all cases.

of a population. is greater than the variability of proThis standard is for EDUCATIONAL USE ONLY.0 0 80% Chance of Detection 52 20 13 9 8 90% 3. 1. —If the problem is to test whether Copyright by ASTM Int'l (all rights reserved).95.5.—The sample sizes for testing the difference between two means are given in Tables 6 and 7.5 4. and the ratio s2/(fixed value)2 is compared with 1/F0. for the new procedure is indeed smaller than the fixed value. 2. some estimate of a must be available.No further . Table 3 can be used as a guide even if no good estimate of a is available.96 in the equation for n (Table 2).5 3. corresponding to °o and n — 1 degrees of freedom for numerator and denominator respectively.975 values from Table 29 should be used instead of the 1.—In order to find the minimum number of specimens needed for determining confidence intervals of stated width for the standard deviation. For example. the sample-standard deviation. since ^0. <r.0 39 15 9 7 6 5 7 Based on Fig. 4. may be above or below a by 0. Escuela Politcnica del Ejercitobeen en Ecuador ESPE pursuant to License Agreement. Two Sample Standard Deviations. Table 4 gives the minimum number of specimens needed to detect a reduction of a stated amount. cedure 2 (the numbers having assigned prior to taking the data). the long-time standard deviation of data based upon an old procedure—and if the other standard deviation is to be computed from data based upon a new procedure that may reduce the variability. However.0 2. whereas an estimate based on n = 30 will not be expected to deviate from the true value by more than 0.—If one standard deviation is a fixed value—for example.—MINIMUM NUMBER OF SPECIMENS" NEEDED TO DETECT IF THE STANDARD DEVIATION OF A POPULATION IS A STATED PERCENTAGE OF A FIXED VALUE.20 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA larger. In some cases. n Percentage of Fixed Value Chance of Detection 7 8 9 12 15 20 28 42 80% TABLE 5. Based on Fig. TABLE 4. 3 from Ferris et al »(14).25<r. say. One Standard Deviation a Fixed Value. if n = 8.—MINIMUM NUMBER OF SPECIMENS0 NEEDED IN EACH SAMPLE TO DETECT IF A STANDARD DEVIATION OF ONE POPULATION IS A STATED MULTIPLE OF THE STANDARD DEVIATION OF ANOTHER POPULATION. (See Section V B4(a) and Table 32. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by the variability of procedure 1. 2 from Ferris et al (14). used to estimate the population standard deviation. Number of Specimens.5 <r. « Multiple Chance Detection 90% of 8 9 12 14 19 26 38 55 40 45 50 55 60 65 70 75 a Chance of Detection 1. since the width of the interval is measured in units of a.) 2. Ta. the principal interest is in the difference between standard deviations. Tests of Significance: Difference Between Two Standard Deviations. These sample sizes apply when the observed standard deviation. Number of Specimens. s. For the Standard Demotion.

0 Taken from Table E. the long-time mean of data based on an old procedure or a commonly used material—and the other mean is to be computed from data based upon a new TABLE 6.975 in Table 29. If the observed value of si is indeed larger than the observed value of sz. stated as a multiple of their equal universe standard deviations. No further reproductio .95 corresponding to (HI — 1) degrees of freedom for numerator and denominator (since n\ = HZ). is given in Table 7. Number of Specimens. <n = az • Number of Specimens.) In this case it is not correct to make the test if s22 is greater than Si2.—MINIMUM NUMBER OF SPECIMENS0 NEEDED TO DETECT A STATED DIFFERENCE BETWEEN A MEAN AND A FIXED VALUE. V B4(6)) is Politcnica compared /0.) No F-ratio test is needed. Escuela del with Ejercito en Ecuador ESPE pursuant to License Agreement. Table 6 gives the minimum number of specimens needed to detect a shift in either direction. <7 = Unknown Standard Deviation of Each Population. measured in terms of the population standard deviation of the new procedure. 2. (See Section V B4(a) and Table 32. Two Sample Means. n 80% Chance of Detection 64 29 90% Chance of Detection 86 39 23 15 11 9 7 5 0 50<r 0 75 1 00 1 25 1 50 1 75 2 00 2 50 0 34 16 10 7 6 5 4 3 44 21 13 9 7 6 5 4 0 50<r 0 75 1 00 1 25 1 50 1 75 2 00 2 50 17 12 9 7 6 4 Taken from Table E of reference (11).—MINIMUM NUMBER OF SPECIMENS0 NEEDED TO DETECT A STATED DIFFERENCE BETWEEN THE MEANS OF TWO POPULATIONS.—The minimum number of specimens needed in each sample to detect a difference in two population means. » Difference Difference 80% Chance of Detection 90% Chance of Detection TABLE 7. One Mean a Fixed Valise. Tue sample Jul 9 10:11:02 EDT 2013 not significantly different and (2) the computed value of / (see Section Downloaded/printed by This standard is for EDUCATIONAL USE ONLY.I of reference (11) • procedure that may shift the mean.NUMBER or SPECIMENS AND THEIR SELECTION 21 ble 5 gives the minimum number of specimens needed in each sample to detect that si is a stated multiple of s%.9?5 in Table 29. (See Section V B4(6).—If one mean is a fixed value—for example. compare s?/s<? with ^0. TheseCopyright sample by sizes apply when the two standard deviations are ASTM Int'l (all rights(1) reserved). Difference Between Two Means: 1. a = Unknown Standard Deviation of the Population Being Estimated. These sample sizes apply when the computed value of is compared with £0.

that is.22 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA V. f(x). is represented by the equation: The Copyright by ASTM Int'l (all rights reserved). Such a distribution curve may be estimated from the raw test data or from transformed test data. In the case of fatigue tests the data observed are usually the lives of specimens tested at a constant applied stress (strain or deflection) amplitude. log log N.to /*. N1'2. No further re . 6.—"Normal" or Gaussian Distribution Curve. The graphical presentation of the distribution of cycle lives for the population of specimens that have lives between certain limits is known as a frequency distribution curve. When the frequency distribution curve has a particular kind of bell shape. and so forth. Escuela Politcnica del Ejercito enthe Ecuador ESPE pursuant License Agreement. Since the cycle life varies from specimen to specimen. as shown in Fig. the populaconstants in the formula are population mean. of samples that could have been taken. The sampling procedure used delimits the-population being estimated. ANALYSIS OF FATIGUE DATA A bask concept of statistics is that a group of one or more specimens is a sample taken from a larger body or population. Such a sample is considered to be just one of a "number. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USEand ONLY. acteristic is not a fixed value and is best described by a frequency distribution. The results obtained from tests on a random sample from the population can be used to estimate the characteristics of the whole population and to measure the precision of the estimates. 6. <r. either from values of N or from values of log N." often very large. the data are said to have a "Normal" or Gaussian distribution. this measurable char- FIG. This Normal probability distribution curve.

Other S-N curves. the median and the mean are equal.11 It should be emphasized that values of the parameters of the population can only be estimated from tests on the specimens in the sample. (a) One Group at Each Stress Level. S-N Curves: These techniques should be used when the actual shape of the distribution of fatigue life values for a given material is unknown or sketchy and the number of specimens tested at each applied stress level is too small. NI is the minimum cycle life value. procedures for testing for statistical control are given in the ASTM Manual on Quality Control of Materials (see footnote 5). have been applied to the analysis of fatigue data. as used by Freudenthal and Gumbel (IS). The median. A7 2 is the second observed value.12 the "extreme value" distribution with and without lower limits. may yield approximately Normal distributions of cycle life.Ejercito en Ecuador . those for p per cent survival (where p is not 50). any set of observations to which these statistical methods are applied is assumed to come from a random sample from the population of interest. generally a transformation to log cycle life is required. to obtain exact values would require that the total population be tested. While references to some . such as the Weibull distribution.of these distributions are included in this Guide. In such cases. or the second order statistic.ANALYSIS or FATIGUE DATA 23 tion standard deviation (a measure of the dispersion). even after various transformations are performed on the data. A. or the first order statistic. to estimate the shape of the distribution. Lack of statistical control in data indicates that the series of samples does not come from the same population. may be fitted to other order statistics if the group size is greater than 1. analysis of the fatigue data has been confined mostly to methods that require no assumptions of distribution shape or to the methods based upon the assumption that the raw data or the transformed data have a Normal distribution. If a series of samples is drawn. Agreement. that are just as normal in the usual sense. these techniques give conservative results. however.del 71. No further rep 12 Escuela Politcnica ESPE pursuant toUSE License See Appendix IV. As stated previously. This is particularly true in the case of tests made at applied stresses near the fatigue limit where runouts are observed. and so forth. This standard is for EDUCATIONAL ONLY. If the group values are arranged in order of magnitude. it is the curve fitted to the medians of the groups at the several applied stress levels. While some fatigue tests.—Usually the first step in the analysis of fatigue data is to draw the S-N curve for 50 per cent survival. and other distributions. say less than 50. as the Normal or Gaussian distribution. an "order statistic." is the middlemost value when the observed values are arranged in order of magnitude. Tue Jul 9 10:11:02 EDT 2013 by InDownloaded/printed the Normal distribution. or the average of the two middlemost values if the group size is even. p. particularly those made in the finite life range of an S-N curve. Others do not yield Normal distributions. Hence. The estimated percentage of survivors for the population at cycle life 11 Copyright by ASTM Int'l (all rights reserved). other distributions. LIFE DISTRIBUTION SHAPE NOT ASSUMED 1.

The value 79 per cent is found opposite sample size 3 in the second column.' 16. 2. 77. the 90. In practice. 23. m = 1. Tue Jul are 9 10:11:02 EDT 2013 not equal to.—MEDIAN PERCENTAGE OF SURVIVORS FOR THE POPULATION. and 10 per cent survival curves may be estimated from the entries in Table 8 and their complements. and the value 21 per cent is obtained by subtracting the value in the second column from 100 per cent.13 Some of these percentages also are given in Table 25 for one group. Nn . andby for the other the time. . TheUSE confidence level associEscuela Politcnica del Ejercito Ecuador ESPE size. unless the data are required for other purposes. where Downloaded/printed standard is for EDUCATIONAL ONLY. Size. 10 " :• 11. 50 70 79 84 87 89 90 91 92 93 94 94 95 95 95 96 30 50 61 69 73 77 80 82 84 85 86 87 88 89 90 S-N curve. which is equal to 1 — i/(n + 1). At least 13 specimens must be tested at each applied stress level to estimate the 95 per cent survival curve. i is the number of the order statistic and nThis is the sample size. or 7V 2 . 13 These are called "median percentages" because. and the 50 per cent confidence level. Examples of how Table 8 may be used follow: 1. the tests may be stopped as soon as the specimen having the median value of fatigue life for the sample has failed. the "expected" by percentage of survivors. ated with expected percentages varies en with the sample whereas it is constant for No furt median percentages. etc. n Sample M- Vajue ^ ^ At the Lowest At the Next Lowest ValuCj 2 3 4 5 6 7 8. c ™~i«. and the 21 per cent survival curves may be estimated from the entries in Table 8 and their complements. 12 13 14 15. the true*percentage will be Copyright larger. 1 :' 9. The 50 per cent survival curve may be estimated from the median of any sample size.". Hence. values of per cent survival less than? 50 usually are not wanted. Table 8 gives the median percentages at Ni and Nz for several group sizes. if all of the specimens in a sample are tested simultaneously.24 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA values of Ni.-. smaller. c. the 79. the value 50 per cent is taken from the median TABLE 8. is 100 — (per cent for N\). depends upon the group size. 3. The median percentage of survivors at the maximum value of the sample. The 50 per cent survival curve may be estimated from the median. pursuant to License Agreement. 50. If 7 specimens are tested at each applied stress level. They close to. If three specimens are tested at each applied stress level. half of the time. but usually ASTM Int'lhalf (allof rights reserved). .

245. Then the median of the particular order statistics (the first. 105. all machines were assumed to be stopped after the second failure. 14 Technically speaking. at Each Stress Level. If S-N curves are based on other confidence levels. but the statement that at least 37 per cent of the population will survive N\ cycles may be made with greater confidence (confidence level = 95 per cent). The probability that at least p Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. and so forth) for the several groups may be used for constructing the S-N curve. The effect of fitting a curve to the same order statistics at several stress levels probably increases the confidence level. #2 162. (5 503. how much is not known.ANALYSIS or FATIGUE DATA 25 As mentioned previously. (140. where Ni is the ith order statistic of the Politcnica del Ejercito en Ecuador pursuant to License Agreement. Escuela is properly called a "tolerance level". per cent of the population lies above Ni cycles. second.) The test data are: Group Life." which are described by Murphy (16). into two or more groups (see references 17 and 18). If the statement is made each time that at least 37 per cent will survive N\ cycles. For each group. 2 281) 105. or Groups. 275 229. Tue statistics Jul 9 10:11:02 EDT 2013 tolerance limits. . so that the time saved could be estimated for this particular set of tests. 668. EXAMPLE. However. at random. they are based on a "confidence level" of 50 per cent. only 5 per cent of such statements are expected to be incorrect. 373 162 229 Entering Table 25.14 Percentage survival values corresponding to higher confidence levels. S-N curves corresponding to a 50 per cent confidence level are usually shown. (Z>) Several Samples. (Actually. the S-N curves on order are "nonparametric Copyright by ASTM Int'l (all rightsbased reserved).—With five testing machines available. the fact should be plainly indicated on the chart. 15 specimens were tested at a constant applied stress level in three groups of 5 each. such as 95 or 99 per cent. under "Lowest Ranking Points. opposite n = 5 in each group. all machines were allowed to run until fracture occurred or until 10 million cycles of fatigue stressing had been applied. (261. 373. 8 695. If estimates of the population percentage are made from a series of samples tested at one applied stress level and the statement is made that at least 79 per cent of the population will survive N\ cycles. Table 25 gives values of percentage survival for several numbers of groups and several confidence levels. 131. the percentage survival values given in Table 8 are median values. if three specimens are tested at an applied stress level. kilocycles 2 3 1 162. 10 000+) Lowest Ranking Points Median From these data we have: 2Vi. the required sample may be divided. and at a confidence level of 50 per cent. 79 per cent of the population are expected to survive N\ cycles (50 per cent confidence level). For example. 10 000+) 275.—If it is not possible to test all the specimens in a sample simultaneously and if stopping the tests before all the specimens have failed is desirable to save time." in the column for m = 3 groups. but ESPE the term confidence level appears to No furth have been used more frequently. are given in Table 25 for a single sample when m = 1. sample. 229. 131. 50 per cent of such statements are expected to be incorrect.

which is the lowest ranking point for m = 1 and n = 15 in Table 25. Similarly. Confidence Interval Estimate.—A point estimate of the population median is the sample median. 45. No further reproductio . For a 95 per cent confidence level. 2. Point Estimate. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for USE ONLY. Estimates of Parameters—Single Stress Level: (a) Median Fatigue Life: 1.03 per cent or more of the population are estimated to survive the 162 kilocycles. at a 95 per cent confidence level.95.61 per cent of the population are estimated to survive 229 kilocycles and.—CONFIDENCE INTERVALS FOR THE MEDIAN.—A confidence interval for the median that does not assume a particular frequency distribution for the population may be computed if the sample size is larger than five. This value is an estimate of the percentage of the population from which the original 15 specimens were selected that will survive 162 kilocycles. S » #30 #3 gives about 80 per cent." at a confidence level of 50 per cent. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. From this.09 per cent for n — 20 TABLE 9. are arranged in order of magnitude as follows: Copyright by ASTM Int'l (all rights reserved). 45 N13 50 " Based on a table in Nair (19). for the "Second Ranking Points. straight-line interpolation between 74. 68." Confidence Level =£ 0. 2.05 per cent. Confidence Limits Sample Size. at a confidence level of 95 per cent. 67.40 per cent or more of the population are estimated to survive 229 kilocycles. Again.11 per cent for n = 10 and 86. #1 N10 #W #11 #12 N12 N15 #W » #24 #27 40. described above in Section V Al(a).26 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA read 87. it is estimated that at a 95 per cent confidence level about 80 per cent of the population will survive 105 kilocycles. N. n Lower Upper 6 7 8 9 10 # #saa # #2 #8 N8 #6 #7 #9 30 12 13 14 15 20 25 35 11 #2 N2 #3 #3 #2 N4 # N6 # #1 #11 #1 . Additional information can be obtained from the preceding test data by considering all 15 specimens as one "group" and determining the percentage of the population expected to survive 105 kilocycles. The n observed values of fatigue life. Ni£N*gNf£ EDUCATIONAL N.

. 226... 224. or 5. if the sample size is 3. p. the following estimates of the popu. J. p...— Confidence limits corresponding to possible values of sample percentage. Point Estimate.—Assume that ten specimens are tested at a particular stress level and the observed values of fatigue life in kilocycles are 201.(N3 . Approximate Confidence Interval Estimate.JVi) and N3 + (N3 .—An approximate confidence interval estimate for the mean that does not assume a particular frequency distribution for the population may be computed as shown in Table 10.. TABLE 10—APPROXIMATE CONFIDENCE INTERVALS FOR THE MEAN. Copyright by ASTM Int'l (all rights reserved).Ni).. namely 235 kilocycles. 245.. 4. Youden." is correct if the sample came from one population. 230. 4 . 238. The point estimate of median fatigue life is the average of the two middlemost values. Ni . p.. 415. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. Point Estimate.ANALYSIS OF FATIGUE DATA 27 The confidence limits corresponding to a confidence level of at least 0.No furth . kilocycles as the stated value of fatigue life.. add (range) /4 to the largest value and subtract it from the smallest value: 3 X range lj£ X range 5 . Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed EXAMPLE. n Procedure". respectively.95 are given by the order statistics designated in Table 9. 2. from which many values in Table 11 were taken. EXAMPLE. 6 Length of Interval 3 . —Using the by data given in This thestandard above example of this Section and 230 is for EDUCATIONAL USE ONLY. for values of n greater than 3. (c) Per Cent Survival for a Stated Value of Fatigue Life: 1.. 2. and 248. Values for other sample sizes may be read from a chart from Dixon and Massey (9). Sample Size. 232.95. 24 244.... add the range of the observed values to the largest value and subtract it from the smallest value: that is.—A point estimate of the population mean is the sample average. Confidence Interval Estimate. for four sample sizes are given in Table 11. 26.—A point estimate of the percentage of the population that has fatigue life values equal to or above a stated value is the sample percentage of observed values equal to or above the same stated value. which are 224 and 245 kilocycles. (b} Mean Fatigue Life: 1. "the median lies between 224 and 245 kilocycles. The population median may be above or below the sample median—235 kilocycles—but the chances are at least 95 in 100 that the statement. Confidence Level = 0.. The interval estimate is defined by A/2 and Ng (see Table 9). Private correspondence from W. use the range: N\ and Ns 0 6 I X range See Youden (20) for n = 3.

larger sample size will give a shorter interval estimate (see Table 11). for n = 40. 204. A particular value is the median.5 65 62. 0 Based on chart from Dixon and Massey (9). the two point estimates would not be exactly equal. per cent).5 55 52.—A point estimate of the population value of fatigue life for a stated value of per cent survival is based on order statistics as TABLE 11. corresponding to 50 per cent survival. per cent » = 20 » = 40 Limits Lower Upper 48 100 Limits Lower Upper P. and.5 70 67. per cent 100 97.5 90 87.—CONFIDENCE INTERVALS FOR PERCENTAGES.5 85 82.per cent 100 n = 10 P. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. Interval estimates for fatigue Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. Point Estimate. Tue Julestimates 9 10:11:02 EDT cent survival) are described in Section V A2(a).28 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA lation value of per cent survival are obtained: (1) point estimate: 70 per cent and (2) interval estimate: 34 to 94 per cent.95. upper: 100 — (tabular value for lower limit cor-' responding to p. (d) Fatigue Life for a Staled Value of Per Cent Survival: 1.5 50 Limits Lower Upper 100 68 100 82 75 68 ." Confidence Level = 0. outlined in the Section on S-N curves: "One Group at Each Stress Level" (Section V A).5 80 77. on chart from Pearson and Hartley (2). per cent 100 95 Limits Lower Upper P.5 60 57. p. —Interval for 2013 medians '(50 per Copyright by ASTM Int'l (all rights reserved). Confidence Interval Estimate.5 75 72. p. cumulative frequency distribution of the observed values. 2. 62 56 51 45 40 36 32 27 100 100 98 97 94 92 88 85 81 77 73 90 54 100 90 85 80 29 99 80 43 98 80 75 70 34 94 70 65 60 15 95 60 25 89 60 55 50 18 82 50 91 87 83 79 77 73 70 67 64 61 58 56 53 51 48 46 43 41 38 36 34 100 100 99 98 97 96 94 93 91 89 87 85 83 82 79 77 *75 73 71 68 66 Where: p = sample percentage (for example. Sample Size »=s #. Confidence limits corresponding to (100 — p) per cent are: lower: 100 — (tabular value for upper limit corresponding to p. per cent). Another point estimate may be derived from the. 415. A. In general. percentage surviving).5 95 92. No further reproductions a .

All the observed values in one group are assumed to come from one population. if the llth. for example. Tue Jul 9 10:11:02 EDT 2013 p = (stated value of per cent survival)/100.15 3. could arise because of differences in material lots or differences in the characteristics of the testing machines. that is. the two groups are said to be significantly different.—-If two or more groups of specimens are tested. The observed differences. If one value appears several times. to have come from two populations with different medians. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. 13th. since the form of the distribution tends to change with change in stress level.ANALYSIS or FATIGUE DATA 29 life values corresponding to other percentage points may be computed by using reference (21) . Rank Test for Two Groups.5. If the observed value falls outside the range of values given in the table. there is a tie. If the observed value falls within the range of values given in Table 26 for the chosen significance level (5 or 1 per cent). The ranks for the two groups are totaled separately and the total for one of the groups (the one with the smaller number of observations if the group sizes are unequal) is compared with the critical values given in Table 26 for sample sizes equal to the group sizes. 12th. they are each given the rank of (11 + 12 + 13 + 14)/4 = 12. The lowest value is given the rank of 1. 1. Int'l and (all rights reserved). For example. the groups may be considered to have come from one population.—In the rank test for two groups the rank of each observation in the two groups combined is determined. Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. the next higher observed value is given the rank of 2. the average of the ranks for those numbers is assigned to each one. Since the populations are assumed to be of the same (though unknown) shape. No further r . only those groups that are tested at the same stress level should be compared. the question of whether the observed differences in the values are due to chance or to some differences in the populations from which the groups were drawn often arises. that is. Tests of Significance: (a) Differences of Group Medians—Single Stress Level. The use of the 1 per cent significance level gives a smaller risk of calling the 15 The interval Nk to Nm may be computed as follows: (1) k is chosen so that (2) m is chosen so that where by ASTM « = 1 Copyright — (confidence level). and 14th values are all equal. and so forth. The rank tests given in this section assume that the several groups are independently and randomly drawn from populations that are of the same shape but may differ with respect to their medians.

5 15. The ranks are totaled separately for each group and the following test-statistic. a discussion of x . would not be expected to occur as often as once in a hundred samples due to chance alone. TABLE 12. N = y^. the number of observations in all groups combined.5 19 20 21 22 23 25 26 17 3 715 811 822 833 841 842 842 869 892 903 944 946 1 .ni. on the average. 87. Copyright by ASTM (all rights approximately reserved).—FATIGUE TEST DATA. 032 1. Thus. Rank Test for More than Two Groups. NZ = 17) for the 5 per cent level of significance. the machines give significantly different fatigue life values.067 1.30 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA groups significantly different when they are actually drawn from one population and the observed difference is due to chance. Hi = number of observations in the ith group.—The method of assigning ranks is the same as for the two-group test. H.—To compare two machines.092 849 667 948 87 Total 27 291 Total 2. see referThis standard(For is for EDUCATIONAL USE ONLY. (See Table 12.) According to Table 26. is computed from the rank totals (22): where: k = number of groups." should be between 101 and 179 (Ni = 10. the value of H calculated from the observed data may be . if the two machines were completely interchangeable. No fur ence (9)). and Ri = sum of the ranks in the ith group. Tue Jul 92 10:11:02 EDT The test-statistic H isInt'l distributed as x with k — 1 2013 de2 Downloaded/printed by grees of freedom if each Ui is at least five. Thus. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. EXAMPLE. the rank test was applied to the data from 27 specimens randomly assigned to two testing machines. and between 89 and 191 for the 1 per cent level of significance. which has "the smaller number of measurements. This means that the actual total. Machine A Rank Kilocycles 624 662 681 Machine B Rank Kilocycles 1 4 5 6 8 2 9 10 24 18 732 688 99 774 781 865 7 11 12 13 14 15. the rank total for Machine A in Table 12. ranking the observations for all the groups combined.

—The test-statistic used to test the significance of the differences among percentage values computed from observed data is x2. TABLE 13.—To compare five machines. corresponding to a 5 per cent significance level or a percentile of 95. the machines may be considered to be interchangeable. Ri. gives x2 = °. randomly assigned to the five machines (see Table 13).. When the sample sizes are unequal: where: k =Copyright numberby of ASTM samples.8 H = 80.56.49. is very much smaller than 9. the observed values of fatigue life may be considered to be from one population. Inspection of the rank totals will usually show which groups are different from the others if the difference is significant. (ft) Differences of Two or More Percentages (for example. Since the computed value of H. Escuela Politcnica del en This Ecuador ESPE pursuant to License Agreement. the rank test was applied to the data from 25 specimens.56 Entering Table 27 with 4 degrees of freedom.49. the populations are said to be different. . Machines A B c (3)539 (12)651 (14)662 (15)675 (19)744 D E Total (5)596 (10)640 (11)646 (18)733 (24)807 Sum of ranks.8 60 3600 720. No fu sum over k Ejercito samples.56 .The formula for x2 may be written in two ways. Int'l (all rights reserved).78 = 2. If H is greater than the x2 value for k — 1 degrees of freedom and the chosen significance level. the groups may be said to have been drawn from two or more populations. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by standard is for EDUCATIONAL USE ONLY.2 4363. that is. . one less than the number of groups. 1. EXAMPLE.ANALYSIS or FATIGUE DATA 31 compared with the values of x2 given in Table 27 to determine whether there may be a significant difference among the populations from which the groups were drawn or not.—FATIGUE TEST DATA.0 63 3969 793.0 49 R* m 2401 480.8 85 7225 1445. the second one is usually better for computation purposes. Bf (6)599 (13)661 (21)760 (22)774 (23)781 (2)530 (8)624 (9)638 (16)684 (25)889 (1)477 (4)568 (7)607 (17)719 (20)757 68 4624 924. per cent survival values). 2.

The computed value of x2 may be compared with the tabular values given in Table 27 for k — 1 degrees of freedom (d.76 6.• = size of ith sample (i = 1. Stress = ±25. EXAMPLE.Xi/£_. The other terms were defined previously. using a significance level16 of 10 per cent..40 3.32 FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA ». the percentages are said to be significantly different. Xi Xi* »i 1 2 3 4 5 6 Total 15 20 17 25 19 14 60. a survival. 2. 2.9 50. The method of computation is the same as that given previously. Another use of the x2 test is to test whether or not the observed percentage values are significantly different from an arbitrary value.—PERCENTAGES SURVIVING 108 CYCLES. ntPer Cent Surviving. and (3) d.000 psi.37 3..ni = average fraction for all samples combined. • • • k). TABLE 14. pp. Significance levels commonly areThis 10standard and is 5 per cent. V)0pi Number Surviving. Tue Jul Downloaded/printed by 16 for EDUCATIONAL USE ONLY. = k.f. replaces p. 175-178). If the computed value of x2 is smaller than the tabular value.f.20 5.50 30. Xi = observed number of "events" in the ith sample. pi = observed fraction for the ith sample: pi = #*/«. If the computed value of X2 is larger than the tabular value corresponding to: percentile = 100 — (chosen significance level). the x2 test was applied to the data given in Table 14.11 = 110 = Copyright by ASTM Int'l (all rights reserved).0 58.0 9 8 10 12 11 7 57 = 5. ESPE pursuant Escuela Politcnica del used Ejercito en Ecuador 9 to 10:11:02 License E A .—To compare six lots of phosphor-bronze strip. except that: (1) the first way of writing the formula for x2 is used for the computations.0 57. etc. that is.0 40. (2) the arbitrary value. which may be called p'. and P = Z_. Lot Sample Size.8 48. an event may be a failure. the samples were drawn from different populations. When the sample sizes are equal the formula reduce to where n = sample size and x = ^ Xi/k (23.). the samples may be considered to have come from one population. such as 50 per cent.88 5.

experience has shown that the per cent survival values tend to lie along a straight line when the data are plotted on Normal probability paper.ANALYSIS OF FATIGUE DATA 33 The tabular value of x2 corresponding to degrees of freedom = 5 and percentile = 90 per cent is 9. While a response curve may be drawn on any type of graph paper. is much smaller than 9. a value of the Normal deviate obtained from Table 28. or Escuela (3) should be consulted for a method of analyzing data. Intercept.. response curves may be constructed from the data if the applied stress levels have been properly chosen. It is based on the values of per cent survival at N cycles that are observed for several (at least 3) values of applied stress. Since the computed value. p = per cent survival in each group of specimens tested. No furthe . Inc.X/k = average of . 4. Norwood. Interpolation between tabular values may be required. The value of 0 corresponding to p is obtained by entering the column headed "Area" with 1 — (/>/100) and reading the corresponding value of z. survive the duration of the test. a = Y 17 No.— If the observed per cent survival values lie along a straight line when plotted on Normal probability paper.. or total number of observed p values. A response curve is constructed for a stated value of the number of cycles-to-failure.18 New York. of course. or No. 2. Y = z. If this has not been done. 358-23.24. 3127. values Downloaded/printed will have approximately This standard is for EDUCATIONAL USE ONLY. such as 107. k = total number of groups tested.28." that is. Mass.. If the computed value of x2 were larger than 9. Copyright by Y.17 (a) Fitting a Response Curve. Codex Book Co.X values. Y is called the transformed value of p. to License Agreement. Politcnica del Ejercito en Ecuador ESPE the pursuant (2). a straight line may be fitted to the points by eye or by the method of least squares.24. The latter method is. and so forth. references (1).24. the lots from which the samples were drawn would be considered to be significantly different. Response Curves: Instead of drawing S-N curves when specimens are tested at several levels of applied stress. Y = ^^ Y/k = average of Y values. 108. and 2^ = sum from 1 to k. no significant difference among samples is indicated by these data and the samples may be considered to have come from one population. ASTM Int'l (all rights reserved). Keuffel & Esser Co. X — ^L. This method of analyzing fatigue data is especially useful when some of the specimens "run out. more precise and not subject to the biases that may be introduced by a person fitting a line by eye. Tue Jul 9 10:11:02 EDT 2013 If the specimens have been allocated as indicated in Table 1. N. the per cent survival by equal weight. The equations for the slope and intercept of a line fitted by the method of least squares to per cent survival values having equal weight are:18 where : X = applied stress value (usually coded to reduce the size of the number) .

05 Yf = -0.67 TABLE 16.78 +0.0. EXAMPLE.00 60.51 -0.00 60.0 24. 46.05 +0.730T .59 X = 3 " The stress values are coded by assigning numbers 1 to 5 to the successive applied stress values. p Transformed Values. 41. Tue Jul 9 10:11:02 EDT 2013 Figure 7 shows the observed values of per cent survival at 107 cycles Downloaded/printed by This standard is for EDUCATIONAL USE ONLY.41 Sum 55 + 6."Area" value) where an Area value corresponds to a value of z = F/ in Table 28. a = —0.00 6.— "PROBIT" TEST DATA. The equation for the fitted line is is used to denote the fitted line values.50 -0.68 25 +7.34 -1.5.X) pf = 100 (1 .50 -1.5 43.00 25.0 44. 40.27 + 0. ksi Number^ Specimens Numberof Specimens 15 Per Cent Survival. .5 46.0 41. AppliedStress.0.25 F .67 -1.-0.33 75. The number of specimens tested at each stress level was chosen in accordance with the procedure outlined in Section III Al. Intercept. 44. 8 5 8 15 14 6 3 2 1 93.8 7. 43. Y „ 1 4 XY Fitted Values F/ Pf 93.—COMPUTATIONS FOR FITTING A RESPONSE CURVE BY METHOD OF LEAST SQUARES. X 1 2 Per Cent Survival. ksi Coded0 Value. as outlined above and demonstrated in Table 16. Ksi Applied Stress. Copyright by ASTM Int'l (all rights reserved).2 52.9 40./.33 75. Slope.—Fatigue tests of specimens of a certain steel gave the data in Table 15 for five applied stress levels.05 9 -0.68 + 1. Applied Stress.0 3 4 5 15 93.34 FATIGUE TESTING AND STATISTICAL ANALYSIS OP DATA fY = a + b (X — X}JThe subscript. so that the per cent survival values would have approximately equal weight.00 6.5. also it is convenient to code the applied stress values when the increments between successive values of applied stress are equal. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.0. TABLE 15.75 16 •> +2.67 + 1. No further reproductions autho .00 25.67 -0.05 + 0. In order to fit a straight line to these data.4 78.50 -0.50 -1. the p values must be transformed to Y values.

(b) Confidence band for fatigue strength values.—Response Curve Tue for a Particular of Steel. No further reproductions authori . Downloaded/printed by Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. FIG. Copyright by ASTM (all rights reserved).Int'l 7. Jul 9 10:11:02Type EDT 2013 This standard is for EDUCATIONAL USE ONLY.ANALYSIS OF FATIGUE DATA 35 (a) Confidence band for per cent survival values.

(This method is an alternate for the one described in Section V A2(c) for a single stress level. the computation of confidence limits by the method outlined here assumes that the Normal transformation of the per cent survival values is suitable. Confidence limits for per cent survival at N cycles for a stated value of applied stress (Fig. Downloaded/printed by The risk. is evenly divided between the two limits. Two types of confidence limits may be computed for a response curve that has been fitted by the method of least squares: 1. connect these plotted values by a straight line.0081 0. Confidence limits for fatigue strength at N cycles for a stated value of per cent survival (Fig. Applied Stress. This derived estimate is called the fatigue strength for p per cent survival at N cycles. is: where F = the transformed value of the stated per cent survival value. This standard is for EDUCATIONAL USE ONLY.called X. ajfunction of F. To draw in the line fitted to the observed values by the method of least squares.«tudies the derived estimate.F/)* — 1 50 — 0 67 0 .0 44.19—The method of computing symmetrical20 confidence limits fora response curve plotted as a straight line on Normal probability paper when the test specimens have been allocated according to Table 1 is presented here. The two types of confidence limits are computed differently. the standard deviation of the ob19 20 Copyright by ASTM Int'l (all rights reserved).36 FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA (open circles) plotted on Normal probability paper. ^(F). For either type of confidence limits.5 46 . Tue Jul 9 10:11:02 EDT 2013 Much of the material in this section is based on Chapter 18 of Hald (24). corresponding to a stated value of p (or F) is often required.0001 0.5 Y Yf — 1 51 (F . select at least two values of applied stress and the corresponding values of pf from Table 16. (The confidence limits are discussed in the next section. 0 Sum + 1 .) In fatigue. 2 5 + 0. 7(6)). 0 4 0 0 0. While no assumption about the distribution of fatigue life values is required. but the resulting confidence bands are numerically equivalent (Fig. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. 7). ksi 4 0 . .0604 (V) Confidence Limits for a Response Curve. that the transformed percentages are Normally distributed about the fitted line. No further re . The equation for the derived estimate.05 + 0 68 + 1 41 0 0001 0 0121 0 . 7(a)). s. 0 41.—COMPUTATION OF STANDARD DEVIATION.67 43. that is. equal to 1 — (confidence level). TABLE 17. 50 — 0 78 — 0.) 2.

27 0. Provided that all Y values have equal weight (see Section III Al).34 0.6 36. . The formula for symmetrical confidence limits is: where: X' X ^(X — X)2 FZ .22 + 0.47 0. 1 2 3 4 5 -1.1 67.88 97.0 0.34 + 1.0 44 5 46.34 0.2)1/2 instead of (2F2.4 85.44 -0.7 17. the computation procedure is given in the following example.4 3.ANALYSIS or FATIGUE DATA 37 served Y values about the fitted line is needed. 1.94 + 1.51 -1.10 0.98 -1.12 -0.3 15. the transTABLE 18.32 + 0.5 43.0 formed values of per cent survival are used.41 + 0. Applied _ Stress. the following equation may be used: The symbols have been defined previously.21 The other symbols have been defined previously.0 41.—METHOD OF COMPUTING 95 PER CENT CONFIDENCE LIMITS FOR PER CENT SURVIVAL VALUES.—In computing such confidence limits for a stated value of applied stress.40 0.77 0.2)1/2.78 -1. Confidence Limits.55 0. The confidence limits should be computed for all the applied stress levels 21 Hald (24) uses t = (Fi.10 0. EXAMPLE. is computed for the X values for the data. = average of X values for the data used in fitting the line.47 Y Values Lower Upper p Values Lower Upper 40.0 Sum . k . .45 0.—Table 17 shows the method of computing the standard deviation of F values about the fitted line.02 + 1. Confidence Limits for Per Cent Survival Values at N Cycles. and = a factor obtained from Table 32 for 2 and k — 2 degrees of freedom for numerator and for denominator.9 62. k . X i 4 1 0 1 4 10 0. Downloaded/printed by This standard is for EDUCATIONAL USE ONLY.6 86.77 0.05 -0. No further . using the values of F and F/ given in Table 16. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.04 -0.55 0.68 + 0.k -2 = the stated value of applied stress (coded). Tue Jul statements 9 10:11:02are EDT 2013 the same fitted line. ksi Coded Value. The wider limits given here ensure Copyright the desired level confidence made using byconfidence ASTM Int'l (all when rights several reserved).40 0.0 41. respectively.

46 4. 96 1.14 (4.69 " 38.99 + 1. is 74 to 92 per cent.82 0. using the value of 5 computed in Table 17 and the Yf values from Table 16.54 Fatigue Strength.50 + 0. for denominator. 38.22 40. EXAMPLE.52 43. the point estimate of the population of per cent survival at 107 cycles is 85 per cent.38 FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA used in the experiment and may be computed for any other value within the range of the applied stress values. and p = 100 (1 — Area value).f.28 -0.82 1.36 ksi. corresponding to a confidence level of 0.50 + (1. Escuelashould Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.58 4.50 corresponds to zero on the coded scale.72 0.5 X coded value).48 0.22 For example. 2.86 = 39. Confidence Limits for Fatigue Strength at N Cycles.68 5.10)1/2 = 4.—Table 18 shows the method of computing 95 per cent confidence limits for per cent survival. 22 This standard is for EDUCATIONAL USE ONLY.95.—These are confidence limits on the derived estimate. X(Y}. 66 3. The method of computation is given in the following example. Tue Jul 9 10:11:02 EDT 2013 The other symbols have Downloaded/printed by been defined previously.95 in Table 32 for 2 d. . 7(a) and connected by dotted lines to form a confidence band. for an applied stress of 41 ksi.57) = 38.37.32 2.62 43.098 0. No further r Caution be used in drawing conclusions from extrapolated portions of the response curve or the confidence band. The exact formula for the confidence limits is: where: F = the transformed value of the stated per cent survival value and Copyright by ASTM Int'l (all rights reserved). Es(2F* i)"t Confidence Limit! CJ • - c p F Coded X(Y) Values Coded Values Lower Upper 1.282 0.81 2.32 42.98 3. For example: lower confidence limit for fatigue strength corresponding to p = 90 is 38.s)1/2 = 0. an Area value corresponds to a z = Y value in Table 28.07 2. s(2F2.61.5 X 0.68 0.93 39. the number of observed p values.62 0.43 0.36 41.51 0.00 0.—METHOD OF COMPUTING 95 PER CENT CONFIDENCE LIMITS FOR FATIGUE STRENGTH VALUES.67 0 + 0.95 46.87 45.37) = 0.31 0. any value of applied stress within the range used in the experiment. 07 4.31 44.55 3. at 107 cycles.67 3.57 1.55. The confidence interval estimate. 06 4. The band applies to TABLE 19.15 3. 09 3.72 0.63 2.072 0. 19 2.39 0.50 + (1.331 0.3)1/2 = (19. the value of F0.46 0. ksi° Lower Upper 90 75 50 25 10 -1. ^2. (2F2.f.000 0. see Table 16. The confidence limits computed in Table 18 are plotted in Fig.95 42. where: n = 5.53 0.3 = 9.28 4. for numerator and 3 d.30 0.

Likewise. For example. the point estimate is 41. It will be noted that this confidence band differs from the one in Fig.—Table 19 illustrates the computation of 95 per cent confidence limits for fatigue strength values at 107 cycles. 7(6) and connected by dotted lines to form a confidence band. using the exact formula. The method illustrated in Table 19 should be used when a confidence band is not presented or when the estimates are needed to more decimal places than can be read from a chart. The confidence limits computed in Table 19 are plotted in Fig. The confidence interval estimate.1 ksi.ANALYSIS OF FATIGUE DATA 39 In many cases C deviates only slightly from b and the approximate confidence limits will be close to the exact ones: EXAMPLE. are Jul always applicable but.7 ksi. may Downloaded/printed by not give results that are precise enough for all pur- Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. the extent of the fitted line. Normal Distribution of Fatigue Life: The preceding analyses. 7(a) only at the ends and that.0 ksi. corresponding to a confidence level of 0.5 to 43. numerically.5 to 42. Y = transformed value of p (See Section V A3 (a). the point estimate of the population value of fatigue strength is 43. LITE DISTRIBUTION SHAPE ASSUMED 1.95. shape of the life distributions. Tue 9 10:11:02 EDT 2013 their generality. because of Copyright by fatigue ASTM Int'l (all rights reserved). No further reproduction . In table 19: p = stated value of per cent survival. B. Thus it is recommended that confidence bands be computed according to the simpler method illustrated in Table 18 and the confidence limits for fatigue strength be read from the intersections of the confidence band and the stated value of per cent survival.4 ksi and the confidence interval estimate is 40. for 50 per cent survival. The band applies to any value of per cent survival within the range of observed values. is 42. for 80 per cent survival. the bands are equivalent. that is. which made no specific assumptions about the This standard is for EDUCATIONAL USE ONLY.

It has been found that replacing the observed fatigue life. A 95 per cent confidence interval on /*. If usable inferences are to be made. but also depend on the amount of variation expected in these estimates from experiment to experiment. In Section B. it is itself a random quantity. (a) Choice of Distribution Shape. empirical evidence gathered to date has not allowed a clear-cut choice to be made among them. In order to sharpen the inferences which may be made from the data. Once this has been done. one on which extensive studies have already been made so that laborious computations can be kept at a minimum. The joint efforts of the test engineer and experimental statistician will be needed to make a specific choice which is suitable for the problem being studied. that the precision so gained is real only when the particular distribution chosen honestly reflects the random process being studied. However. EDT 2013 lated from the sample values in a prespecified way. however. it is sensible to choose a form for the fatigue life distribution according to the following two criteria: 1. are assumed to have been transformed so that they are Normally distributed. The process for constructing the interval from the sample can This standard is for EDUCATIONAL USE ONLY. These two conditions can be fulfilled by finding a transformation which. N. the data transformation should not be so variable as to be a function of tjje individual sets of data. There should be satisfactory agreement between the data gathered and the results predicted by the "theoretical" distribution. both as posiEscuela Politcnica Ejercito en Ecuador ESPE pursuant to License Agreement. all random variables. however. In many situations it is sufficient to obtain point estimates of n and a. In this position. for example.+40 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA poses. More meaningful estimates. when applied to a set of fatigue life data. but the problem of what transformation to use is not a cut-and-dried statistical one. can be described in the following way: Once the data from a particular experiment have been collected.—On the assumption that the variable being studied is Normally distributed. The particular transformation which will do the job mentioned above is not necessarily the same for all cases. all the results of Normal distribution theory are available to analyze the transformed data. Some situations may arise. 9 10:11:02value. particular functional forms can be assumed for these distributions and the statistical analyses carried out relative to these forms. where this transformation is inappropriate. the information in the sample can be used to obtain estimates of the parameters for the particular population. however.—In the literature are several specific fatigue life distributions which have some basis in theory. so that it will not be certain to include /x in any single experiment. be determined. unless otherwise noted. in such a way that the chance of \L being covered . can be calcuCopyright by is ASTM Int'l (allto rights reserved). (b) Estimates of Parameters. at worst. 2. an interval which presumed include the Tue trueJul parameter p. It should be pointed out. however. The distribution form should be mathematically tractable or. since these utilize not only the point estimates. by log N will often give a distribution which is approximately Normal. Noto further reproduction tion and length. results in new data which closely follow the Normal or Gaussian distribution. Since this interval is a Downloaded/printed by function of thedel observations. may be gotten by the use of confidence intervals.

where 7 is a confidence level chosen in advance. K. /*. If these estimates are in terms of transformed variables. and the standard deviation. k. One Copyright by ASTM Int'l (all rights reserved). then for any preassigned percentage.ANALYSIS OF FATIGUE DATA 41 by the interval in a given experiment is exactly 0. The numbers. the true value. The standard deviation. with a confidence level of 7. that at least p per rcent of the population is greater than x — ks. The sample median is the less efficient of the two. /x and a. a number. These quantities as they stand serve as point estimates of the population parameters. S-N Curves: Suppose the cycle life (or some transformation of cycle life) of specimens from a certain population is a Normally distributed random variable. has EDT the folDownloaded/printed lowing meaning: if the by applied stress is Sz . p. at least 90Politcnica per cent del of Ejercito the specimens to be tested will survive Nz Agreement. Since the parameters of the fatigue life distribution are not known. then with a confidence level of standard is for EDUCATIONAL USE ONLY. and n. Table 33 lists some values for k. for 90 per cent survival. in the sense that it experiences a larger variation from experiment to experiment but has the advantage of being simpler to compute. however. It is then said. however. "With a confidence level of 0." are functions of p. which are called "one-sided tolerance factors. Consider the problem of constructing a specific S-N curve.. is included in the interval so computed. Any point (Si. where x and s have been computed from a sample of size n as described in Section V Bl.95. What is done in this case is to determine a number k such that the probability of the random variable x — ks not exceeding /x — Ka is exactly 7. of the Normal population. which is equal to the median of the population." The mean. construct a curve whereon any point (£2. the above defined S-N curve cannot be constructed. n and a are unknown and can only be estimated by information obtained from a sample drawn from this population. 2. say. If the parameters of this distribution. — Ka. >. Escuela en This Ecuador ESPE pursuant to License 7. where Xi is the tth sample value from a sample of size n: where ^ is the sum from I ton. 7. It is then said. cycles. In general. s—are summarized below. Tue Jul 9 ^2) 10:11:02 2013 can. The construction of appropriate confidence intervals is described in Section V B3. No furt . is estimated by the sample standard deviation. may be estimated either by the sample mean or by the sample median. a. x. then 90 per cent of the specimens to be tested will survive Ni cycles. were known.95. the applied stress being fixed. NI) on the curve should give the following information: If the applied stress is Si. they should be converted to cycle life values for presentation. could be determined so that p per cent of the population would have cycle lives exceeding /j. The sample quantities—the mean.

<») as given here.—These are discussed in reference (9).—The statistics. Select a confidence level. x and 5. may be chosen in advance in accordance with the specific purpose of the S-N curve.) for several values of ft. (b) One-sided Tolerance Intervals of the Form (-co. say 7 (where 7 is a proportion.f. As in other instances. that are being considered. The value x — ks is then the appropriate abscissa for the ordinate. x + ks). of the transformed cycle lives. on the S-N curve. Tue 9 10:11:02 EDT 2013 bylevel. Keep in mind that there is a risk of 1 — 7 that the interval being constructed will not contain p. The tolerance factors. These factors are not applicable directly to fatigue data analysis. The construction of a confidence interval for n is done as follows: 1. From Table 29. the confidence level. in addition to serving as point estimates of n and a. 2. where t$ is tabulated as a function of degrees of freedom (d. not a percentage). the wider the interval will be. s. and standard deviation. compute fti = (1 — 7)/2 and j82 = (1 + 7)/2. Other sources discuss tolerance limits.—A confidence interval for a Copyright has analogous meaning that for /x and isJul similarly constructed: by ASTM Int'l (allto rights reserved). S. n. These are basically of two types: (a) Two-sided Tolerance Intervals. The 1. taking d. Downloaded/printed Select a confidence remarks about 7 made before are still This standard is for EDUCATIONAL USE ONLY. read fa and fa . Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.) 3. which are different from those given here. Confidence Intervals: (a) Confidence Interval for the Mean*. No further re pertinent. 7. The mechanics of constructing a point for fitting an S-N curve may be described as follows: Given a sample of n cycle lives for a fixed stress level S. nor is it possible to transform them in an elementary way to obtain one-sided factors. k. are also used in constructing confidence intervals for these parameters.f. 7. (Note that fa is negative.—These are completely symmetric with the intervals of the form (x — ks. The confidence level associated with the fitted curve cannot be specified precisely but is probably larger than the confidence level used at each value of applied stress. Having chosen 7. Tables are given there for two-sided tolerance factors. An S-N curve is determined by fitting a curve of appropriate shape to points computed from data at several stress levels as described in the next paragraph. . which is appropriate. p.42 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA The confidence level. for presentation this value may be transformed back to read cycle life directly. = n — 1. and the sample size. read the value of k corresponding to the per cent survival. The desired confidence interval has the limits: (b) Confidence Interval for the Standard Deviation. compute the mean. From Table 33. 3. x. 7. are the same for both types of onesided intervals. 0 < 7 < 1. Also remember that the greater the confidence.

9243 4. License Agreement. The following quantities are relevant (see above): The desired confidence interval for /* has the limits or 4.1304 by ASTM (all rights reserved). 3.0512. 0.9661 and s = 0. The desired confidence interval has the limits EXAMPLE. It is now desired to have 90 per cent confidence intervals on both n and <r. Escuelamedian Politcnica del en X Ecuador ESPE pursuant to cycles.8388 4. with d. the desired confidence interval for <r has the limits: or 0.0726.f. These are obtained using the denning formulas from Section V Bl.—This example will illustrate the construction of confidence intervals. From Table 30.0334 5. Compute ft\ and (82 as before. 5.0828 The estimates are readily determined to be: x = 4.0147 Similarly. This standard is for EDUCATIONAL USE ONLY.9731 4. No further reproductions authori Sample = Ejercito antilog = 92 thousand (This value may be . The entries in the following table are logarithms of fatigue life values as observed in a test on eight specimens: i x{ = log Ni 1 2 3 4 5 6 7 8 4. read: using percentiles 100 ft and 100 ft respectively.9445 4.9777 5. = n — 1.9175.9542 4.ANALYSIS OF FA 43 2. Tue Jul 9 10:11:02 EDT 2013 Note Copyright that the pointInt'l estimates and confidence interval estimates are logDownloaded/printed by arithms and may be converted to cycle life values for presentation.

Let $i be the standard deviation for a sample of size n\ from the first Copyright by ASTM Int'l (all rights reserved).400 cycles. and Soyero and Olds (25).23 Since o-2 is unknown. No further rep See Hald Politcnica (24). 161. it is important to investigate whether the standard deviations are significantly different or not. classified according to the questions they are designed to investigate. If.9175 = 82. Replacing N' by N gives: Sample standard deviation Approximate confidence limits for the population standard ^deviation are derived from the confidence limits for v.700 cycles and antilog 5. Compute the ratio Downloaded/printed by 23 This standard is for EDUCATIONAL USE ONLY.1513<r2)." this section is concerned with a comparison between two sets of data.792 cycles where = means "approximately equal to". p!pursuant 671. .1513 s2) = 93. Escuela del Ejercito en Ecuador ESPE to License Agreement.—Before testing whether or not the means of two samples are significantly different.44 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA found from the original. (a) Difference Between Two Standard Deviations. A summary of these techniques. N = antilog (X + 1. Except for the part on "Differences Among k Means. an approximate value of the sample average may be found by replacing er2 by s2. the principal interest is a comparison of distributions in the sense of deciding whether their paranreters are the same. Then: sample average. however. p.) Confidence limits for the population median are derived directly from the confidence limits for JLI: antilog 4. Tue Jul 9 10:11:02 EDT 2013 population and let s2 be similarly defined for the second.0147 = 103. Sample average. N = antilog (X ± 1. Sample standard deviation: where N' is the population mean in terms of cycle life. 4. Tests of Significance: If cycle life data are collected at several stress levels for the purpose of estimating the life distributions of these several levels. imtransformed observations. then a different set of analyses is needed in order to use the information in the data more effectively. the techniques of analysis just described are applicable. is given in this section.

NOTE. these values are obtained for significance levels of one-half of the level stated.f. respectively.) If the samples are large enough so that the test for the standard deviations would have Copyright by difference ASTM Int'l between (all rights two reserved). while the other half is in the lower tail. the hypothesis that the population means are equal may be tested by the following procedure: compute s2. With a preassigned significance level. when using tables to obtain Fp and //j values. If the chosen significance level (as a proportion) is denoted by a. conclude that the populations are different in mean or in variance or both.95 or 0. compute 0 = 1 — (a/2) (see Note) and. .) If V > Fp . a value Downloaded/printed standard is for EDUCATIONAL USE ONLY.975.ANALYSIS OF FATIGUE DATA 45 of the squares of the standard deviations. a. —-In "two-sided" tests. read from Table 32 the value of Fp corresponding to n\ — 1 ''Degrees of Freedom for Numerator" and n% — 1 "Degrees of Freedom for Denominator. if then In order to decide if the difference is statistically significant. If V ^ F$ . V. If \t \ > t$. an estimate of the common variance. consequently. if /3 = 0. Standard Deviations Not Significantly Different. (On the average. \ 11 >Escuela fy canPolitcnica be attributed to different population means. identical populations will be erroneously judged different about 100 a per cent of the time. (Z>) Difference Between Two Means: 1. Tue Jul 9 10:11:02 EDT 2013 probably detected any by important difference in the variances. by Next compute the test-statistic where x\ and £2 are the sample means for the first and second samples. one-half of the critical region of the distribution curve of the statistic is in the upper tail. compute 0 = 1 — (a/2) and find tp . from Table 29. then the standard deviations are considered to be significantly different.— -If V ^ Fp (see (a) above) . corresponding to d. the degrees of freedom are reversed. + nz — 2. With the same No fur del Ejercito en This Ecuador ESPE pursuant to License Agreement. a significance level must have been chosen in advance. putting the larger standard deviation in the numerator. = n\." (If s-? > Si2. the standard deviations are considered to be not significantly different.

4 0.0060 -0. TABLE 20. after testing specimens with one surface finish.0037 + 0.5 4..9542 4. No further reprod .—COMPUTATIONS FOR SIGNIFICANCE TESTS.1167 45.0418 -0. First Surface Finish Xi = log Ni Second Surface Finish *.7 1. use the nearest smaller integer.6532 4.8388 4.—Ifthe sample standard deviations are significantly different.4 0.0407 +0.5)* 162. another lot is fabricated with a different finish and tested at the same stress level. 2.—Assume that. the hypothesis that the population means are equal may be tested by the following procedures.9243 4..6435 4.2 16.0407 +0..6 56.3 4 *< = log Ni Xi — X (*< -i)« 4. pp. 397-398.5315 4.9777 5.0334 5..1 X 10~ 17.2 EXAMPLE.6902 4.6902 4.24 Compute the numbers: and From Table 29 read a value /j».7243 46.0828 39.0263 -0.0263 -0.x (*.0116 +0. one may conclude that the population means are not significantly different. a = 2(1 — /3) is the approximate proportion of the time when the means are actually equal but will be incorrectly judged unequal.9445 4.7288 -0.0 248.1180 -0. Sample Standard Deviations Significantly Different.4 0.9 6. they are judged unequal. the population means are judged equal and if I /' I > tp.1 x io-4 369.9731 4. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.1273 -0.46FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA reservation about sample size. if I / 1 < tp. If I /' I < tp. The distribution of This standard is for EDUCATIONAL USE ONLY. see Hald (24).0216 -0. The tests are to be analyzed to determine whether the Tue change surfaceEDT finish significantly Copyright by ASTM Int'l (all rights reserved). Jul 9in 10:11:02 2013 affects the fatigue life at Downloaded/printed by the stress level used for the tests.6721 4.2 4.1 5.3 136.6435 4.9 0.0748 139.4949 -0.6232 4.0226 + 0. corresponding to If this value is not an integer.0060 +0.6232 4.0 6. 24 This is an approximate test.0070 +0.0673 +0.5 1.6 16.0119 +0.

In the matter of fatigue life at the test stress level. (See Table 20. Compute the ratio by Copyright by ASTM Int'l (all rights reserved).5 = 2. average variance within samples.ANALYSIS OF FATIGUE DATA 47 fatigue lives at this stress level is roughly log-normal. To test them for significant differences. and sb*. choosing a significance level of 0. each of which is a random sample from a Normal population. and the variance. the mean of the first sample is judged to be significantly larger than the mean of the second sample. (c) Difference Among k Means.) We now have the mean. these populations being taken as having a common standard deviation.m (d. Xi be the mean of the ith group.f.05. s2. Thus the test given in Section V B4(6)l may be used: From Table 29. Escuela Politcnica del Ejercito en This Ecuador ESPE pursuant to License Agreement.: If k (k > 1) sets of data have been obtained. F0. x. = 16) = 2. From Table 32. use the F-ratio test and then the /-test.8 is larger than t0. the second surface finish appears to be inferior to the first. Let:% be the/th observation from the iih group. No fur .976 (d.y. by and where Downloaded/printed x is the mean of the numbers xt .f. and Ui be the size of the tth group. of each sample.12. therefore. variance among samples. denominator = 9) = 4.20. The variances of the two samples are considered to be not significantly different.f. Tue Jul 9 10:11:02 EDT 2013 standard is for EDUCATIONAL USE ONLY. numerator = 7) (d. Since t = 10. the computations are in terms of log N rather than N.12. Define sw2. then one can test the hypothesis that these populations have a common mean. t0.

. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. The Staircase Method: As pointed out in Section III B2. 3. (a) OtUline of A nalytical Method: 1. and S0 = first stress level. Determine whether failures or runouts are the less frequent event. on which the less frequent event occurs. N. Only the less frequent event is used in the analysis. S0. And use +1 if the less frequent event is a run out.. corresponding to k — 1 "Degrees of Freedom in the Numerator" and T^ w. assigned to the stress level for i0 + d. or — ^." The hypothesis that the means of the k populations are equal is accepted only if V < Fp .— jfe "Degrees of Precis dom in the Denominator. Estimate the mean fatigue strength by where: N = total number of the less frequent events. The procedure for experimentation and some discussion of the advantages and limitations are given in Section III.48 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA From Table 32 read the value of F$ (/3 = 1 — a). 5. Ni = number of the less frequent events at the corresponding levels. It is recommended that a statistician be consulted if a more complete exploration of this situation is considered necessary. i = 1 is.—For sample sizes less than 15. etc. the staircase method has been found useful for estimating the mean fatigue strength at a preassigned cycle life. use the fol(a) Downloaded/printed For Small Sample This standard is for EDUCATIONAL USE ONLY. if the less frequent event is a failure. 6. iNj = product of i and Ni at each level. The Modified Staircase Method: Copyright by ASTM Int'l (all rights reserved). d = preselected stress increment. and A = "£iNi. 2. Tue Jul 9 10:11:02 EDT 2013 by Sizes. No further lowing simplified procedure: . Number the stress levels and make a table as follows: Stress i Ni iNi Sz Si So 2 1 0 N2 Ni No 2N2 N! 0 Sum: A where: i = 0 is assigned to the lowest stress.

equal to the preselected stress increment. as described previously.25 ksi n 16 (V) For Reducing Testing Time—For staircase runs involving more than 30 specimens. Test as many groups simultaneously as desired. Subsequently. the time required to complete the staircase method is long because of the sequential nature of the procedure.0).—The following data were obtained in a modified staircase fatigue test: Stress. The n is taken as 1 with the first specimen of the pair which gave opposite results. use a step size. d. 2. Downloaded/printed h This standard is for EDUCATIONAL USE ONLY. the total C is composed of the stresses from n = 2 (62. C 980 m — — = -— = 61. . The (n + l) th . The last n — 1 tests started at n = 2. In the analysis. T. Let n = total number of specimens tested. so that rn = T. d. C m — — n or merely the average stress used on the last n — 1 tests plus that which would have been used on the (n + l)th test. 3).0).0 ksi. equal to twice the preselected stress increment until the first pair of opposite results is obtained. Test each group as a separate small-sample staircase program. independent staircases and conducting these simultaneously. test would have been run at 61. to n = 17 (61. No further reproductions authorized. 1. Therefore. ksi »= 1 2 h 3 4 h 5 6 h h 7 8 h 9 10 11 12 13 14 h 15 16 h 17 62 61 60 59 h h h h h h h h h h h 58 Where x = specimen has failed and o = specimen has not failed (as in Fig. Proceed by dividing the total number of specimens. disregard the specimens used up to the first pair which give opposite results.ANALYS h 1. This long time can be reduced by dividing one long staircase program into several shorter. The best estimate of m for the whole set of T specimens is then: Copyright by ASTM Int'l (all srights iMii reserved). 2. Let C = sum of the stresses used on the last n — 1 tests plus the stress which would have been used on the next test had it been run. Tue Jul 9 10:11:02 EDT 2013 mby= = arithmetic average of r values of m. Start the staircase procedure as before. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. into r groups of n each. or 17th. Then. EXAMPLE. but use a step size. starting with the first pair of opposite results.

Tue In each the surviving specimens are by ASTM (all rights Jul 9 10:11:02case. 1 m 101 102 97 99 99 TABLE 21. 5 Group No. EDT 2013 subjected to an applied stress that is 5 ksi higher than the preceding value. In this test. 4 No. R. and the data arose from testing five groups of 15 specimens each. This is repeated until specimens have failed. 140 ksi (Nominal). STEP0 TESTS OF 42 SPECIMENS. No further reproductions authorized. 3 No. The estimate of the mean fatigue strength of the population is given by: 7.50 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA EXAMPLE.—The step size is 5 ksi. The mean between the last Escuela Politcnica del Ejercitoall en Ecuador ESPE pursuant to License Agreement.—R. The Step Method: The data from the step method described in Section III can be analyzed by the use of arithmetic or logarithmic probability paper. 2 No.Continued. all 42 specimens were first tested at an applied stress of 55 ksi. . Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. Int'l and so reserved). forth. at which stress level all survived 107 cycles. Run = 107 cycles or to failure. Steps = 5 ksi. nex ! column) 0 Starting stress = 55 ksi. the 31 surviving specimens were testedCopyright at 65 ksi. MOORE ROTATING BEAM. SAE 4330 Steel: Tensile Strength. This is illustrated by analysis of the data presented in Table 21. ksi Kilocycles at Failure Stress 950 692 1 077 1 177 5 184 7 661 997 1 524 566 630 108 815 3 850 1 166 681 967 606 680 886 735 1 479 65 60 70 65 h (. Failure Stress. No. ksi Kilocycles at Failure Stress 928 585 2 082 665 969 < 1 567 1 522 490 1 085 1 285 10 390 289 595 384 479 1 039 < 973 329 1 489 1 265 2 076 Failure Stress. The applied stress was then raised to 60 ksi and the test repeated.

5 (77. FIG.2 9.2 38.—Per Cent of Specimens Having At Least the Indicated Fatigue Strength at 107 Cycles. The data are then tabulated as indicated in Table 22.8 35. larger Escuela number of had been some of them would probably .5) 26. 8.5 0 stress survived and the failure stress is the estimated fatigue strength for each specimen.5 67. The data in the last column can be plotted on arithmetic probability paper.5 62. if a very much This standard is for EDUCATIONAL USE ONLY. as shown in Fig. It Tue is Jul interesting to note that the dash-line exCopyright by ASTM Int'l (all rights reserved). 8 by short arrows. Politcnica delspecimens Ejercito en Ecuador ESPE pursuant totested.7 9. Estimated Fatigue Strength.5 72. License Agreement. 8.ANALYSIS or FATIGUE DATA TABLE 22—ANALYSIS OF DATA IN TABLE 21.1 26. ksi 51 Number of Specimens Per Cent of 42 with This Per Cent of 42 Having at with This Fatigue Least This Fatigue Fatigue Strength Strength hhh 11 16 11 4 0 57. 9 10:11:02 EDT 2013 trapolations of the Downloaded/printed by "reasonably straight line" suggest that. The stress levels of 100 per cent and 0 per cent survivals are indicated on Fig. No further reproductions authorized.5 0 100 73.

3 9. further . If the straight line is acceptably close.0 ksi. provided that (1) the power is properly selected and (2) this straight line will intersect the axis of failure stresses at the fatigue limit. Extrapolation on probability paper of the type used for Fig. . If a straight line does not reasonably represent the plotted points. psi. Extrapolation to very low or very high probabilities is entirely unjustified and may be very misleading.9241 1. In such cases.208 C . W. psi « (Mean). 8.7723 .B B — A C . however.50. TABLE 23. 11.0592 0. K = a constant. a statistician should be consulted to determine whether some transformation will Normalize the data.5 ksi and some might not have survived a stress of 55. but it is not necessarily the mean. 9 10:11:02 EDTand 2013the optimum value of n. Otherwise. If the straight line is acceptably close to the plotted points. . This may be expressed by the equation: where: . this assumption is almost certainly not safe. the approximate fatigue strength corresponding to any given mortality ratio (per cent failure) can be estimated by reading directly from the line. the mean fatigue strength is approximately the value of the abscissa at P = 0.5 = fracture stress (final breaking stress).00705 0. a = rate of increase of stress. that value is the median. The Prot Method: The Prot method is based upon the assumption that the curve relating the failure stresses and some power of the rate of increase of stress per cycle is a straight line.3181 . on a straight line. psi. E = fatigue limit. psi per cycle 0.52 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA have a fatigue strength higher than 72. S (Mean). but only between the first and last plotted points. that is.3181 Line'i Change in Log a (A Log a) A C 0 C B 92 700 80 600 71 800 92 700 See Fig. 0.5458 0.8482 . anddel for the standard deviation of thetofatigue is No Escuela Politcnica Ejercito en Ecuador ESPE pursuant License strength Agreement.A 0.208 Log a . and n — exponent of a that makes S a linear function of an.—PROT TEST COMPUTATIONS. psi per cycle.3 9.4699 8. Group Fracture Stress. depends critically on the assumption of a Normal population.3 1 . G.3 0 . Snedecor (26) says that extrapolation is mostly a guessing game. the value above which half of the specimens survived. the Downloaded/printed by power of a that will place the plotted test values This standard is for EDUCATIONAL USE ONLY. A mathematical solution for rights the value of Tue the Jul fatigue strength Copyright by ASTM Int'l (all reserved). the population distribution is probably not nearly Normal enough to warrant analysis by logarithmic or arithmetic probability paper.

and E s c24.6173 0.6812 4. u e The la P o l i t cof n ilines ca d e This l standard E j e is r for c i t oC-A e n are E ccomputed. 21 700 9 600 800 21 700 23 11 2 23 25 13 4 25 700 600 800 700 700 600 800 700 4. i g and h t s A.4334/1.4099 C -B B -A C -A C -B B -A C -A C -B B -A C -A 0.4523 0. Assumed E. B-A.6173/0. but the difference would usually be unimportant. .3102 0.9275 0. The procedure is to assume values of E. f° B ic 69 000 . .3542/0.9031 4. as shown wnloaded/printed by EDUCATIONAL USE ONLY. The arithmetic mean of the individual values of a in each of the three groups is determined.57 0.9241 = 0.3365 3.7287/1.4472 4. Tue Jul value D ofoE.. 9.0645 3.3747 4. in Table slopes C-B.9275/1. n = A log (S . r efor ser v e correct d).ANALYSIS or FATIGUE DATA 53 proposed by Corten et al (5).9823 2. .17 1. the geometric means should be used instead of arithmetic means.—PROT TEST COMPUTATIONS.9241 = 0.A log a 71 000. c The cut-and-try method depends upon the fact that Prot's equation can be written thus: log (S-E) = log K + n log a This plots as through the Copy r iag straight h t b y line AST M I n t points ' l ( a l C.49 0.3365 4.65 1. .E) -5. TABLE 24. The test data may also be plotted as shown in Fig.5458 = 0.9241 = 1. Stress Cycles FIG. The following "cut-and-try" method provides an alternative to the mathematical solution.0792 1.98 0.496 f° B h 1C f° B h 67 000. . psi Group S -E Log (S-E) Line Change in Log (S — E) [A log (S .0792/0. as well as the arithmetic mean of the individual values of S—the fracture stress—in each group.5458 = 0.E)] Slope.1335 3.2764/0.4699 = 0.3102/0. such as those included for illustration in the second and third columns of Table 23.7287 0.4699 = 0.5458 = 0.67 0.3542 1.4334 0.2764 0. with slope n. 9. (Strictly.63 0.3747 4. They represent the entire set of test data fairly well.506 0.—Prot Test: Stress as Linear Function of Stress Cycles.4523/0. l rB.4699 = 0.) This calculation gives three pairs of values. u a d o r E Since SPE pursu .4099 4.

A value for E close to but a little less than the value of the mean failure stress for A must first be assumed. (b) Correct assumed value of fatigue limit.54 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA C. The "cut-and-try" procedure described for three groups can easily be adapted to four or more groups. the slopes of the three lines must come out the same when the correct value of E has been determined. another value of E must be assumed and the calculation repeated. all lie on a straight line. B. in some cases more groups with fewer specimens per group may be advisable. using values of S and A log a given in Table 23. some case-hardened steels—for which the Prot method gave no solution. 25 Although theoretically only three groups of specimens need be used.25 Table 24 is a suggested form for systematizing the trial computations. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. slopes are not within 1 or 2 per cent of one another (see Fig. Some materials have been found—for example.—Log-Log Plot of Prot Data. for the correct value of E in the logarithmic equation. FIG. The quantities indicated in Table 24 are then computed. Copyright by ASTM Int'l (all rights reserved). 10. No further r . Whether or not this means that they have no true fatigue limit is not known. The value of E that makes the slopes sufficiently close to one another so that the three points lie on the same straight line is the estimate of the fatigue limit as* given by the Prot method. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. If the (a) Incorrect assumed value of fatigue limit. and A. 10).

STP91A-EB/Feb. 1964

APPENDIX I MISCELLANEOUS REFERENCE TABLES

Copyright by ASTM Int'l (all rights reserved); Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. No further reproduc

Copyright© 1964 by ASTM International www.astm.org

h h h h h
Minimum Per Cent of PopuNumber Confiof Points dence in Each Level, Group, n per cent
1 3

Lowest Ranking Points Number of Groups, m

Second Ranking Points Number of Groups, m

5

7

9

15

25

1

3

5

7

9

15

25

1

50.00 50.00 50.00 50.00 50.00 50.00 50.00 50 75 25.00 32.64 35.94 37.88 39.20 41.50 43.35 10.00 19.58 24.66 27.86 30.10 34.15 37.51 90 95 5.00 13.53 18.94 22.53 25.14 30.00 34.14 98 2.00 8.40 13.52 17.27 20.10 25.61 30.48 1.00 5.89 10.56 14.23 17.10 23.14 28.14 99 99.9 0.10 1.84 4.76 7.67 10.25 16.12 22.06
50 75 90 95 98 99 99.9 50 75 90 95 98 99 99.9 50 75 90 95 98 99 99.9 50 75 90 95 98 99 99.9 50 75 90 95 98 99 99.9 50 75 90 95 98 99 99.9
0

3

79.37 79.37 79.37 79.37 79.37 79.37 79.37 50.00 50.00 50.00 50.00 50.00 50.00 50.00 63.00 68.84 71.10 72.35 73.18 74.58 75.68 32.64 38.20 40.51 41.84 42.74 44.30 45.55 46.42 58.06 62.71 65.31 67.01 69.89 72.12 19.58 28.37 32.38 34.76 36.39 39.26 41.59 36.84 51.34 57.41 60.85 63.11 66.94 69.89 13.53 23.09 27.82 30.73 32.73 36.32 39.26 27.14 43.80 51.33 55.68 58.57 63.50 67.29 8.40 17.83 23.08 26.43 28.79 33.09 36.67 21.54 38.90 47.27 52.20 55.50 61.15 65.53 5.89 14.75 20.16 23.73 26.28 31.00 34.97 10.00 26.38 36.22 42.47 46.80 54.41 60.42 1.84 8.04 13.18 16.97 19.84 25.43 30.36 84.09 84.09 84.09 84.09 84.09 84.09 84.09 61.42 61.42 61.42 61.42 61.42 61.42 61.42 70.71 75.58 77.42 78.45 79.12 80.26 81.14 45.63 50.91 53.04 54.25 55.06 56.45 57.55 56.23 66.52 70.47 72.65 74.06 76.44 78.26 32.05 41.40 45.38 47.68 49.22 51.90 54.02 47.29 60.65 65.95 68.89 70.80 74.00 76.43 24.86 35.91 40.85 43.77 45.73 49.16 51.88 37.61 53.84 60.64 64.46 66.95 71.13 74.30 17.94 30.04 35.90 39.43 41.83 46.07 49.48 31.62 49.26 57.01 61.41 64.30 69.15 72.83 14.09 26.38 32.71 36.59 39.27 44.03 47.87 17.78 36.81 46.69 52.61 56.58 63.36 68.53 6.40 17.41 24.41 29.04 32.34 38.38 43.40 87.05 87.05 87.05 87.05 87.05 87.05 87.05 68.61 68.61 68.61 68.61 68.61 68.61 68.61 75.79 79.93 81.49 82.35 82.91 83.87 84.60 54.58 59.38 61.28 62.35 63.07 64.29 65.26 63.10 72.17 75.58 77.44 78.65 80.66 82.19 41.61 50.65 54.35 56.46 57.85 60.26 62.15 54.93 67.03 71.68 74.22 75.86 78.59 80.65 34.26 45.40 50.13 52.85 54.67 57,79 60.25 45.74 60.93 67.02 70.38 72.55 76.15 78.85 26.71 39.60 45.40 48.78 51.05 54.98 58.09 39.81 56.75 63.79 67.70 70.23 74.44 77.60 22.21 35.84 42.26 46.07 48.64 53.10 56.63 25.12 44.95 54.37 59.82 63.41 69.41 73.91 12.20 26.10 33.78 38.58 41.90 47.79 52.52 89.09 89.09 89.09 89.09 89.09 89.09 89.09 73.55 73.55 73.55 73.55 73.55 73.55 73.55 79.37 82.97 84.32 85.06 85.54 86.36 86.99 61.05 65.38 67.08 68.03 68.67 69.76 70.61 68.13 76.20 79.19 80.81 81.86 83.60 84.92 48.97 57.45 60.84 62.75 64.01 66.17 67.86 60.70 71.65 75.77 78.00 79.44 81.81 83.60 41.82 52.56 56.97 59.48 61.13 63.96 66.16 52.10 66.18 71.64 74.62 76.53 79.68 82.03 34.17 47.04 52.56 55.72 57.82 61.42 64.23 46.42 62.37 68.75 72.25 74.49 78.20 80.95 29.43 43.38 49.59 53.19 55.58 59.70 62.91 31.62 51.36 60.19 65.17 68.41 73.76 77.73 18.14 33.54 41.35 46.06 49.25 54.80 59.17 93.30 93.30 93.30 93.30 93.30 93.30 93.30 83.77 83.77 83.77 83.77 83.77 83.77 83.77 87.06 89.40 90.27 90.74 91.05 91.58 91.98 75.26 78.28 79.45 80.09 80.52 81.25 81.82 79.43 84.95 86.93 88.00 88.68 89.81 90.66 66.31 72.68 75.11 76.46 77.34 78.82 79.97 74.11 81.87 84.66 86.15 87.10 88.65 89.81 60.59 69.07 72.33 74.14 75.32 77.30 78.82 67.62 78.06 81.86 83.89 85.17 87.26 88.79 53.98 64.81 69.06 71.42 72.95 75.52 77.48 63.10 75.33 79.86 82.28 83.80 86.28 88.09 49.64 61.87 66.79 69.54 71.31 74.30 76.57 50.12 67.05 73.74 77.34 79.63 83.31 85.97 37.63 53.40 60.19 64.02 66.53 70.73 73.92 96.59 96.59 96.59 96.59 96.59 96.59 96.59 91.75 91.75 91.75 91.75 91.75 91.75 91 75 93.30 94.55 95.01 95.26 95.43 95.69 95.90 87.10 88.77 89.41 89.77 90.00 90.40 90.70 89.12 92.17 93.24 93.81 94.17 94.76 95.21 81.90 85.63 87.01 87.76 88.25 89.07 89.70 86.09 90.48 92.01 92.81 93.32 94.15 94.76 78.39 83.53 85.43 86.46 87.13 88.23 89.07 82.23 88.35 90.48 91.59 92.29 93.41 94.23 74.12 80.99 83.53 84.90 85.78 87.24 88.33 79.43 86.79 89.36 90.71 91.54 92.88 93.85 71.12 79.19 82.19 83.81 84.84 86.55 87.83 70.79 81.88 85.87 87.94 89.23 91.27 92.72 62.24 73.74 78.14 80.52 82.03 84.51 86.33

4

5

6

10

20

This standard is for EDUCATIONAL USE ONLY. From Schuette (27). Downloaded/printed by curves faired through test data, subtract three from the number of NOTE—In evaluating fatigue Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. No further reproductions author groups before entering table.

Copyright by ASTM Int'l (all rights reserved); Tue Jul 9 10:11:02 EDT 2013

56

Thhhhhhhhh lation Exceeding Median of: Third Ranking Points Number of Groups, m
1 3 5 7 9 15 25 1 3

Fourth Ranking Points Number of Groups, m
5 7 9

15

25

20.63 9.14 3.45 1.70 0.67 0.33 0.03 38.57 24.30 14.22 •9.76 6.01 4.18 1.31 50.00 35.94 24.66 18.93 13.53 10.56 4.76 57.85 44.68 33.32 27.13 20.93 17.31 9.40

20.63 20.63 20.63 20.63 20.63 20.63 12.33 13.79 14.67 15.28 16.36 17.25 7.00 9.00 10.31 11.25 13.00 14.50 4.73 6.75 8.15 9.20 11.20 12.99 2.88 4.72 6.12 7.20 9.39 11.41 3.65 4.98 6.05 8.29 10.42 2.00 0.61 1.60 2.62 3.54 5.69 7.97

38 . 57 28. 7& 20.96 16.90 12.94 10.65 5.75
50.00 40.51 32.36 27.81 23.04 20.12 13.11 57.85 49.06 41.17 36.58 31.61 28.45 20.44

38.57 30.63 24" . 09 20.54 16.90 14.70 9.50 50.00 42.38 35.73 31.91 27.80 25.19 18.57 57.85 50.82 44.47 40.71 36.57 33.88 26.74

38.57 31.72 25. 99" 22.80 19.46 17.39 12.30 50.00 43.46 37.70 34.35 30.71 28.37 22.25 57.85 51.84 46.38 43.13 39.52 37.15 30.75

38.57 32.46 27.29 24.38 21.29 19.35 14.45 50.00 44.18 39.04 36.03 32.73 30.58 24.90 57.85 52.51 47.65 44.76 41.53 39.39 33.57

38.57 33.76 29.61 27.23 24.66 23.01 18.69 50.00 45.44 41.37 38.98 36.32 34.58 29.85 57.85 53.67 49.88 47.60 45.04 43.35 38.65

38.57 15.91 15.91 15.91 15.91 15.91 15.91 15.91 34.80 6.94 9.40 10.53 11.22 11.69 12.54 13.24 31.51 2.60 5.30 6.83 7.83 8.56 9.92 11.09 5.11 6.18 6.97 8.52 9.91 29.60 1.30 3.56 0.51 2.16 3.56 4.62 5.45 7.12 8.68 27.51 26.15 0.25 1.50 2.74 3.76 4.58 6.28 7.93 0.03 0.46 1.21 1.97 2.67 4.29 6.03 22.51 50.00 31.38 46.44 19.38 43.25 11.21 7.64 41.37 39.26 4.69 37.87 3.27 34.05 1.02 57.85 54.60 51.65 49.87 47.87 46.54 42.83

31.38 31.38 31.38 31.38 23.06 24.63 25.55 26.17 16.64 19.21 20.77 21.84 13.35 16.30 18.14 19.44 10.17 13.35 15.42 16.90 8.35 11.58 13.75 15.33 4.48 7.43 9.66 11.38

31.38 31.38 27.26 28.15
23.78 25.37 21.80 23.77 19.67 22.03 18.31 20.90

14.80 17.91

42.14 42.14 42.14 29.71 33.67 35.32 20.09 26.61 29.50 15.32 22.73 26.22 10.88 18.74 22.73 8.47 16.31 20.54 3.79 10.54 15.02
64.49 54.23 44.83 39.33 33.43

42. 14 42.14 42.14 42.14 36.27 36.91 38.03 38.93 31.22 32.38 34.43 36.09
28.32 29.76 32.33 34.42 25.20 26.92 30.02 32.57 23.21 25.09 28.51 31.36 18.07 20.29 24.47 28.06 64.49 60.38 56.59 54.29 64.49 64.49

74.14 64.46 55.04 49.31 42.95 38.82 28.15
86.84 81.33 75.52 71.74 67.26 64.17 55.26

74.14 67.80 61.67 57.87 53.51 50.58 42.41
86.84 83.27 79.66 77.31 74.53 72.59 66.86

74.14 69.11 64.29 61.30 57.86 55.53 48.92
86.84 84.03 81.22 79.43 77.31 75.83 71.48

74.14 69.85 65.76 63.24 60.33 58.35 52.73
86.84 84.45 82.10 80.60 78.84 77.62 74.02

74.14 70.33 66.74 64.52 61.95 60.21 55.26
86.84 84.73 82.66 81.36 79.83 78.77 75.66

74.14 71.18 68.41 66.70 64.74 63.41 59.61
86.84 85.20 83.62 82.64 81.49 80.70 78.40

74.14 71.84 69.71 68.40 66.90 65.89 63.00
86.84 85.58 84.37 83.62 82.76 82.17 80.45

64.49 64.49 64.49 57.69 59.07 59.86 51.40 54.06 55.58 47.60 51.02 52.98 43.34 47.59 50.05 29.71 40.54 45.30 48.08 20.46 32.94 38.98 42.59 75.79 77.93 78.76 79.23 69.58 73.98 75.68 76.63 65.63 71.48 73.74 75.00 61.04 68.54 71.47 73.10 57.93 66.52 69.91 71.80 49.13 60.64 65.36 68.00

61.28 61.99 58.34 59.71

45.04 49.33 52.74

51.68 54.52 56.76 49.94 53.16 55.71

56.55 58.33

81.94 81.94 81.94 81.94 81.94 81.94 81.94

79.55 80.08 80.50 77.26 78.32 79.14 75.83 77.23 78.31 74.17 75.97 77.36 73.03 75.11 76.71 69.73 72.63 74.85

Copyright by ASTM Int'l (all rights reserved); Tue Jul 9 10:11:02 EDT 2013 USE ONLY. This standard is for EDUCATIONAL Downloaded/printed by Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. No further reproductions authorized.

57

05 0.01 0.01 0.58 FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA TABLE 26. 25 26 27 28.05 0.05 0.01 0.01 0.05 0. 22.01 0.05 0.01 0.01 0.05 0.01 0.05 0. 9.01 0.01 0.01 0.01 0.01 0.05 0.01 0.05 0. 8. 14.01 0.01 0.01 0.05 0. No further reproductio .—UNPAIRED Critical lower and upper rank totals for the 5 per cent and 1 per cent levels of significance.05 0.05 0. 15 16 17. 18 19 20 21.01 0.01 0.05 0.01 0.05 0. 30.05 0. Wilcoxon (28). 12.05 0.01 0.05 0.05 0. 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY.01 0. 10 11.01 0.05 0.05 0.05 0.05 0.01 11 25 12 28 18 37 13 10 13 11 14 11 15 12 16 12 16 13 17 13 18 14 19 14 31 34 35 37 38 41 41 44 44 48 48 51 51 55 54 58 57 62 15 19 16 20 17 21 18 22 19 24 19 25 20 26 21 27 22 28 23 30 23 40 41 26 44 23 45 28 48 24 49 29 52 25 53 31 56 27 56 33 61 28 60 34 65 29 64 36 69 30 68 37 73 31 72 39 77 32 75 41 82 33 42 35 52 55 56 37 60 33 61 39 65 34 65 41 69 36 69 43 74 37 74 44 79 38 78 46 84 40 83 48 89 41 87 50 94 43 91 52 99 44 96 54 103 46 56 47 68 72 73 78 78 83 83 89 89 95 94 100 99 106 104 111 109 117 114 122 119 128 49 44 51 46 54 47 56 49 58 51 61 53 63 55 65 56 68 58 70 60 73 62 87 92 93 98 98 105 104 111 110 117 115 123 121 129 127 136 132 142 138 148 143 154 63 57 66 59 68 61 71 63 74 65 77 67 79 70 82 72 85 74 88 76 91 79 108 114 114 79 131 121 71 189 121 82 138 96 157 128 74 146 88 165 127 85 145 100 164 116 184 135 76 154 90 174 106 194 133 88 152 104 171 120 192 142 79 161 93 182 109 203 139 91 159 107 179 124 200 149 81 169 96 190 112 212 146 95 165 111 186 128 208 155 84 176 99 198 116 220 152 98 172 114 194 132 216 162 87 183 102 206 119 229 158 101 179 118 201 136 224 169 89 191 105 214 122 238 164 104 186 121 209 140 232 179 92 198 108 222 126 246 170 107 193 125 216 144 240 182 94 206 111 230 129 255 111 199 129 223 148 248 97 213 114 238 133 263 132 231 152 256 117 246 136 272 156 264 139 281 137 126 142 129 146 133 150 137 155 141 159 144 163 147 168 151 172 155 177 158 181 162 214 226 222 235 231 244 240 253 248 262 257 272 266 282 274 291 283 300 291 310 300 319 "Adapted from of Dr. 0.01 0. Values h 4 5 6 7 8 9 10 11 12 13 Ni 4 5 6 7. Tue Jul Copyright by work ASTM Int'l Frank (all rights reserved). 13. 29.01 0. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. 23.05 0.01 0.05 0.05 0.05 0. 24.05 0.

Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. No further reproduction .MISCELLANEOUS REFERENCE TABLES RANK TEST.0 in the body of the table refer to the group with the smaller number of measurements. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. NI. 14 15 59 16 17 18 19 20 21 22 23 24 25 160 147 165 151 170 156 175 159 179 163 184 167 189 171 194 175 199 179 203 184 208 188 246 259 255 185 269 171 264 190 278 175 273 196 289 179 283 201 299 184 292 206 309 188 301 211 319 193 310 216 329 197 319 222 339 202 329 227 348 2 0 6 338 232 358 211 237 215 280 294 290 212 305 196 299 218 316 200 309 223 326 205 319 229 337 210 329 234 347 215 339 240 358 220 348 246 368 225 358 251 3 7 9 230 368 257 389 235 378 263 400 239 268 244 316 332 326 241 344 223 337 247 355 228 347 253 366 233 358 259 377 238 368 265 388 244 378 271 399 249 389 277 410 254 399 283 421 260 409 289 433 265 420 295 444 270 301 275 354 372 365 271 384 251 376 278 396 257 387 284 408 263 398 290 419 268 409 297 431 274 420 303 443 2 8 0 431 310 455 286 442 316 466 291 453 323 478 297 464 329 490 303 336 308 395 415 407 303 427 282 418 310 439 288 430 317 452 294 441 324 464 300 453 331 4 7 6 306 464 338 4 8 9 313 476 345 501 319 487 352 513 325 499 359 525 331 510 366 538 337 373 343 438 459 450 338 472 315 462 345 485 321 474 352 498 328 486 360 511 334 498 367 524 341 510 374 536 347 522 382 549 354 534 389 562 360 546 396 575 367 558 404 588 373 411 380 482 505 495 374 519 349 508 381 532 356 521 389 546 363 533 397 559 370 546 405 573 377 558 412 586 383 571 420 600 390 584 428 613 397 596 436 627 404 4 4 609 4 640 411 529 554 543 412 568 385 556 420 582 392 569 428 0 0 596 4 582 436 611 407 4 4 596 4 625 414 609 452 639 422 622 461 653 429 635 469 667 436 6 4 8 477 681 444 579 605 592 451 620 423 606 460 6 3 4 431 620 469 649 438 6 3 4 477 664 446 648 486 678 454 661 494 693 462 675 503 708 469 689 512 722 477 630 658 644 493 673 463 658 502 689 471 673 511 704 479 687 520 719 487 702 529 734 495 716 538 750 504 730 547 765 512 683 713 698 536 729 505 713 5 4 6 745 513 728 555 761 522 743 565 777 530 758 575 792 539 773 584 808 547 739 771 7 5 4 787 770 803 785 820 800 836 816 852 Copyright by ASTM Int'l (all rights reserved).

216 0. .2107 0." Percentiles 0.43 9.00 26.81 7.01 8.57 4.34 63.33 3.26 7.83 3.46 100.30 88.58 1.05 3.67 23.53 83.85 18.24 1.23 5.TueJul 910:11:02EDT2013 Downloaded/printedby This standard is for EDUCATIONAL USE ONLY.84 5.69 66.36 23.61 2.67 40 60 120 a 6 51.63 9.65 2.909)2 = 87.02 20.73 2.70 3.56 50.49 11.34 13.f.79 20.59 12.92 20 24 30 16 18 32.711 1.—PERCENTILES OF THE Degrees 6 of freedom 1 2 2 x DISTRIBUTION.872 1.000039 0.Nofurtherreproductionsauthorized.28 15.86 12.l)i/»p where Za is the normal deviate and n is the number of degrees of freedom.6 for the 99th percentile.53 19.80 9 10 11 12 13 1.57 4.26 3.554 0.58 7.5 3 4 5 7 g 6 0.49 26.412 0.24 1.484 0.09 2.48 20.91 8.99 7.63 6.21 24.15 0.40 79.584 1.57 5.51 16.34 1.0039 0.14 12. EscuelaPolitcnicadelEjercitoenEcuadorESPEpursuanttoLicenseAgreement.31 19.75 18.79 24.95 140.12 27.94 4.21 11.0717 0.01 5.34 24.79 8.02 7.49 28.48 21.58 6.26 7.00 34.25 3.16 0.05 46.38 9.92 18.06 22.40 16.95 22.32 32.68 21.83 14.54 25.16 37.69 29.98 6.82 31.44 15.85 31.41 33.99 17.62 14 15 2.71 4.57 40.07 15.64 From Dixon and Massey (9) by permission of the publishers.42 43.59 25.98 •45.56 0.40 5.28 18.85 3.55 19.84 14.88 10.989 1.86 14.064 1.00 42.60 12.21 158.51 43.77 5.92 23.30 28.61 6.43 40.64 2.53 34.99 28.30 29.23 5.18 2.23 146.26 6.82 4.04 7.87 5.81 9.24 10.Za + (2n .60 5.01 17.48 91.76 59.1026 0. hhhh Yl (2.35 11.95 163.207 0.96 9.326 + 10.39 10.73 26.87 31.41 36.16 37.78 9.14 30.20 2.76 28.45 16.57 152.31 23.19 26.38 91.11 4.5 1 2.60 FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA TABLE 27. CopyrightbyASTMInt'l(allrightsreserved).89 13.86 16.17 2. For large values of degrees of freedom the approximate formula may be used: X«2 = Yz\.0100 0.17 4.89 53.73 3.03 22.09 21. for 60 d.64 12.08 83.0506 0.5 99 99.81 55.23 9.66 5.36 46.55 9.74 26.68 15.85 13.00016 0.68 25.297 0.69 2.02 13.676 0.14 6.81 18.22 27.00098 0. For example.57 7.30 7.831 0.49 4.26 10.0158 0.28 21.89 6.27 34.352 0.96 23.17 39.07 3.20 40.5 5 10 90 95 97.115 0.36 14.70 2.0201 0.81 37.25 7.19 95.66 20.81 21.59 14.07 4.09 16.77 74.60 29.71 35.48 86.60 3.07 12.55 20.31 10.

9 ff 2.650 0.9 — 0.950 0.1 ff h h A* M M A* M M A* At M A* 2.5398 0.5 — 0.090 + 2.750 0.1 0.8 -2.2 2.126 0 - 3.4 0.9987 0.hhh TABLE 28.2 ff 0.025 0.9 2.0035 0.4 ff h — 1. + 0.282 1.0 ff 1.—AREAS OF THE "NORMAL" CURVE.326 1.282 1.282 -1.5 2.036 0.524 + 0.001 0.6 -2.4602 0.4 ff 0.0026 0.4 -2.8643 0.7. + 0.7 — 0.842 0.0013 0.9 ff 1.8413 0.5 2.7881 0.8849 0.9 — 1.5 u — 1. At At At At At At At At At At + + + + + + + + + + 3.8159 0.990 0.9641 0.8 — 1.0287 0 0359 0 0446 0 0548 0.674 0.0 ff 2.842 0.1357 0.4 ff 2 3 ff 2.9332 0.385 At + 0.576 2.300 0.9 3.7257 0.500 At + 0.282 + 1.0 1.5 ff 0.5000 0.0 ff — 2 9 ff — — 2 7 ff — 2 6 ff — — — — — — — — — 2.3 ff h — 0.090 -2.674.2743 0.6 ff 1.3 ff 1 4 ff 1 5 ff 1.700 0.1 1.524 - 1.385.524 -0.6554 0.7 ff 1.5 0.1587 0.1.1 ff 0.6179 0.4 ff 2.326 + 1.0 2.576 2.250 0.0 ff 0.1151 0.5 -2.5 — 1.9918 0.9893 0.995 0.0 -2.9938 0.3 — 0.6 1.253 0.2 — 2.7 ff 2.3 — 1.674 0.6 0.2 — 0.4 — 1.3085 0.960 1.6 — 3.1 ff 2.6 2.8 — 0.2 ff 1.6 ff 0.9 ff 3.3 2.0082 0.9965 0.524 0.326 1.036 0.0 -3.1.350 0.8 f 1.400 0.576 + 2.150 0.645 + 3.1 2.0228 0.6 ff 2.0668 0.674 -0.0062 0.126 From Dixon and Massey (9) by permission of the publishers.9974 0.6 — 0.2 1.253.8 1.385 0.3 ff h — 1 2 ff M A* — 1.090 2.1 —2.2 0.253 -0.8 ff 2.0 — 1.1 ff 1.8 2.126 At At M - n 0.8 0 7 0 6 h — 0 5 ff h — 0 4 ff /* — 0.0019 z X Area -3.7 1.0808 0.200 0.4207 0.3446 0.9 0.7580 0.6915 0.5793 0. +0.8 0.010 0.5 ff 2.960 + 1.0047 0.645 -1.3 0.550 n h h h h h h h h h -1.0139 0.9772 0.3 1.842 -0.999 0.4 — 0.2 ff 2 1 <r 2 1 1 1 1 0 «r 9 <r 8 ff 7 ff 6 ff 2 8 <r 0.9821 0.4 2.1 ff ff ff ff h h h h h — — — — 0.9981 0.9452 6.2420 0.9032 0.4 15 1. Escuela .576 -2.2119 0.960 -1.9713 0.253 At + 0.03 + 0. — 2.126.9554 0.9 1.2 ff h — 0.1 — 1.090 2.7 0.3821 0.3 ff 0.645 1. Copyrig Downlo This standard is for EDUCATIONAL USE ONLY.0968 0.960 1.7 At At At At At At At At At At At At At H At At At /t At M At At At At At At At At At At + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + 0.450 0.645 + 1.0179 0.800 0.100 0.385 -0.005 0.9953 0.850 0.0 — 0.9192 0.2 ff 2.9861 0.7 — 1.036 -0.600 0.0 ff 1.326 -1.3 ff 2.2 — 1.050 0.900 0.8 ff 0.842 + 0.0107 0.7 ff 0.1841 0." z 61 X Area 0.3 -2.975 0. 0 u .9 — 2.

0025 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 00 Degrees of Freedom When the table is read from the foot.40 2.72 1.39 2.69 3.35 2. the tabled values are to be prefixed with a negative sign.62 2.49 2.72 1.97 2.57 2.06 3.04 3.08 3.37 2.12 3.71 1.71 1.81 2.62 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA TABLE 29.5 6.oit* 63.83 3.18 3.36 2.76 2.—VALUES0-5 OF t.45 5.42 2.58 3.01 1.36 2.76 2. Interpolation should be performed using the reciprocals of the degrees of freedom.05 3.81 1.03 2.70 2.92 2.nt ^0.50 3. Copyright by ASTM Int'l (all rights reserved).09 3.9975 1 2 3 4 5 6.03 3.33 3.88 2.13 2.7 9.41 2.07 2.14 2.976 /0.50 3.11 3.77 2. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.75 1.86 2.70 1.98 1.83 1.29 3.73 1.05 3.95 2.07 2.70 1.15 3.58 /0.17 3.81 /0.64 'O.92 2.75 2.60 4.16 2.85 2.23 2.83 2.09 2. Degrees of Freedom to.75 2.00 1.32 4.38 2.76 1.10 3.71 1.38 2.98 2.22 3.20 2.09 2. 0 From Dixon and Massey (9) by permission of the publisher.75 1.63 2.73 1.OB 12.47 2.026 25.86 2.05 2.26 2.53 2.68 1. See Pearson and Hartley (2).02 2.05 2. No furthe .05 2.70 1.1 7.59 2.11 2.69 2.27 2.25 3.18 2.41 2.80 2.51 2.71' 3.96 <0.13 2.36 2.43 3.84 2.08 2.33 2.20 3.77 1.07 3.14 3.92 5.06 2.45 2.16 2.7 4.37 3.66 1.89 1.70 1. 6 The values in this table were computed from percentiles of the F distribution.04 2.90 2.78 2.18 2.36 3.97 2.60 4. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY.06 2.03 3.25 3.94 1.10 2.01 2.50 3.84 4.37 2.24 to.43 2.21 4.56 2.006 127 14.86 1.9i to .31 2.12 2.9875 to.06 2.72 1.30 3.78 1.78 2.67 1.46 2.17 3.30 2.74 1.66 2.82 2.80 1.91 2.71 1.77 4.31 2.79 2.45 2.

55 1.00 1.88 1.52 2.0719 0.00 Interpolation should be performed using reciprocals of .363 0.113 0.57 1.94 1.229 0.543 0.839 1.13 2.507 0.60 1.57 1.39 1.307 0.111 0.00 14 15 0.85 1.90 2.000039 0.1054 0.0158 0.52 1.325 0.178 0.44 1.00098 0.216 0.37 2.67 1.469 0. DISTRIBUTION.298 0.69 3-12 2.25 2.0253 0. Escuela Politcnica del Ejercito en This Ecuador ESPE pursuant to License Agreement.518 0. The values in the table were computed from percentiles of the F distribution.405 0.83 1.99 1.00 20 24 30 40 16 18 0.0517 0.5 1 2.93 1.11 2.94 1.274 0.498 0.04 2.48 1.413 0.300 0.49 1. 0 6 Copyright by ASTM Int'l (all rights reserved).61 3.266 0.60 1.75 1.141 0.32 3.0823 0.369 0.43 2.52 1.36 1. No fu .00 6.5 4 5 2 3 0.763 1.MISCELLANEOUS REFERENCE TABLES 63 TABLE 30.79 1.14 2.402 0.193 0.32 1.88 5.394 0.436 0.72 3.542 0.256 0.177 0.487 0.5 5 10 90 95 97.71 1.556 0.57 1.74 2.46 1.24 1.64 2.611 0.83 1.64 1.0-6 Degrees of Freedom 1 h 0.19 2.432 0.121 0.94 1.79 1.457 0.00 2.582 0.390 0.00 3.367 0.71 2.00 5.720 0.09 2.28 3. From Dixon and Massey (9) by permission of the publisher.39 1.47 1.278 0.38 1.291 0.436 0.41 2.652 0.206 0.0101 0.604 0.87 1.577 0.63 1.34 1.79 2.17 1.663 0.51 2.02 3.63 4.59 1.480 0.342 0.30 2.90 1.349 0.80 2.367 0.232 0.30 4.0383 0.32 1.08 2.72 1.321 0.0513 0.560 0.19 2.168 0.01 1.687 0.60 2.498 0.10 2.145 0.00 60 120 00 0.06 2.272 0.347 0.412 0.525 0.310 0.0743 0.5 99 99.570 0.f.32 2.00 7 8 9 10 11 12 13 6 0.69 1.724 1.416 0.90 1.the degrees of freedom.77 1. See Pearson and Hartley (2).333 0.75 1.41 2.316 0.726 0.30 1.72 1.36 2.625 0.24 2.57 2.08 1.78 3.798 1.522 0.88 1.372 0.117 0.517 0.79 1.460 0.241 0.322 0.18 2.29 2.206 0.463 0.256 0.47 1.22 1.452 0.02 2.27 1.348 0.385 0.0039 0.70 1.273 0.166 0.616 0.67 1.62 2.29 2.35 3.42 1. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by standard is for EDUCATIONAL USE ONLY.417 0.195 0.00 1.453 0.592 0.554 0.84 3.622 0.05 1.80 1.21 2.0239 0.237 0.484 0.00501 0.—PERCENTILES OF THE xVd.699 1.00 2.50 1.00 7.53 1.64 1.00016 0.675 0.774 0.67 1.

96 (9. For example.-«o.os6 AND «o.5 percentiles are 21 + 1.96 (9.976 Ni = Nz 40 42 Mo . a Normal approximation may be used.5 per cent of the time. No further reproductions authorized. Escuela Politcnica del Ejercito en (29).02B MO.74. Copyright by 0 14 15 16 16 17 18 19 20 21 22 28 29 31 32 33 34 35 36 37 27 44 48 50 46 31 33 35 37 38 40 45 49 54 58 50 52 54 56 59 61 66 72 77 83 The values listed are such that a number less than or equal to the Wo. 75 63 22 39 88 24 41 80 68 93 26 43 72 85 99 28 45 77 90 104 30 47 (all95rights reserved).74) / = 14. 82 Tue Jul 109 ASTM Int'l 100 86 115 .1 and 21 Downloaded/printed by 1 2 With permission from Eisenhart and Swed 1. 975 20 21 22 23 24 25 26 27 28 29 30 32 34 36 38.975 will occur not more than 2.The mean is 55 60 65 70 and the variance is .5 per cent of the time and a number greater than Mo.975 FOR RUNS AMONG ELEMENTS IN SAMPLES OF SIZES Ni AND K0.20.74)1/2 = 27.025 «0. for Ni = N2 . Ecuador ESPE pursuant to License Agreement.. the mean is 21 and the variance is 9.TABLE 31°. 9 10:11:02 EDT 2013 The 2.9. For values of Ni and -/V2 larger than 20.5 and 97. 975 Nl tf 3 4 5 7 g 9 10 11 12 15 13 2 3 4 5 6 7 8 9 10 11 12 13 14 IS 16 17 18 19 20 2 4 5 5 5 3 6 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4 4 4 8 g g 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 10 10 11 10 11 11 12 11 12 11 11 11 11 11 11 11 11 11 11 12 12 H 13 13 13 13 13 13 H 1? 1? 13 13 13 13 14 14 15 IS 15 15 16 16 16 16 16 6 2 2 3 3 3 3 3 3 3 3 4 4 3 3 3 4 4 4 4 5 3 4 4 s 4 5 5 5 6 6 6 6 6 7 s 5 5 6 6 6 7 7 g g 8 7 7 S 6 6 7 7 7 7 g g 9 8 g 5 5 7 7 7 g g 8 9 9 9 9 7 g g g g 9 9 5 S S 5 S 11 11 11 12 12 s 14 15 15 15 16 16 17 17 17 17 17 17 15 16 16 17 17 17 18 18 18 19 19 16 17 18 18 18 19 19 14 16 17 19 20 2 2 2 2 3 3 3 5 s 5 5 5 6 6 6 s 9 9 10 10 11 H 1S 14 14 14 15 IS 15 15 IS 18 18 19 19 20 20 19 19 20 20 21 20 21 21 22 21 22 22 18 2 2 2 3 3 3 4 4 4 5 5 5 s s 5 6 6 6 6 6 7 7 7 9 9 10 9 10 10 9 10 10 10 10 11 10 10 11 11 ii S 12 13 11 12 12 12 n n n H s 14 s 5 S 19 20 20 20 21 71 21 22 22 22 22 23 23 23 24 22 23 24 24 ?4 74 24 25 7S 25 25 26 76 76 77 Nl = #2 «0 .o26 value will occur not more than 2.025 MO . This standard is for EDUCATIONAL USE ONLY.

74 3.86 3.57 2.02 1.34 2.17 2.70 3.54 2.01 1.93 1.68 3.54 2.30 2.01 1.4 19.26 3.09 6.59 8.09 2.32 2.00 4.01 8.94 8.64 10.25 2.38 3.86 2.54 2.4 19.92 1.76 4.38 4.49 2.25 2.98 1.20 3.40 3.67 3.42 2.89 1.07 3.99 2.87 2.98 1.28 2.77 2.35 2.14 2.25 2.41 2.81 2.20 2.95 1.54 4.69 2.52 3.64 2.90 1.5 19.07 2.61 2.96 1.05 4.66 2.35 4.23 3.85 2.56 4.67 3.82 2.58 1.05 2.1 5.93 2.75 3.77 2.24 2.80 2.92 Copyright by ASTM 00 3.68 3.66 1.64 2.98 3.89 3.35 2.42 2.4 19.34 2.0 19.34 2.5 19.02 2.52 3.21 2.57 2.13 2.29 2.62 1.92 1.32 3.74 4.23 2.01 2.08 4.19 2.74 2.49 3.18 3. (Continued next page) .46 2.24 5.23 2.81 1.37 2.63 7.62 2.88 1.74 2.72 5.79 1.10 3.21 2.76 1.39 2.63 3.32 4.77 3.13 2.14 4.07 2.12 3.35 1.60 2.53 2. No further reproductions authorized.79 1.26 4.91 1.91 1.55 1.4 19.71 1.84 4.89 1.47 3.53 4.72 2.68 4.64 1.45 2.87 1.85 2.70 2.59 2.14 3.67 2.79 2.06 2.91 5.80 2.53 2.68 2.00 1.84 3.36 3.42 2.53 2.12 2.59 3.24 3.96 2.61 2.03 3.95 2.09 3.71 3.54 2.84 1.53 4.10 2.55 2.79 5.95) • 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16 17 18 19 20 21 22 23 24 25 30 40.18 2.04 2.92 2.12 2.22 3.36 4.59 2.37 2.44 3.96 5.46 2.17 2.81 8.66 5.49 2.67 4.49 4.68 1.32 2.20 3.63 2.84 1.26 4.96 1.70 1.30 2.02 1.96 1.15 2.12 9.99 1.55 2.55 3.76 2.66 2.83 1.99 5.71 2.3 19.27 2.11 2.69 3.48 3.30 4.48 2.01 2.46 3.39 2.5 19.10 3.2 19.92 1.00 3. 60 120 12 252 242 254 249 248 244 246 239 241 234 237 216 225 253 250 251 230 161 200 19.08 2.38 2.82 1.20 2.49 2.27 2.31 2.01 1.03 2.TABLE 32.71 2.49 3.44 2.21 3.00 3.49 4.31 2.12 4.90 2.— F DISTRIBUTION.53 2.58 2.25 2.55 8.87 1.70 2. This standard is for EDUCATIONAL USE ONLY.43 5.30 3.5 19.75 4.91 2.12 2.70 2.40 4.74 1.96 1.01 2. Tue 2.84 1.34 2.84 2.82 4.90 2.39 3.62 4.98 2.85 8.63 3.15 2.85 2.01 2.06 3.97 1.40 4.20 2.99 1.66 2.43 1.26 6.79 3.77 1.33 2.76 4.38 2.28 2.00 2.42 3.3 19.18 2.51 3.22 2.87 3.75 5.83 2.16 2.42 2.01 1.49 2. from Table 18 of Pearson & Hartley (2).07 2.96 2.93 2.46 4.87 3.94 2.11 2.44 3.84 1.01 1.66 8.01 2.92 2.48 2.4 19.29 reserved).41 3.15 2.57 8.50 4.75 2.69 2.45 2.5 8.29 3.06 2.41 5.58 2.05 3.15 3.11 2.55 9.07 3.68 rights 2.30 2.78 2.16 2.33 3.91 2.74 4.—Interpolation T hhhhhhhhhhhhhhhhhhhhhhhhhhh 0 Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.46 2.11 2.42 2.70 8.06 2.84 1.74 8.53 2.31 2.11 3.50 1.16 2.65 2.97 2.81 1.51 2.79 2.53 9.38 2.74 2.28 3.79 1.83 4.46 2. This table is taken.46 6.39 3.28 4.10 2.40 2.94 6.73 3.97 3.05 2.5 19.03 2.94 3.43 2.61 1.34 4.45 4.78 1.86 5.90 2.07 Int'l (all 3.65 1.84 2.34 3.08 2.22 3.00 5.24 2.19 2.60 4.04 2.10 3.45 2.77 2.23 2.41 4.33 2.75 1.90 2.51 2.86 1.79 2.36 2.98 1.77 4.07 2.23 2.80 5.35 2.19 5.29 3.53 1.89 1.28 4.06 3.34 3.40 2.94 1.94 1.47 2. 40 60 120 00 Upper 5 per cent Points (^0.84 2.50 3.10 2.38 2.62 2.08 1.16 3.96 EDT 1.22 3.03 2.16 2.74 3.41 3.39 1.57 3.71 2.4 19.85 2.11 2.2 19.73 1.18 3.11 3.01 2.5 18.37 3.92 1.25 1.64 3.0 Degrees of Freedom for Numerator 1 2 3 4 5 6 7 8 9 10 12 IS 20 24 30 .88 1.32 5.69 1.09 2. with permission of the Biometrika Trustees.21 2.19 2.59 3.61 5.51 2.92 1.42 2.62 8.48 3.47 1.25 2.84 3.30 2.95 4.59 5.37 2.39 3.65 2.74 1.71 2.53 2.61 2.81 3.60 2.23 3.57 3.04 6.79 8.18 2.51 1.15 2.13 3.60 2.71 6.28 3.45 2.88 4.59 6.83 2.60 4.27 2.85 2.34 2.15 3.35 4.04 2.40 2.75 1.44 3.28 9.01 1.17 4.35 3.45 2.39 6.37 2.77 4.81 3.59 1.07 2.4 19.16 6.66 2.03 1.09 Jul 9 10:11:02 2.84 1.89 8.58 3.10 2.54 2.32 3.69 5.75 2.51 2.27 2.18 2.93 1.25 2.15 2.39 3.60 4.62 2.96 4.94 2013 Downloaded/printed by NOTE.27 2.76 2.00 4.

9 8.3 39.11 2.56 2.39 3.93 2.22 2.37 3.06 5.18 3.94 2.84 2.82 4.46 4.18 6.86 2.70 2.84 2.38 3.03 3.47 2.39 3.76 4.75 8.04 2.02 2.1 17.— Interpolation should be performed using reciprocals of the degrees This ofstandard freedom.12 6.90 2.90 2.5 14.91 1.25 3.51 5.56 2.82 1.03 2.98 2.97 5.44 3.07 3.65 4.12 2.5 14.78 2.52 EDT Copyright Int'l (all rights reserved). Tue Jul 9 10:11:02 20132.56 8.07 2.48 1.64 5.5 per cent Points (^0.79 1.48 3.9 14.85 3.40 2.20 9.22 3.29 3.75 3.49 2.7 14.20 3.72 3.69 1^57 5.08 2.04 3.87 2.33 2.36 3.24 2.4 39.96 3.26 2.68 5.48 3.67 2.24 4.2 39.24 2.72 6.43 6.51 3.52 6.39 3.4 14.64 2.14 2.55 6.23 3.30 6.14 2.36 4.25 2.1 14.37 3.14 3.8.04 2.76 2.0 13.72 2.56 4.22 2.2 10.58 .3 14.61 5.01 2.11 ' 2.05 2.15 2.85 2.18 2.32 2.9 2.51 4.42 3.20 2.39 2.33 2.26 5.34 3.77 2.07 4.46 3.16 2.12 2.67 2.09 3.32 4.69 3.18 2.64 1.84 3.41 8.4 39.31 3.94 2.5 39.5 14.05 3.51 12.06 2.31 1.01 2.29 6.20 6.25 3.12 6'.29 3.74 1.00 3.94 1.14 2.05 1.95 3.33 2.42 4.68 2.06 3.99 5.81 2.32 2.06 2.68 6.05 2.13 3.31 2.62 2.0 14. 23 8.27 2.89 3.97 2.L-43 1.83 5.90 4.57 2.47 4.43 7.29 2.38 3.12 4.82 4.82 3.41 3.10 3.76 7.82 2.51 3.01 3.01 „ 1.65 30 5.83 4.78 3.53 40 5.44 3.17 4.13 5.33 3.67 3.33 6.73 3.29 ' 2 .73 2.57 4.60 2.30 2.02 3.00 5.42 4.21 3.01 1.17 3.08 4.75 5.6 8.89 3.08 2.98 6.39 2.92 2.4 39.73 3.07 8.70 2.59 3.45 2.28 3.36 8.72 2.25 3.21 6.35 4.04 5.80 1.37 2.87 2.27 4.50 2.61 1.30 3.88 1.5 39.4 39.82 1.42 5.36 9.76 1.75 5.51 2.77 4.99 2.975) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 969 957 977 948 864 997 985 993 1001 1006 800 922 937 1014 648 900 963 1010 1018 39.47 4.07 10.24 4.36 2.57 4.96 2.05 4.77 2.38 2.46 2.60 4.71 5.87 1.31 8.83 2.87 2.77 3.32 2.07 1.90 8.51 2.11 2.70 4.89 5.99 2.4 39.09 2.33 3.TABLE 32— F DISTRIBUTION— Continued Degrees of Freedom for Numerator 1 2 3 4 s 6 7 8 9 10 12 IS 20 24 30 40 60 120 00 Upper 2. No 2.93 3.41 2.52 3.62 2.93 2.98 9.90 4.89 2.81 8.41 60 2.25 2.47 4.41 5.76 6.20 2.80 2.23 2.00 3.3 14.56 3.98 5.80 120 by ASTM5.50 3.80 3.85 4. .57 2.63 4.33 3.89 3.67 2.0 6.64 2.7.52 2.15 3.51 3.5 39.38 2.51 2.98 1.28 6.02 8.67 1.02 2.61 3.08 2.20 3.85 6.66 3.86 4.46 5.66 3.79 2.61 3.18 3.84 2.67 3.18 3.92 5.01 1.78 3.07 1.78 2.44 2.4 39.61 2.4 16.62 6. 1 8 2.97 4.95 3.54 6.41 2.71 5.88 1.26 3.29 4.12 4.22 3.46 9.84 8.62 3.78 3.16 3.79 2.31 2.39 2.57 2.31 3.28 4.62 4.46 4.67 4.16 2.31 3.15 3.91 2.90 3.44 2.6 14.05 1.51 2.29 ^.18 2.60 2.42 3.12 3.20 3.5 39.98 8.37 4.61 3.89 3.52 5.39 NOTE.3 39.73 3.20 3.66 8.87 5.06 2.67 2.57 2.01 2.30 2.53 2.84 2.43 3.57 2.27 2.73 2.13 2.5 38.69 6.30 2.15 6.73 2.50 3.53 4.10 3.95 3.4 39.5 39.32 4.57 7.43 4.05 2.12 3.8 2.26 6.89 2.68 2.75 2.59 3.72 2.99 4.00 1.46 2.72 5.77 3.19 2.0 15.38 2.21 2.88 3.41 3.41 2.33 2.91 2.55 2.60 3.72 3.62 2.76 3.4 15:1 14.17 3.23 5.79 2.63 2.45 2.38 4.74 2.63 2.59 2.72 1. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.01 4.96 4.48 1.45 2.0 39.42 2.27 4.12 4.17 ^"2.54 2.96 2.50 2.2 39.72 2.94 1.85 5.27 2.56 3.64 2.44 2.70 2.95 2.88 2.05 3.44 3.60 5.97 1.35 2.39 7.94 6.50 2.19 Downloaded/printed by S further reproductions authorized.60 9.25 3.69 7.27 2.07 7.10 3.45 3.15 3.26 6.13 2.35 5.48 4.21 2.87 2.66 3.95 2.38 3.05 3.22 3.87 3.58 3.08 4.10 4.86 3.72 4.82 5.79 2.35 4.53 3.15 4.94 1.41 6.53 1.«2 2.96 2.94 1.94 1.29 3.52 4.76 2.12 4.79 5. is for EDUCATIONAL USE ONLY.29 3.15 00 5.34 3.66 2.9 13.28 3.69 4.2 14.20 3.00 2.05 4.

167 1.895 1.641 3.698 1.353 4.966 1.804 3.520 3.498 0.106 3.736 1.674 1.805 4.134 1. prepared at the Applied Mathematics and Statistics Laboratory of Stanford University under Contract N6 ONRThis standard is for EDUCATIONAL USE ONLY.552 7.210 2.436 2.927 3.602 1.756 2.304 3.087 1.641 4.952 3.118 4.395 2.668 1.242 3.400 3.526 2.068 2.007 0.340 5.691 1.403 2.379 2.780 1.112 5.119 1.355 2.852 3.905 1.702 5.175 3.974 1. p = per cent survival.302 0.514 1.006 2.684 1.455 2.876 0.396 3.854 0.337 0.682 1.686 1.139 1.028 3.860 1.274 4.158 13.544 1. Lieberman was received from the Office of Naval Research.646 1.895 1.183 1. 25126 (NR-042-002).448 2.435 1.171 4.361 2.238 4.776 2.025 1.582 2.215 4.319 4.682 1.358 2.012 1.420 4.483 2.341 4.078 3.257 3.772 4.783 3.904 3.561 3.204 4. and 7 = confidence level.155 2.750 1.212 3.301 0.187 2.329 1.710 2.346 1.140 3.143 3.750 1.188 1.007 2.667 1.052 3.421 2.899 0.566 1.421 3.388 2.299 0.149 1.680 1. with an explanation on the construction of the tables.671 1.471 4.924 3.617 1.219 1.683 1.979 3.206 3.454 2.308 0.201 4.464 3.577 1.919 0.333 2.200 1.684 1.649 2.615 3. After this table was prepared.126 3. H.712 1.065 2.496 3 152 2 680 2 463 2 336 2 250 2 190 2 141 2 103 2 073 2 048 2 026 2 007 1 991 1 977 1 964 1 951 1 942 1 933 1 923 1 916 1 907 1 901 1 895 y 4.607 4.114 1.742 2.494 2.402 2.689 • 1.233 3.061 5.008 3.258 3.220 3.276 4.534 4.528 1.891 0.883 0.085 1.726 3.556 5.804 2. S.95 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 2.705 1.228 2.133 2.781 1.741 5.947 0.691 1.666 4.791 1.407 3.606 1.747 3.683 1.568 2.143 0 In which: n = sample size.311 0. 1.172 3.497 = 0.683 1.363 2.242 1.927 2.701 0.866 1.690 3.220 1. Technical Report No.310 3.957 3.651 7.949 1.687 1.296 1.411 1.723 2.532 1.668 1.091 2.887 1.969 2.842 1.046 1.690 1.415 3.150 3.146 3.—k FACTORS0 FOR S-N CURVES (NORMAL DISTRIBUTION ASSUMED).9 n 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 0.219 2.324 0.414 5.256 1.193 3. Table 33 was originally prepared by D.TABLE 33.637 3.320 0.760 3.672 1.268 1.894 2.042 2.800 1.001 1.205 3.009 3.647 5.629 4.696 1.666 2.265 3.933 0.536 1.755 2.765 2.732 1.316 0.739 2.532 3.961 1.928 1.298 0.062 4.444 3.128 6.585 3.078 4.820 1.822 2.203 5. contains similar tabular values for sample sizes ranging from 3 to Copyright by ASTM Int'l (all r 50.190 2.136 3.300 0.882 3.765 1.577 3.503 2.926 1.515 4.163 3.685 1.520 3.016 1.360 2.508 1.678 1.138 1.624 1.071 1.684 1.670 1.164 4.698 1.243 3.693 1.370 3.464 1.656 2.972 3.619 2.964 0.152 1.303 0.152 1. This report.310 3.532 3. 1 1957.010 0.297 •y = 0.307 0.554 1.853 1.952 9.859 0.719 1.437 4.174 2.686 2.360 0.9 75 90 95 •y = 0.702 1.709 1.595 3.739 1.239 3.364 2.181 3.507 4.675 2.275 2.697 2.042 5.520 1.740 1.858 3.103 6.313 0.208 2.972 1.384 2.724 1.688 1.101 1.299 2.157 3.299 0.849 0.118 4.041 4.857 9.851 2.50 99 99.999 4.108 2.865 0.360 1.688 3.591 1.302.981 3. on "Tables from One-Sided Statistical Tolerence Limits" by G.659 3.502 1.138 3.838 7 5 4 3 3 3 3 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 655 145 202 707 399 188 031 911 815 736 670 614 566 523 486 453 423 396 371 350 329 309 292 10.463 3.273 4.910 4.419 0.415 4.036 4.601 2.032 2.366 1.722 3.159 2.702 1.977 2.773 1.183 3.163 3.465 1.665 2.329 2.686 5.909 0.939 1. 34 of Nov.722 1.331 3.755 2.661 3. P 75 90 95 T = 0.787 4.509 3.058 1.132 7.869 1.364 4.132 1.612 6.830 1.162 1.845 0.90 1.153 3.987 2.370 2. Downloaded/printed by 67 Escuela Politcnica del Ejercito .686 1.376 2.984 0.842 2.545 3.305 0.671 1.126 1.035 1.955 4.371 3.329 0.145 2.357 3.690 4.043 1.732 1.367 2.257 1. Shaffer of the Westinghouse Research Laboratories.158 4.712 1.796 2.676 1.75 99 99.168 3.382 0.501 6.262 3.501 2.215 7.870 0.134 3.900 4.885 2.136 3.143 3.540 1.272 2.373 2.366 2.362 3.411 2.249 2.295 3.694 1.

there is another simple test called the run test that may be used.BB.B . and so on. In the example used. For example. EXAMPLE—Inspection of the data in the rank test example. it may be concluded that the samples are probably different.STP91A-EB/Feb. eight or less runs should not occur. No further reproduc 68 Copyright© 1964 by ASTM International www.AAA . However. First arrange all individuals of both samples in ascending or descending order. Table 12. 8 to 10. such as already has been provided by the rank test. the run test does not indicate clearly that the machines are not interchangeable. and.A .org . According to Table 31.BB . rank 3 counts as a run. on the average. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. suppose the observations from the two samples are identified by A and B. the fact that the ten runs found are close to the critical number of eight runs implies the desirability of another check. fof* NI = 10 and Nz = 17. 7. although it is probably not as sensitive as the rank test. From Table 31. similarly. 4 to 6. If the number of runs is too small.5 per cent). Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. and the ordered series gives ABBAAABABB Here the number of runs is six: A . can be established whether this number of runs is too small for the observations of both samples to have been drawn from one population." Ranks 1 and 2 count as a run. Then count the number of "runs" from each sample. 1964 APPENDIX II ADDITIONAL TECHNIQUE FOR DISTRIBUTION SHAPE NOT ASSUMED RUN TESTS In addition to the rank test for two groups given in Section V A3. shows a total of ten "runs.astm. more than once in 40 times (2. for example. Copyright by ASTM Int'l (all rights reserved).

69 Copyright© 1964 by ASTM International www.STP91A-EB/Feb. quickly applied test is described by Wilcoxon (28) and credited by him to Olmstead and Tukey. "A Corner Test for Association. counting the plotted points passed over before a point is reached on the other side of the z-median. move in from the left. per cent 2 1 Quandrant Sums ±13 ±14 to 15° Use 14 for 14 or more points. No further reproduc Quoted with permission of Frank Wilcoxon (28).astm. Some of the tied points Copyright may be on the side of the median favorable being included in the quadby ASTM Int'l (all rights reserved). Four values are thus obtained to which are attached the signs of the quadrants in which they lie.org ." Table 34 gives critical values of the quadrant sum. This simple. lay a ruler or pencil parallel to the vertical axis and move it to the left. Although statistical analysis of correlation between two variables is not discussed in this Guide. counting the plotted points passed over until the next point lies on the other side of the y-median. substitute for these points a new point. TABLE 34— WORKING SIGNIFICANCE LEVELS FOR QUADRANT SUM. When it is necessary to deal with an odd number of points. The plotted points now lie in four quadrants. one of the coordinates of a point will lie on the ^-median and one of the coordinates of another point will lie on the j-median. These critical values are almost independent of the number of points. Commencing at the right side of the diagram. with those coordinates taken from the original points which do not involve the medians. the following test is very useful in determining whether or not a correlation between two variables probably exists. 15 for fewer points. In a similar manner. Next lay the ruler at the bottom of the diagram parallel to the horizontal axis and move it upwards. Another difficulty arises when tied values are encountered. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. After the scatter diagram has been plotted. the lower left and upper right being taken as plus (or minus) quadrants while the upper left and lower right are taken as minus (or plus) quadrants. horizontally and vertically. referring to their paper. Significance Levels. The algebraic sum of these four values is called the quadrant sum and its expected value is zero if there is no association of the two plotted quantities. Then proceed in the usual manner. the following procedure is to be followed:1 "Two median lines are drawn so as to divide the points into two groups of equal numbers. Tue Julto 9 10:11:02 EDT 2013 rant sum. 1964 APPENDIX in ANALYSIS OF CORRELATION BETWEEN TWO VARIABLES "QUADRANT SUM" CORRELATION TEST . indicating significant association at various probability levels." (30). and one or more Downloaded/printed by of the other members of the tied group may be on the 1 This standard is for EDUCATIONAL USE ONLY. and down from the top. per cent 10 5 a Quadrant Sums ± 9 ±11 Significant Levels. In this case.

11 shows inclusion size plotted against specimen life to failure and suggests a possible correlation. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. No further . Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. EXAMPLE.70 FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA other side of the median. The above test was applied and shows a quadrant sum of —16.—The scatter diagram of Fig. The conclusion is that a definite correlation probably exists between the two variables. Copyright by ASTM Int'l (all rights reserved).—Scatter Diagram. 11. as indicated at the right of the figure. In this case treat the tied group as if the number of its points before crossing the median were: number favorable for inclusion in quadrant sum 1 + number unfavorable FIG.

Moyer.718].. going on to infinity. 71 Copyright© 1964 by ASTM International www. No ^ 0 = minimum life parameter. S. J. J.32). Weibull (31. is for EDUCATIONAL USE ONLY. Gohn and Miss M. N. EDT this 2013 distribution has some theoretical basis. which is the life value where the highest number of failures occur. Timken Roller Bearing Co. Inc. U. that is. chairman. for b > 1. General Motors Corp. and B.by assuming that fatigue failures are examples of extreme values. The distribution is said to have a nonzero minimum life if the curve touches the life axis at a value of life greater than 0. T. General Motors Research Laboratories. B.astm. Robert A. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. it is sometimes referred to as "Fisher-Tippett Type III for smallest values. Semenek. There has been a demand from the roller bearing industry for the inclusion of an additional section covering the use of the extreme-value distribution originally proposed for the analysis of fatigue data by W. composed of: C. chairman. New Departures Div. and. prior to publication in its present form. T. International Harvester Co. It has been revised. Bush. 1964 APPENDIX IV THE WEIBULL DISTRIBUTION FUNCTION FOR FATIGUE LIFE* On the assumption that fatigue failures are initiated at the "weakest link. Since Fisher and Tippett (33) are often credited with first showing that this distribution was one of three limiting types of the extreme-value distribution. Physical Laboratories. 12) is usually skewed to the right. Downloaded/printed that is. The curve representing this function (Fig. they are smallest-strength or weakest-link It has also been used by others This standardvalues. Na = characteristic life parameter occurring at the 63. R. A. Applied Research Laboratories. J. and b > 0 = Weibull shape (or "slope") parameter." the fatigue lives of a group of specimens tested under a given set of conditions may be represented by one of a family of frequency distribution functions: where: N = specimen life. This function is a simple exponential distribution function when 6 = 1. Ruley. the Rayleigh distribution function when b = 2." As pointed byreserved).org . any specimen from the population represented by such a distribution will have zero probability of * This description of the Weibull distribution function. and G. H.57. was originally prepared by a Task Group in Subcommittee VI on Statistical Aspects of Fatigue of ASTM Committee E-9 on Fatigue. Heller. composed of: John K. e = 2. Columbia University. Torrey. General Motors Corp. No further reproduc in the analysis of life test data.STP91A-EB/Feb.2 per cent failure point for the population [63.. Bell Telephone Laboratories. when the mean and the median values are equal. as an addition to ASTM STP 91-A. In other words. Holt. by another Task Group in Subcommittee VI. and a good approximation of the Normal distribution function when b = 3. reaches zero frequency (touches the life axis) to the left of the mode. M. P.\/e). New Departure Div.2 = 100(e .. New York University. Steel Corp. Ruley.... M. Freudenthal and9Gumbel Copyright by ASTM Int'l (allout rights Tue Jul 10:11:02 (34). Kao.

1675 0.097 1.8539 0 9208 .222 1. from Kao (35).0706 0.0458 0. Tue prior Jul 9 10:11:02 EDT Downloaded/printed by standard is for EDUCATIONAL USE ONLY.2076 0.4949 0.2676 0.398 1.0555 0.F(N) 2 4 5 6 8 10 12 14 15 16 18 20 22 24 25 26 28 30 32 34 35 36 38 40 42 44 45 46. Since the data are usually obtained in an ordered manner in fatigue testing.1308 0 1427 0 1549 0. F(N) X 100 log 1 .0969 0.8239 0. 3979 0 4202 0 4341 0 4437 0 4559 0. but if it is reasonable to assume N0 = 0. Later it will be shown how to test for N0 values greater than zero.PUt) 1 F(N) X 100 52 Iog 1 1 .4685 0.2840 0.301 1.0757 0.0862 0.1192 0.7447 0.2366 0. it is easy to fit a cumulative distribution function to fatigue life.—Typical Weibull Distribution Curves.2518 0. 48 50 0. 1 . 000 1. the frequency distribution function is simplified.1871 0.0655 0.1079 0.2218 0. TABLE 35. No further reproductions au .—ORE)INATE LOCATIO NS CORRESPON DING TO PER CENT FAIL!]D VALUES.72 FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA FIG.1805 0. 12. failure prior to N0 life.1249 0.6990 0. Escuela Politcnica del Ejercito en EcuadorThis ESPE pursuant to License Agreement.—All logs are to the base 10.7959 0.2596 0.5528 0 5850 0 6021 0 6198 0 6576 0.5229 0.0088 0. The cumulative function for the of population failed to life N is 2013 Copyright by fraction ASTM Int'l (all rights reserved).0177 0 0223 0 0269 0 0362 0.699 NOTE.3010 54 55 56 58 60 62 63 2 64 65 66 68 70 72 74 75 76 78 84 85 86 88 80 82 90 92 94 95 96 98 0 3188 0 3372 0 3468 0 3565 0 3768 0 .1938 0.

WEIBULL DISTRIBUTION FUNCTION FOR FATIGUE LIFE This function can be transformed into the straight-line relationship 73 which allows a simple graphical method for fitting the Weibull distribution to the data and the subsequent graphical estimation of the parameters (b. it can be constructed rather simply from square log-log paper.—Construction of Weibull Probability Paper from Log-Log Paper. value is 1. 104 Belrose Ave. Y. For example. Columbia University. N. N0. or Technical and Engineering Aids for Management. 13. Similarly. Construction of Probability Paper Although Weibull probability paper can be purchased from a source such as Cornell University. Y.. in Fig. that is. 95 FIG... the ordinate of the 90 per cent failure Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. Tue Jul 9 10:11:02 EDT 2013 Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. 13.000 on the vertical logarithmic scale.. Ithaca. the ordinate for the 20 Copyright by ASTM Int'l (all rights reserved). log-log paper in which the cycles are the same size in both directions. Lowell. N. No further r . Mass. New York. and Na) in the formula. The paper is prepared by the marking off on the vertical logarithmic scale of the probability percentages F(N) corresponding to the values of given in Table 35.

67 53.67 13.00 95.67 6.57 42.35 88.29 28.89 80.00 26.71 80.00 20.42 73.00 62. No No No No 1 2 3 4 5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 5 10 15 20 25 30 35 40 45 50 20 50.00 54. .67 93.15 53.47 82.00 16.78 70.18 27.00 90.71 6.67 50.00 80.46 46.14 71.44 89.00 40.36 45.33 64.00 83.33 41.00 94.00 81.67 83.67 7.50 25.00 58.73 81.89 63.88 11 76 17 65 23 53 29.67 33.33 60.50 43.86 86.08 30.00 88. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed Mean-rank estimatesby = 100 Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.00 66.62 78..68 78.00 56.67 72.00 37.33 66.77 38.11 10.26 10 53 15 79 21 05 26.18 47.85 61.50 50.75 1 12 1 14 15 16 17 18 19 20 91.67 25.00 66.38 23.54 69 .95 55 00 60 00 65.56 11.00 50.43 12.09 18.57 35.00 33.00 57.29 71.12 88.78 83. Order No. 7 No 8 No 9 No 10 No No No No No No No No No No 0 85.67 75. No further reproductions authorized.69 15.74 90.00 31. n.00 92.33 25.16 68.55 63.94 58 82 5.TABLE 36. Samp' e Size.86 57.59 76.67 84.24 61.78 33 33 38.11 47.33 20.50 18.00 70 00 75 00 94.41 35 29 41.25 62.27 36.00 9.00 40.00 46.11 66.31 85.92 7.33 92.25 87. 23 76.00 33.71 70.00 60.00 87.14 64.00 76 19 80 95 85.56 11 11 16 67 22 22 27.25 12.72 75.00 30.— MEAN -RANI C ESTI MATE 3° OF THE PER CE NT PO PULArDION i^AILEI) COR]RESPONDINCJ TO FAILUEJE ORIDER IJf SAM!PLE.67 33.50 77.57 73.22 33.32 31 58 36.75 75.67 60.25 5.75 25.33 14.44 55.33 57.33 40.50 68.00 66.22 77.44 50 00 55 56 5.24 Copyright by ASTM Int'l (all rights reserved).33 44. q No.71 90.43 37.33 16.84 42.00 20.82 90.00 94. This standard is for EDUCATIONAL USE ONLY.33 42.00 66.75 50.91 50.89 44.29 21.37 52 63 00 4 76 00 9 52 00 14 29 00 19 05 00 23 81 00 00 00 00 00 28 33 38 42 47 52 57 61 66 71 57 33 10 86 62 38 14 90 67 43 No 6 No.50 22.86 50.00 75.33 50.48 95.14 14.64 72.43 28.47 85.06 52.50 93.45 8.21 80.89 84.

. Specimen Number of Revolutions to Failure No.0 7. Mean rank. such estimates are recommended by Gumbel (36) and Weibull (37). 1.0 order numbers are given in Table 36 for sample sizes ranging from 1 through 20. No. . the tangent of the angle 6 is an estimate of the Weibull "slope. q. Plotting Positions on Probability Paper: The fatigue data for any one sample are first ordered from shortest to longest life. 7 No 8 0 X 106 7 0 0 5 0 20 0 23 5 No. . as illustrated in Section V A4 of this guide.0969 on the logarithmic scale. An estimate of the population cumulative distribution that corresponds to the data plotted in Fig.0 9. It is possible to calculate this line by the method of least squares. No. for the population line. 2 .5 No. Number of Revolutions to Failure Order. 3 . For example: and Copyright by ASTM Int'l (all rights reserved). Mean-rank estimates of the percentages of the population failed at successive TABLE 37. No. 1 . WITHOUT RUNG UTS. 8 No. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement." b. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. . . 8. 13) is the estimate of the per cent of the population failed. 7 No. q/(n + 1). ." The vertical plotting position of the per cent failed (Fig.1 X 106 revolutions and the ordinate at F(N) X 100 = 16.1 X 106 No.3 No. 2. 5. q No. No. or Kao (35). . 2 No. 4 . 7. 6 No.—TYPI CAL FATIGUE 1PEST DATA.0 6. No further reproduct . . . 4. Estimates of the Distribution Function Parameters: 1. the abscissa for the first specimen is plotted at its life value of N = 1. No. Blom(38) suggests modified mean-rank estimates. but such data are treated separately below under "Estimates of the Distribution Function Parameters. . On such paper. based upon the specimen order number.67. 5. For the data given in Table 37 for the sample taken from lot 1.3 10. 8. 3 No. from 1 through n. .6 13.6 No. the plotting position for the first of a sample of five based upon mean ranks given in Table 36. . . 4 4. F(N).0 8.WEIBULL DISTRIBUTION FUNCTION FOR FATIGUE LIFE 75 per cent failure line is 0. 6 No.0 X 105 5. or the slope of the line may be computed. . 1. No. 6. 2 . All runouts are assumed to have longer lives than the last ordered specimen that failed. . Lieblein (39). . . The angle 0 may be measured with a protractor.0 No. .. 13 can be fqund quickly by drawing a line by eye through the failed points. 3 . 1 No. No. each specimen being given an order number.—TYPICAL FATIGUE TEST DATA. q Lot 1 Lot 2 2 3 5 8 11 13 TABL. is an unbiased estimate of F(N). 6 . More refined techniques for calculating this line can be found by referring to Gumbel (36).E 38. . The horizontal plotting position is its individual life value. 4. 5 No. . Plot of N Versus 1?(N) Nonlinear Order.

14(a)). the minimum life.2 per cent. Thus estimate of b Copyright by ASTM Int'l (all rights reserved). on the fitted line. 13. FIG. b. In Fig.—Estimation of Weibull Distribution Function Parameters for Data in Table 38. No furth Politcnica delin Ejercito enmay Ecuador ESPE pursuant to License Agreement. N0 : (1) note the life value which the curve approaches asymptotically. is obtained from Fig. 13 by reading off the life value corresponding to the intersection of the fitted line and a horizontal line corresponding to F(N) X 100 = 63. 13. is equal to the tangent of the angle 6 shown in Fig. (2) obtain the quantity N — N0 for each point by subtracting the N0 value from each individual specimen life.76 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA Comparisons using these methods as against the graphic method sho. Downloaded/printed This standard is for EDUCATIONAL USE ONLY. the best estimate of N0 will be found so that the data shown in Fig. plot as a straight line. thus the existence of a finite minimum life value greater than 0 would be suspected. is assumed to equal zero. An estimate of the median life is obtained by reading off the life value corresponding to the intersection of the straight line of Fig. The slope parameter. that the latter is usually adequate for small samples. N0 . since the plot of the fatigue data is approximately linear. The skewness of theby Weibull distribution varies with the shape parameter. The plotted data from Table 38 result in a line which curves downward (Fig. To find an estimate of minimum life. Tue Jul 9 10:11:02 EDT 2013 6. b. An estimate of the characteristic life. occur at various per-cent-failed values. andEscuela the Weibull mean. Na . when transformed. 14(6). 14(o) will. 13 and a horizontal line corresponding to F(N) X 100 = 50 per cent. Thus. and (3) plot this life difference on Weibull paper versus the same per cent failed values as before. 4. 2.. by trial and error. 5. . general. 14. Another estimate of b can be made by computing the tangent of 6 from the logarithms of the ordinates and abscissas of two widely separate points. NI and Nz. as shown in Fig.however.w. 3.

Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY.WEIBULL DISTRIBUTION FUNCTION FOR FATIGUE LITE TABLE 39. 16. WITH RUNOUTS. Copyright by ASTM Int'l (all rights reserved).—TYPICAL FATIGUE TEST DATA. FIG. 4 No. 8 1. 4 No 1 No 6 No 3 No. q 77 Specimen No.30 X 106 1. 5 No.'l No.10 2. 3 No. reproduc Escuela Politcnica Ejercito en Ecuador ESPE pursuant to License No further . 7 No. —^Estimation of del Weibull Distribution Function Parameters for Agreement. 6 No. Order.—Per Cent Failed at Weibull Mean. 2 Number of Revolutions to Failure No. 5 No. 7 No. 15. 8 No.60 1. 2 No.70 runout runout FIG.35 2.75 2. Data in Table 39.

78

FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA

that is, the mean does not coincide with the median. Using the estimated Weibull slope, b, it is possible to read from Fig. 15 an estimate of the per cent failed at the Weibull mean and then refer back to the estimated population line on Weibull probability paper, as in Fig. 13, to read off the estimated mean life from the curve. Gumbel (36) and Kao (35) give methods for calculating the Weibull mean1 when the characteristic life Na and the slope b are known. 7. For data containing run-out specimens (Table 39), the n' broken specimens (6 in the example, Fig. 16), out of a total of n specimens tested, are plotted on probability paper at the mean-rank plotting positions, corresponding to a sample size n (8 in the example, Fig. 16(a)). The line drawn through these points will approach a horizontal asymptote, F/racture , which is equal to the ratio of the first plotting positions corresponding to sample sizes n and n', respectively (Fig. 16(a)). The parameters of this distribution may be obtained graphically by plotting only the n' broken specimens at mean-rank plotting positions, corresponding to a sample size n' versus N — N0 , where N0 is again the estimate of the vertical asymptote approached by the curve. The slope of the resulting straight line (Fig16(6)), tan 6 = b, can be obtained as described in this Section. Na , at the probability level of 63.2 per cent, is taken directly from the plotted line. The estimated equation of the probability function for the complete sample of size n will then become

where F/ = rf rac ture . The curve of Fig. 16(a) may now be replotted by using, as ordinates, Fracture times the ordinates of the straight line and, as abscissas, N0 plus the abscissas of the straight line. Note that Na is, in this case, no longer the estimate of the? characteristic life parameter of the complete distribution, F(N). The value of N at the 63.2 per cent probability of failure level may be obtained from the plot in Fig. 16(o).
1

Weibull mean:

where T — the gamma function; and for Weibull variance:

Copyright by ASTM Int'l (all rights reserved); Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. No further repro

STP91A-EB/Feb. 1964
FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA REFERENCES
(1) D. J. Finney, Probit Analysis, Cambridge University Press, 1952. (2) E. S. Pearson and H. O. Hartley, Biometrika Tables for Statisticians, Cambridge Uni versity Press, 1954. (3) R. A. Fisher and F. Yates, Statistical Tables for Biological, Agricultural and Medical Research, Fourth Edition, Oliver & Boyd, London, 1953. (4) T. S. Dolan, "Certain Mechanical Strength Properties of Aluminum Alloys 25S-T and X76S-T," NACA TN914, October, 1943. (5) H. T. Corten, Todor Dirnoff, T. J. Dolan, and Masaki Sugi, "An Appraisal of the Prot Method of Fatigue Testing, Part II," Technical Report No. 35 on the Behavior of Metals under Repeated Stress, ONR Contract N6-ori 071(04), University of Illinois. June, 1953. (6) E. Prot, "Fatigue Testing Under Progressive Loading, A New Technique for Testing Materials," translated by Edward J. Ward, Captain, USAF, WADC TR 52-148, September, 1952. (7) E. J. Ward and D. C. Schwartz, "Investigation of Prot Accelerated Fatigue Test," WADC TR 52-234, November, 1952. (8) A. P. Boresi and T. J. Dolan, "An Appraisal of the Prot Method of Fatigue Testing," Technical Report No. 34 on the Behavior of Metals under Repeated Stress, ONR Contract N6-ori-71, T.O. IV, University of Illinois, January, 1953. (9) W. J. Dixon and F. J. Massey, Jr., Introduction to Statistical Analysis, McGraw Hill Book Co., 1957. (10) E. L. Crow, F. A. Davis, and M. W. Maxfield, Statistical Manual, Dover Publications, Inc., New York, N.Y., 1960. (11) The Design and Analysis of Industrial Experiments, edited by O. L. Da vies, Hafner Publishing Co., New York, N.Y., 1956. (12) D. B. Owen, Handbook of Statistical Tables, Addison Wesley Publishing Co., Inc., Reading, Mass., 1962. (13) J. A. Greenwood and M. M. Sandomire, "Sample Size Required for Estimating the Standard Deviation as a Per Cent of Its True Value," Journal, Am. Statistical Assn., Vol. 45, 1950, p. 258. (14) C. P. Ferris, F. E. Grubbs and C. L. Weaver, "Operating Characteristics for the Common Statistical Tests of Significance," Annals of Mathematical Statistics, Vol. 17, 1946, p. 178. (15) A. M. Freudenthal and E. J. Gumbel, "Minimum Life in Fatigue," Journal, Am. Statistical Assn., September, 1954. (16) R. B. Murphy, "Non-Parametric Tolerance Limits," Annals of Mathematical Statistics. Vol. XIX, 1948, pp. 581-589. (17) E. H. Schuette, "A Simplified Procedure for Obtaining Design-Level Fatigue Curves," Proceedings, Am. Soc. for Testing Mats., Vol. 54, 1954. (18) E. H. Schuette, "The Prediction of Exceedances in Limit-Value Testing," Statistical Methods in Materials Research, proceedings for a short course conducted by the Pennsylvania State University, June, 1956. (19) I£. R. $air, "Table of Confidence Intervals for the Median in Samples from Any Continuous Population," Sankhya, Vol. 4, 1940, pp. 551-558. (20) W. J. Youden, "Systematic Error in Physical Constants," Physics Today, Vol. 14, September, 1961, p. 32. (21) "Tables of the Binomial Probability Distributions," Applied Mathematics Series 6, Nat. Bureau Standards, U. S. Government Printing Office, Washington, D. C., 1949. (22) W. H. Kruskal and W. A. Wallis, "Use of Ranks in One Criterion Variance Analysis," Journal, Am. Statistical Assn., Vol. 47, 1952, pp. 583-621. (23) P. G. Hoel, Introduction to Mathematical Statistics, Second Edition, John Wiley & Sons, Inc., New York, N. Y., 1954. (24) A. Hald, Statistical Theory with Engineering Applications, John Wiley & Sons, Inc. Copyright by ASTM Int'l (all rights reserved); Tue Jul 9 10:11:02 EDT 2013 New York, 1952, pp. 550-551. Downloaded/printed by (25) N. C. Severe and E. G. Olds, "A Comparison Tests on the USE Mean of a LogarithicoThis standardof is for EDUCATIONAL ONLY.

79

Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. No further reproduc

Copyright© 1964 by ASTM International www.astm.org

80

FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA
Normal Distribution with Known Variance," Annals of Mathematical Statistics, Vol. 27, No. 3, September, 1956, p. 670. G. W. Snedecor, Statistical Methods, Fifth Edition, The Iowa State College Press, 1946. E. H. Schuette, "The Significance of Test Results from Small Groups of Specimens," Proceedings, Am. Soc. Testing Mats., Vol. 57, 1957. F. Wilcoxon, Some Rapid Approximate Statistical Procedures, American Cyanamid Co., New York, N. Y., 1949. C. Eisenhart and F. Swed, "Tables for Testing Randomness of Grouping in a Sequence of Alternatives," Annals of Mathematical Statistics, Vol. 14, 1943, p. 66. P. S. Olmstead and J. W. Tukey, "A Corner Test for Association," Annals of Mathematical Statistics, Vol. 18, 1947, pp. 495-513. W. Weibull, "A Statistical Distribution Function of Wide Applicability," Transactions, Am. Soc. Mechanical Engrs.; and Journal of Applied Mechanics. Vol. 73, September, 1951, pp. 293-297. W. Weibull, Fatigue Testing and the Analysis of Results, Pergamon Press, New York N. Y., 1961. R. A. Fisher and L. H. C. Tippett, "Limiting Forms of the Frequency Distribution of the Largest or Smallest Member of a Sample," Proceedings, Cambridge Philosophical Soc., Vol. 24, Part 2, 1928, p. 180. Reprinted in Fisher's Contributions to Mathe matical Statistics, John Wiley and Sons, Inc., New York, N. Y., 1950. A. M. Freudenthal and E. J. Gumbel, "Physical and Statistical Aspects of Fatigue," Advances in Applied Mechanics, Vol. 4, 1956, pp. 117-158. J. H. K. Kao, "A Summary of Techniques on Reliability Studies of Components Using Weibull Distribution," Proceedings, Sixth Symposium on Reliability and Quality Control, Cornell University, Ithaca, N. Y., January, 1960. E. J. Gumbel, "Statistical Theory of Extreme Values and Some Practical Applications," Applied Mathematics Series 33, Nat. Bureau Standards, Feb. 12, 1954. W. Weibull, "A Statistical Representation of Fatigue Failures in Solids," Acta Polytechnica, Mechanical Engineering Series, Vol. 1, No. 9, 1949. G. Blom, Statistical Estimates and Transformed Beta Variables, John Wiley and Sons, Inc., New York, N. Y., 1958. J. Lieblein, "A New Method of Analyzing Extreme-Value Data," Technical Note 3053, National Advisory Committee for Aeronautics, Washington, D. C., January, 1954.

(26) (27) (28) (29) (30) (31) (32) (33)

(34) (35) (36) (37) (38) (39)

SUPPLEMENTARY READING FOR APPENDIX IV
(40) J. H. K. Kao, "The Design and Analysis of Life-Testing Experiments," Transac' tions, 1958 Middle Atlantic Conference, Am. Soc. Quality Control; and Reliability Training, Inst. Radio Engrs., 2nd Edition, Chapter II, March, 1960. (41) E. J. Gumbel, Statistics of Extremes, Columbia University Press,. New York, N. Y., 1958. (42) E. J. Gumbel, "Probability Tables for the Analysis of Extreme-Value Data," Applied Mathematics Series 22, Nat. Bureau Standards, July 6, 1953. (43) W. Weibull, "New Methods for Computing Parameters of Complete or Truncated Distributions," FFA Report 58, Aeronautical Research Inst. of Sweden, February, 1955.

Copyright by ASTM Int'l (all rights reserved); Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. No further reproductions authori

73) Correlation between two variables. Table 16 (p. 22) Group. 4 Computation of significance tests. Table 38 (p. pp. 22 Areas of the Normal curve. distribution. Table 11 (p. 38) Limits. sample. Table 27 (p. 75). 27. p. p. 14). p. 6 X .astm. 76) Fatigue. Table 19 (p. p. 5 Distribution shape. confidence. 61) Arithmetic mean. Appendix III Increasing amplitude tests.STP91A-EB/Feb. 4 Interval. 34. Estimate. p. 28) Fatigue tests. p. Table 17 (p. 22 Fatigue life. 72). 27. further reproductions Mean. 4. Table Confidence interval. Appendix III (p. pp. 6 (p. 8 (p. Table 20 (p. p. 65) Confidence limits (see confidence interval) F-ratio test. 14 Construction of Weibull probability paper from Gaussian distribution curve. p. confidence limits for. p. 2 Interval estimate. 13 (p. 30). 1. p. p. 4 log-log paper. 21. 13. 51) 46) Computation of standard deviation of values Fatigue strength for p per cent survival at N cycles. 6 (p. 27) Estimate. 47 Constant amplitude tests. p. Table 1 (p. 11) Analysis of fatigue data. 14 (p. 5. Downloaded/printed by Fig. 27 Estimate. pp. p. 9-13 Frequency distribution. 2 Fatigue data (see analysis of).by 4. 45 Difference between two standard deviations. p. p. pp. p. 27). 22. percentiles of. tolerance. pp. pursuant to Licenseinterval Agreement. 4 Fatigue strength. 6 Difference among k means. 69) Allocation of test specimens—probit method. 76) Estimation. 1. 5 Definitions relating to fatigue tests and test ^-factors for S-N curves. 28 X2 distribution. 5 Mean fatigue life. rights 60). p. 4 (p. p. Table 39 (p. Fig. 2. p. Table 33 (p. Table 13 (p. 2 curve. 76 Estimates. 5 F-distribution. 5 Limits.f. 6 Level. Fig. 6 about fitted line. 5 (p. p. Table 30 Tue Copyright reserved). p. p. p. Table 10 (p. 28. Table 10 (p. 20.ASTM TableInt'l 27 (all (p. Fig. p. single stress level. p. 6 method of least squares. 5.org . p. 5 Distribution. Table 36 (p. 15) (p. pp. Table 37 (p. 38 pp. 75). 6 (p. significance. 18. 81 Copyright© 1964 by ASTM International www. pp. p. p.del 12 (p. pp. p. Jul 9 10:11:02 EDT 2013 Maximum stress. 36) Fatigue test data. Fig. definition. 21 Abbreviations. 9 Definitions. 5 Mean. Fig. 22) Distribution curves. Fig. p. fatigue strength at N cycles. method for fitting a response methods. 5 fatigue data.P-32 Fatigue limit. Table 28 (p. point. 63). 1964 INDEX Estimates of parameters. p. method of computing.No p. 19 Escuela Politcnica Ejercito Ecuador ESPE Mean. 5 Average. 22) confidence for. choice of. Table 30 (p. 5 31). 4 Constant life fatigue diagram. 45. Weibull distribution function parameters. p. p. 6. 1 Confidence level. 14 (p. 40. p. 26). 34) Fatigue notch factor. Confidence coefficient. 77) 9 (p. 3 Choice of distribution shape. 6. p. pp. 5 X2/d. 60) 2 Fatigue life for p per cent survival. mean rank—Weibull distribution function. p. Table 32 (p. Limits. 74) Estimation. p. 3 (p. p. 44 Limits. 7 Analysis of correlation between two variables. percentiles of. Table 16 (p. p. Fig. 39 Fatigue life for a stated value of per cent sur63) vival. 67) Least squares. confidence. 65). 34) Definitions relating to statistical analysis. p. 42. Fig. Fatigue notch sensitivity. 47 Level. 4 Definitions relating to statistical analysis of Level. p. interval (see confidence interval) This standard is for EDUCATIONAL USE ONLY. p. 40 Mean. 26. tolerance. Table 12 (p. Table 32 (p. 40 Computations for fitting a response curve by Fatigue limit for p per cent survival. 5 Difference between two means. p.en Fig. sample. 69) Cycle ratio.

19). 34) Response or survival tests. 30. 12. 5 Sample means. 61) Normal distribution curve. p. confidence limits for. 48. 35). 26-29 Population. p. Table 25 (p. 22) (p. 23.Sample average. 13. Table 25 (p. pp. 5 (p. 64) Nominal stress. Table 28 (p. 10 (p. 11) Median percentage of survivors for the popula. 5. 6 Fig. differences between two. 12. 29 Method of least squares. Table 7 S-N diagrams. 10. 2013 Test minimum p. p. 16. 6 (p. pp. Table 2 Range of stress. 6. Fig. pp. 74) "Quadrant sum. 14 (p. 15 (p. p. p. 5 Mo. 45. R. 44-46. 79 Mean. p. R. 17. 26. 36. p. 1 (p. 19. 1. p. p. S-N curve for 50 per cent survival. 40. 20. 5 Normal distribution of fatigue hie. 1. 25 Statistic. p. pp. Table 33 (p. 26) (p. distribution. 3 Probit test method. 69) Significance level. 18. Fig. 2 (p. 50). Fig. 3 Step test method. Table 36 (p. 43 Run test. Table 16 N2. p. 3 Staircase test method. Fig. 4 (p.QTS for runs among elements in sam. 8 Standard tests. p. 8 (p. 5 Minimum stress. differences of group. 24). p. 21 dian of low ranking points. 33. 1. 49 Sample standard deviation. 40. 7.Rotating beam tests. 13. Table Means. pp. 1. 70) Selection of test specimens. p. 50. 50) Median. 3 Percentiles of the x2/d. 4 (p. 46) 1 (p.f. 69) Significant. Table 34 (p. 3 60) Stress amplitude. 16 Prot test. 56). 35. 6." Table 34 Normal curve. 15. pp. pp. 5 Moore rotating beam step test. 4 Stress ratio. 47 26 (p. 16-21. Fig. 10) Survival tests. 38. Standard deviation. 76). ^-factors for S-N curves. 53) Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. Fig. 51). p. 25. Table 2 (p. Table 31 (p.Standard deviation. 11) Probability paper. 10. 3 (p. p. 67) Number of test specimens. 3 (p-19) Means. Table 21 tion. 12) (p. pp. confidence Step tests. 20). 46. Table 4 (p. Table 5 (p. p. 64) Sample. 6 S-N curve for p per cent survival. p. p. p. 13. Fig. 45 Rank test. limits for. p. Table 27 (p.Scatter diagram. Table 21 (p. Downloaded/printed by Test specimens. p. Table 8 (p. 58) Mean stress. 18. selection of.69) Table 36 (p. 9 This standard is for EDUCATIONAL USE ONLY. 69) Means. Moore rotating beam specimens. confidence interval for. Table 15 (p. 2 Significance level. 56). 51) Per cent survival for a stated value of fatigue Steady component of stress. 21) Parameter. 52). Table 22 (p. Table 31 (p. 17. p. 7 Probit test—allocation of test specimens. 68) Runs along elements in samples of sizes NI and Medians. Weibull. 16 (p. 1. 47. Jul 9 10:11:02 EDT number. 29. 37 Table 21 (p. Fig. 42 cated fatigue strength at 108 cycles. Fig. Fig. 5 Table 22 (p. 41. 20). 23. 11 (p. areas of. p. pp.11) Probit test data. 18. 19. 42. 31. Tue specimens. Table (p. p. Stress. 50). p. 33. Table 3 (p. 7 Median. Table 16 (p. Fig. 34. p. 6. 27 Per cent survival values at N cycles. (p. pp. 6. p. 7 Table 33 (p. pp. pp. 4 (p. 34) Minimum per cent of population exceeding me. Table 20 (p. p. 34. pp. 39 Normal distribution. 21). 13. Fig. Fig. 56) Sample median. ASTM 52. 77) 19) Per cent of specimens having at least the indi. 4 Stress cycle. 7. pp. Table 30 Stress concentration factor. Table 34 (p. 4. Fig. Appendix II (p. pp. Median fatigue life. No further reproduc . 12). pp. Fig.025 and MO. 3 Modified staircase test method. 51) Percentiles of the x2 distribution. 14T. pp. 19 ples of sizes NI and N%. 48. 19). 73 Symbols. by 15. p. 34) Test procedures. R. p. minimum. 24) (p. 9. 25. 63) Stress cycles endured. S-N curves. 4 (p. p. 8. p. 3 Point estim_ate. Sample percentage. Table 9 (p. p. sample. 2 (p. 67) Table 2 (p. 9. Fig. p. Moore. 3 Probability-stress-cycle curve. p. 26 Median fatigue strength at N cycles. pp. 77) Response curves. p. 2 (p.82 FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA Mean rank estimates: per cent of population "Quadrant sum" correlation test. confidence interval for. Table 8 (p. for "quadrant sum. 14) life. p. 1. Table Test of significance. differences among k. 30. 3 References. Appendix III failed corresponding to failure order in sample. Table 25 (p. 51) Sample variance. 15) Copyright Int'l23 (all rights reserved). 16 (p. Table 6 (p." working significance level. p. pp. 74). 31. R.

p. 77). 62) Variable component of stress. 41. 5. 6 Theoretical stress concentration factor. p.INDEX Test-statistic. 3 Weibull distribution. Footnote p. 78 Wohler test method. p. 42 Tolerance level. Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY. 5 2-test. p. p. pp. Fig. 15 (p. 62) 83 Universe. No further reproductions au . 71 Weibull mean. Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. p. 58) Values of t. p. 4 Unpaired rank test. Table 26 (p. Table 29 (p. 30. Table 29 (p. p. p. 1 Copyright by ASTM Int'l (all rights reserved). p. 5 Tolerance limits. 4 Tolerance interval. 45 /-values.

No further reproduc . Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by This standard is for EDUCATIONAL USE ONLY.This page intentionally left blank Copyright by ASTM Int'l (all rights reserved). Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement.

Copyright by ASTM Int'l (all rights reserved). reports. Tue Jul 9 10:11:02 EDT 2013 This standard is for EDUCATIONAL USE ONLY. Much of the data result from the voluntary contributions of many of the country's leading technical authorities from industry. Over the years the Society has published many technical symposiums. and special books.THIS PUBLICATION is one of many issued by the American Society for Testing and Materials in connection with its work of promoting knowledge of the properties of materials and developing standard specifications and tests for materials. These may consist of a series of technical papers. A list of ASTM publications and information on the work of the Society will be furnished on request. No further reproductions authorized. Downloaded/printed by Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. or compilations of data developed in special Society groups with many organizations cooperating. scientific agencies. reports by the ASTM technical committees. . and government.

This standard is for EDUCATIONAL USE ONLY. No further reproductions au . Tue Jul 9 10:11:02 EDT 2013 Downloaded/printed by Escuela Politcnica del Ejercito en Ecuador ESPE pursuant to License Agreement. Copyright by ASTM Int'l (all rights reserved).