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

Uploaded by gudemaranahalli

0.0 (0)

By WALODDI WEIBULL,I STOCKHOLM, SWEDEN This paper discusses the applicability of statistics to a wide field of problems, Exa mples of simple and complex distributions are given. Fa variable X is attributed to the individuals of a population, the distribution function (df) of X, denoted F(x), may be defined as the number of all individuals having an X ~ x, divided by the total Dumber of individuals. This function also gives the probability P of choosing at random an individual having a value of X equal to or less than z, and thus we have

P(X

I

sary general condition this function has to satisfy is to be a positive, nondecreasing function, vanishing at a value Xu, which is not of necessity equal to zero. The most simple function satisfying this condition is

(x xu)m

:1'0

**and thus we put
**

F(x) = 1- e

x,

S x)

= 1-

= F(x)

[1]

...............

[51

**AJiy distribution function may be written in the form
**

F(x)

e-<P(x)

[2]

**This seems to be a complication, but the advantage of this formal transformation depends on the relationship
**

(1 P)"

= e-""'(z) .........•.•..••.

[3]

The merits of this formula will be demonstrated on a simple problem. Assume that we have a chain consisting of several links. If we have found by testing, the probability of failure P at any load x applied to a "single" link, andif we want to find the probability of failure P; of a chain consisting of n links, we have to base our deductions upon the proposition that the chain as a whole has failed, if anyone of its parts has failed. Accordingly, the probability of nonfailure of the chain, (1 - Pn), is equal to the probability of the simultaneous nonfailure of all the links. ThUB we have (1- Pn) = (l-P)". If then thedf of a single link takes the form Equation [2], we obtain P; = 1e-""'(z)

[4]

Equation (4) gives the appropriate mathematical expression for the principle of the weakest link in the chain, or, more generally, for the size effect on failures in solids. The same method of reasoning may be applied to the large group of problems, where the occurrence of an event in any part of an object may be said to have oocurred in the object as a whole, e.g., the phenomena of yield limits, statical or dynamical strengths, electrical insulation breakdowns, life of electric bulbs, or even death of man, as the probability of surviving depends on the probability of not having died from many different causes. Now we have to specify the function ",(x). The only necesI Professor at the Royal Institute of Technology; ScientificAdviser to AB Bofors; mailing address Bofors, Sweden. Contributed by the Applied Mechanics Division for presentation, by title, at the Annual Meeting, Atlantic City, N. J., November 25-

The only merit of this df is to be found in the fact that it is the simplest mathematical expression of the appropriate form, Equation [21,which satisfies the necessary general conditions. Experience has shown that, in many cases, it fits the observations better than other known distribution functions. The objection has been stated that this- distribution function has no theoretical basis. But in so far as the author understands, there are-with very few exceptions-the same objections against all other df, applied to real populations from natural or biological fields, at least in so far as the theoretical basis has anything to do with the population in question. Furthermore, it is utterly hopeless to expect a theoretical basis for distribution functions of random variables such as strength properties of materials or of machine parts or particle sizes, the "particles" being fly ash, Cyrtoideae, or even adult males, horn in the British Isles. It is believed that in such cases the only practicable way of progressing is to choose a simple function, test it empirically, and stick to it as long as none better has been found. In accordance with this program the df Equation [5), has been applied not only to populations, for which it was originally intended, but also to populations from widely different fields, and, in many cases, with quite satisfactory results. The author has never been of the opinion that this function is always valid. On the contrary, he very much doubts the sense of speaking of the "correct" distribution function, just as there is no meaning in asking for the correct strength values of an SAE steel, depending as it does, not only on the material itself, but also upon the manufacturer and many other factors. In most cases, it is hoped that these factors will influence only the parameters. However, accidentally they may even affect the function itself. The purpose of this paper has been to illustrate with a few examples the experience that the df, Equation [5), may sometimes render good service. The number of examples has, by space, been limited to the following: 1 2 3 4 5 Yield strength of a Bofors steel Size distribution of fly ash Fiber strength of Indian cotton Length of Cyrt.oideae Fatigue life of a St-37 steel

Discussion of this paper should be addressed to the Secretary, ASME 29 West 39th Street, New York, N. Y., and will he accepted until October 10, 1951, for publication at a later date. DiSCUSSIOn received after the closing date will be returned. NOTE: Statements and opinions advanced in papers are to be understood as individual expressions of their authors, and not those of the Society. Manuscript received by ASME Applied Mechanics Division on January 15, 1951. Paper No. M-A-6. 293

30, 1951, 01

THE

AMERICAN

SOCIETY

OF MECHANICAL

ENGINEERS.

In the Appendix: 6 Statures for adult males, born in the British IslE'S 7 Breadth of beans of Phaseolus Vulgaris

3 = 9.6-2 __ l FIG. 1 /.ash N. 5. Then X' = 8. 3 gives the curve. x.831284 values. N.40. Koshal and A. 2 SIZE DISTRlBUTION OF FLY ASH Observed values 3 14 34 56 85 126 150 175 188 197 202 208 ~11 = x 2 3 4 5 6 7 8 9 10 11 12 13 14 particle diameter in 20 microns) Expected values 14 34 62 92 122 150 172 188 199 205 209 211 "3 " . the d of fare 8 . If the classes 2-3 and 13-14 are pooled. New York. In this equation y represents the frequency of any strength expressed in grams.008. The authors. Koshal and A. 2 SI~E DISTRIBUTION OF FLY AsH I . 325-370.f-.40 gives P = 0. The parameters are x~ = 38.59 gram..73 grams.. observed and calculated. having the following equation y = 599. J.74 kg/mm2 f----- . YIELD STRENGTH OF A BOFORS STEEL The observed values are taken from J.934. It seems obvious that the components of examples 4 and 5 are due to real causes. Then X· = 5.275 kg/mm') Expected Observed values values :z. Dalla Valle.0-. the values expected on the hypothesis of a normal distribution have been computed and are given in the last column of Table 1. J.+ ~ log(. showing a wellmarked predominance of weak fibers.57 kg/rnm".0-0 of Fly.45 gives a P = 0. Turner. n 32 10 10 33 36 33 34 84 81 35 150 161 36 224 224 291 37 289 340 38 336 369 369 39 383 40 383 42 389 389 .43 gives a P = 0. 2. Fig. However." by R. It may be said that any distribution may be represented by a sum of a sufficiently great number of simple distributions. N. p. Journal of the Textile Institute Transactions.2. and Table 3 the values./ ~6 Xv" Xo. TABLE (:z. In examples 6 and 7 it is impossible to decide whether the division is a formal one or a real one.49. The specifie data for the examples follow.876Tl6 ( . m = 2.J OF A i The observed values are taken from R.07. vol. M. Dalla Valle's work..2- ! '" V .~ I YIELD STRENGTH BOFORS + -18. • "Studies in the Sampling of Cotton for the Determination of Fiber Properties.1947 ~- X )13. i-6 p • 0.35. whereas the four remaining populations have to be split up into two components. the d of fare 13 -3 = 10. In any case. 38. If the classes 14 to Hj are pooled. if the number of the components be small and the number of observations sufficiently large.29. Without pooling. Pitman Publishing Corporation. The agreement is thus very satisfactory.777 )O. As a comparison. = 3. .'' by J. the d of f are 12 . 1930.-xu} m· 29342 x. The fundamental question now arises.74 kg zmmt. or whether it might unveil Borne hidden real causes. To = 7. 57.3 X (1 .44 gives a P = 0. the likelihood of real causes seems to increase. The first type will be called a "simple" and the second type a "complex" distribution.456.1 - 29. The parameters are X~ = 30IL. S.294 JOURNAL OF APPLIED MECHANICS SIZE DISTRIBUTION SEPTEMBER.2 = 6. If the classes 9-10 are pooled.have pointed out that the most striking feature about the frequency curve is its asymmetry.! Fig.57 kg/mm2 7.195] OF FLY ASH The correctness of fit has been checked by applying the chisquare method Of those populations. 2 gives the curve and Table 2 the values.3 = 6.8-2- r _!!'o--+--.?-o. 3 Fig. the d of fare 7 and X' = 8. Then X' = 14. whether this splittingup is a purely formal operation. Then X' = 11.Lf"---I " FIG.- FIBER STRENGTH OF INDIAN COTTON . which is not satisfactory at all. S. The d of fare 13 . V . m = 1. Nos. = 0. 1948. pp. Fig. M. 21.44 gives a P = 0. Then a X' = 18. Xo = 128 IL. '''Micromeritics.8-. before such an agreement is obtained.211 1 2 3 4 5 7 8 9 10 G (x ~ yield strength in 1.5 = 8 (as there are 5 parameters).4·0 The observed values are obtained as routine tests of a Bofors steel.2-1-1l: -+--+--'(.. Y.~ ·ll .49. It was found-they saythat the observation curve would be well fitted by a theoretical curve of Pearson's Type 1. x. the quality of which was chosen at random for purposes of demonstration only. m = 2. but the fact itself may be a valuable stimulus to a closer examination of the observed material. Turner. The parameters are x.17 gives a P = 0. it is very easy to produce real complex distributions by syntheses. just as any periodical function may be developed in a Fourier series. 1 gives the curve and Table 1 the TABLE 1 YIELD STRENGTH OF A BOFORS STEEL Normal distribution n 8 28 71 141 225 301 351 376 386 388 .:+--'-----'--i . Without pooling. 1-3 are distributed in good agreement with the df Equation [5 J.4-'l! .389 I I I of a Yield strength Bofors Steel ~ /' l/'8 .49 11° STEEL .288. the degrees of freedom (d of f) are 9 . The values computed from this not very handy equation are shown in the last column of Table 3.

035 The observed values are taken from Muller-Stock.1: Xu = 3.5 67. About 20 per cent of the populations showed a simple distribution.-T0~. 55 gives DO impression of a complex distribution.WElBULL-A TABLE 3 FIBER (x = STATISTICAL OF INDIAN DISTRIBUTIO~ COTTON Pearson Type 1 n FUNCTION OF WIDE APPLICABILITY 295 STRENGTH tensilestrength in grams) Observed Expected values values n In spite of the greater number of parameters. TABLE 5 FATIGUE LIFE OF ST-37 (Rotating-beam test at =32 kgZmm t) N 10' 17.9 165.020 (jOJu - I 0...9 235.1938)." by W: Weibull.3 = 4. reproduced in Fig. 9 165'._4 __ ~ ~ FIG. ~_UL-5 I 1 1\ II \ ~ ~ 0. LENGTH OF CYRTOIDEAE :r n 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1232 1751 2161 2461 2802 2886 2937 2966 2982 2991 2996 2999 3000 Z067 646 118 177 667 1219 1729 2153 2465 2664 2813 2887 2933 2962 2985 2991 2995 2999 3000 1255 1777 2184 2480 2683 2816 2899 2949 2978 2994 3003 3007 3009 3010 659 127 r· 1 40 ' x2 i 1 log(·-·u) . (March.0 165. the fit of this distribution function is not as close as that of the first one.9 233. The measurements were made by Dr. vol.4 47. This is the first example of a complex distribution.0 222.5 215.0 69.1 169.6 233.097 3. on the other hand. The parameters are: Componen t No.5 92. and that it is necessary to split up the population in two parts.5 22. i:: 10 P . o FIG.0 70.5 ----Expected \'all1es-n.0 226.2 )1.080 0.3 4. 0 216. m nl+2 4. and P = 0.45.010 0.2 165. Nature.1 125.0 61.2 225.. 4 gives the curves and Table 4 the values of one of the complex populations. MUller-Stock.9 232. t .7 ·2-0 Length I '-. Xo = 124.59.6 165.1 )1.0 235.4~7-. 1047.0 JI.1+1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 100 80 60 40 82.47. LENGTH OF CYRTOIDEAE Observed values = length in microns) ". ~ __ I 22 7-3 20 I log (. Fig.9 228. FATIGUE LIFE OF AN ST-37 STEEL =- 11.--T-IT-~T I (\ 8-1 2.479.7 228.0 65.5 233.7 230.75)1.5 231.1.6 47. corresponding to an age interval of about 100.10 5 LIFE OF ST-37 STEEL FREQUENCY CURVE OF FATIGUE 4 LENGTH OF CYRTOIDEAE (Number of specimens versus number of stress cycles. schwingender uberbeanspruchung auf die Entwicklung des Dauerbruchs. Jungk. n.0 66. "A Statistical Analysis of the Size of Cyrtoideae in Albatross Cores From the East Pacific Ocean.9 165. + Nz• It is easy to see that the distribution is a complex one.0 57.0 51.040 0.0 63. nl+ 2 1 5 5 13 13 23 23 35 35 47 47 58 58 67 67 74 74 79 79 82 82 85 85 86 86 90 4 86 93 86 7 86 9 95 86 11 97 12 86 98 13 86 99 86 14 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 z 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 m. 13.100.0 165. taking samples from each 10 em of the core.6 230.chung.1 169. The observed values have been obtained from investigations by Dr..070 0. on submarine cores from the Swedish Deep-Sea Expedition With Albatross. p..000 years.---Expected val uee-r-r-r-rn.m = 2.050 0. Mitteilungen Koble.090 0. and 11-13 gives X' = 3.0 207.S 32.6 4.' The remaining samples showed a twocomponent distribution.und Eisen/or.2 192.7 165. The number of individuals is too small for the X2_t<!St. Gustaf Arrhenius.5 FIG.0 0 5 12 24 38 45 58 69 70 80 82 84 86 90 93 95 97 98 99 100 Observed values 71.3 164 9 28.3 .4 125.5 47. Component 1\0.5 42.0 192.6 165. Xo = 63. Some fifty populations have been analyzed statistically.1 234.4 125.9 232.! The frequency curve in Fig.9 165.S 52.9 231. by measurements from his Fig.1 161 2 8. • "Der Einfiuss dauernd und unterbrochen wirkender.5 37.5 57." by H.'u) ___12_____1.9 41.6 47. 1949.0 67. Pooling the classes 2-3. 9-10.5 87. which.1 165. The undivided sample gives the curve marked N. 164. .0 206.9 222. as exemplified in a previous paper. 2: x" = 122.59 2_2-t~~~ __ +-__~ __-+~P_. and 14 to component No. By trial it was found that 86 of the individuals belonged to component No.097.m = 1.5 77. 5 of this paper..5 72.2.5 62.0 68. C.060 I -i Ii I I l i i 0.) . may easily be seen when COTTON 3 FIBER STRENGTH OF I~DIAN TABLE 4 (. The d of fare 7 -.5 27.

giving a P = 0..2: Xu = +7. making 70 per cent.95 rum).032 m• 5. the following two are given with the tables only: STATURES FOR ADULT ~fALES BORN IN THE BRITISH ISLES FIG. we obtain 14 specimens out of 20. The real causes of this splitting up in two components may be found by examining the frequency curve of the yield strength of the same material. Lippincott Company. quoted from Charlier" to exemplify the expansion in Edgcworth's series. oI 22 7 1 / \I \ ~ 26 \.484.3671 11571 7900 3908 11808 7900 4022 11922 7900 4071 11971 40P] 7900 11991 7900 4098 11998 7900 4100 12000 Observed values "1+1 32 135 374 998 2185 3835 5718 7648 9286 10416 11153 11580 11801 11911 11968 11992 11998 12000 6 5 I 3 2 j / ( 1\ \ \ r.20. m = 1. The reason why this partition is so easily seen in Fig. m = 6. is composed of two different kinds. quoted by Harald Cramer in his book:' Mathematical Methods of Statistics. which gives a P The d of fare 8 = 0.11. each of them may be very well fitted to a simple distribution function with the following parameters: Component No.6865. If the population is divided into two parts.)EPTEMBER. Charlier. 5 seems to be the result of a smoothing operation on the cumulative frequency curve. I1l+ :2 nl+ ! 2 2 2 6 6 6 20 20 20 56 56 61 143 144 143 333 313 333 702 702 707 1350 1350 1376 2351 2351 2366 3641 3641 3589 1917 4917 4918 6126 5787 339 6148 7213 7211 6134 1079 6197 1671 7868 7857 6200 2039 8239 8249 2233 6200 8433 8451 6200 2324 8524 8530 6200 2363 8563 8562 2378 6200 8578 8578 6200 2383 8583 8583 2385 8585 6200 8585 n. = 4.31. 7. = 0. 2: Xv = 4.235 life r··o-I Xu· 4.·I0800 Component .• Xo = 2. V..2 (= 8. H)51 using the plottings in Fig. 1945." by C. = 81/2. 94 and 111.032. as 165/235 = 70 per cent. 2 Nu·30S00 m·1. 1937. Fig.2 )(U·--4·494. .\TlJRES . Component Ko. B. Without pooling we have the x. m = 1.215. we get X2 = 11.3 in. m = 5.) 29 STEEL FREQUENCY CURVE OF YIELD STRENGTH OF ST-37 (Number of specimens versus yield strength The observed values are taken from Yule and Kendall.1. By trial it was found that the population had to be split up into two parts: N.215 z 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 FOR ADuLT MALES BORN IN THE BRITISH ISLES (z = height in inches) Observed -----Expected val ues-----~ values n. which gives a P = 0.80. BREADTH OF BEANS OF PHASEOLUS VULGARIS This is a classical example. 7 "Die Grundlagen der Mathematischen Statistik. X' = 5. G.3' /. n. eleventh edition. Table 5 shows the close agreement between the observed and the calculated values. = 7.2: Xu = 67. N] = 7900 and N. second edition.2 in. of course. Yule and M.p. upon the much larger scatter in fatigue life than in yield strength. 1'0 = 16." Princeton University Press. L.70 moo) It may be pointed out that the frequency curve in Fig.0 (= 5.6098. If we suppose that all the specimens with a yield strength of less than 25 kg /mm! belong to Component No. N." This distribution is classified by the authors as being approximately of the symmetrical type. and there is no mention of its being composed of two parts. If the classes 1-2 and 14-15 are pooled. • "An Introduction to the Theory of Statistics. If the classes 20--21 arc pooled._. The d of f are then 12 . p. \. As we have 7 parameters altogether.296 JOURNAL OF APPLIED MECHANICS TABLE6 ST. U. 7 and not at all in Fig.xu) ·+·1 FIG.25:r: VULGARIS + 6. Exactly the same proportion has been found by the statistical analysis.4 in . pp.. log (x ." by G. = 6200 and N1 = 2385. 0·0 Fatigue N. 73. Pa .' = 6'/.2 5718 5718 7140 486 7626 7761 1525 9286 7890 2510 10400 7900 3229 11129 7900 . probably not being killed. r--------r--------r--------r--------.956. 31/. J.0 in.6 FATIGUE LIFE 6-1 OF ST-37 STEEL ·8-1 TABLE (Breadth :r: 7 PHASEOLUS of bean. It is easy to see that the material. Philadelphia. Kendall. = 4100. J.70. m = 0. nl+ t 32 32 130 130 400 400 1011 ion 2145 2145 3832 383.1: x" = 50. depends. The parameters are: Component No.~ 27 28 In kg/moo".4662. Component No. and the d of f 10 . 5.-------. (one of them is the partition of the population). 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 -----Expected value5---~ n. 440.1: Xu = -3.956 J Nu. the sampling errors of the observed values in Table 5 have been eliminated almost entirely (without affecting the function).. and for this reason. m = 1.29. It may perhaps be of interest to have examples of this kind. = 4'10. which explains the really too good representation of the observed values.2805.1: x.. we take 3' /2 to each of the components. The parameters are as follows: Component No. Component No..35.~------_. 1920. 6. Princeton. Accordingly._ ?5 Appendix The foregoing statistical methods have been applied to many problems outside the field of applied mechanics.50 mrn )..

5 d or f 13 P < 0. Discussion of his paper was reported in the ASME Journal of Applied Mechanics. the agreement with the Rormai di.t.34. Wallodi Weibull published "A Statistical Distribution Function of Wide Applicability" in the ASME Journal of Applied Mecbanics.001 Firat approximation x' . operation il certainly of a purely formal character.re this reault with thOlJe of Charlier and Cramm-. 297 The If the cl_ 17-18 a. . but that a clOlJer examination sbow.50. September 1951. tive kurtoai•.56.hu calculated the valuell of X· on the bypoth_ of Normal dilltribution x' . pages 233234 as described on the following pages. pages 293-297 as described above.WEIBULL-A STATISTICAL DI8TRIBUTION FUNCTION OF WIDE APPLICABILITY normal distribution and uymptotic I'9IIUlt wu &8 follow8: exp&lllioni from it.ribution IMInI very I&tiafactory. Transactions of the American Society Of Mechanical Engineers.0. and the doff 9 .196.19 The agreement is eatiafactory in the third caee only.S d of f 12 P < 0.5'/.re pooled. a rnnall negative akewn_ and a I1D&11 poai. Cramm.9 d of f 11 P . . at the fim look. Transactions of the American Society Of Mechanical Engineers.3'/. June 1952.0.4. Thi. Charlier eay. that. requirina: four ~ Df the Jeriea.001 Second approlEimation x' a. . It lI1&y be of interest to compa. give a P . the value of X.

. Trans. 13171324. DO. Tbe pa.. than the minimum. This haa been illustrated ill the case of the lognormal distribution by Vim Uvens and more recently by Mugele and Evalls. Quarterlll of Applud Mathematics. f _ dF _ '!: (x - exp [. Calif. deulach.rning 1\00 should be Added in regard to the examples in the paper: They contain some arithmetical and dimensional errors. as evidenoed by tbe seven examples presented in his paper. and to the Rosiu-Ramruler type of equation' as x~ vanishes. . and 3 in the paper.x. Would he care to dlecuss this point briefly? 2 When the funotion i8 applied to an unknown distribution..)"'J. Tsu.e JO.. pp. in the second ex. n. Mem.. to represent the equation which will again reduce to Equation (5: &8 x . In order t.. one would expect the maximum particle to be more tangible. " by M. Su-te Colle. xo)"-' ~o .. the discrepancy being appreeiable when x is small. D.Applicahiiity ' T. .. vol.rameter~. The range of fields treated in his examplel! is also impressive. ma...i«ure. and m. v.. .. and ita implication of a lower limit to a distribution. • "Feinheit und Rtruktur des KohlenstAub8..(':: x..o apply the dilltribution impartially. Akad. should be used. published in the September. In thi" connection he would like to ask the following qUestions. some systematic means of fitting it to experinlental data. Cambridll". chu . ASME. becomes infinite. the ease Ylith which it can be applied to studying the aize effect.. :1:.yivBlliB • Auociate Profeaoor of Encineerinc Research. Now. [1) gives the frequency dlstributfon x.. ~usele and H. Of course one may start with any distribution function where the argument has infinite range. Rosin and E." by D.ry. However. A word of wa.' The distribution Iunetdon suggested by the author is attractive because of it3 simplicity..) in Fig. 'Asoistax>t Prol . Van Uven. issue of tb.ns. Could there be BOrne numerical errors? If 80. pp. If the distribution function is 8. Kon. vol.. Ev .on of the observcd data. the reason for introducing the minimum value X~. vol. State CoUese. appear 1. is not entirely clear.. A.. ho.. F.. :1:. vol.. however. 1961.. 381-389. IndU81rial"nd Knqinurinq Cloemut'1l. J. pp." by P.. 2 of the paper..r Inlle . Since the writer is currently concerned with the problems of particJe.fUGELE. .' On changing the variable or integration to The author's treatment is definitely a con- x-xo)" ( ----r. tte Inatitute of Technology.· The a.meten :1:. for such a case as Fig. Prager. C. .. z" [2) rather than _ (Z-Zu)'" P-I-e . .." by R. true representat. However.ample (size distribution of fly ash).i.. J. In the author's examples he did no~ specify how his parameters were obtained... 1927. be is interested in the posBible utilization of the author's method to reduce the necessary lIoIllount of experimental work. 670.. 2 do not correspond to the given values of :t and x. In applying the author's distribution function it is nec_ry the para. """'7 293-291.' The latter reference also gives a oritical review of the Rosin-Rammler and other distribution Iunctioae. Greenberg. Wete. The Penn.DISCUSSION 233 tributioD to the literature on distribution functions.uthor should be congratulated for having devised a distribution function of truly wide applicability. 9. ZeilBchri/t d •• V..)" ""-' A Statistical Distribution Function of Wide . vol... 73. 2. Drucker. For a simple distribution the following procedure appears useful. M......r of Mechanical EncineeMnIi.. 19.ny observed data arc neceeeary to yield the parameters reliably? Considerable practical value would be added to the author'a function if ite application could result in a saving of experimental work.size distribution in aerosols. Probably it relates to the original applications. • "Droplet Si"" Distribution in Spray s. and ignoring the maximum x .. A. Rammler. 1952. . and convert it to one where the range is finite... 43. Proe. !'. J&tlua...<"ig8. 3 The relatione shown in J. when these are corrected. pp.(:=_==l')"J .0 Could that be a misprint? The values for log (x . Ineorporation of both a maximum and a minimum value of x will bring Equation [5 J into the form F(x) = l-e _k(~-Z. The nth moment of the thoorotical distribution is fust calculated from the ClImulative distribution F _ 1Differentiation dx to determine exp[... • Oakland. and H. the examples illustrate excellently the general statements of the text. W.. ASME. would the author kindly MOW a corrected figure? R... Ske" Frequency Curv . C. 1917.i"".. •By Waloddi Weibull. ich may have been the h Cystoidea or the yield strengths and fatigue-life data of steels.."UtAL 0'" AF'L1tD MBCBA. Pa.. Zo. and fit can be chosen so that the Dr8t three moments of the distribution function coincide with those of· the data. A. Jun MME. • "Extended Limit Design Criteria fOT Continuous Medill. p..1-7.. MCCLINTOCK.>lIca. then any three sets of the values of P and :I: would be sufficient to evaluate these three parameters.. and also more significant practically. M . 4. 1951.

facilitates the otherwise tedioWl computa· tions. the value or a. 1 10 . As to the third question. It was not found necessary to introduce this new parameter in the field of strength of materials. J1. .. also including the standardized Gaussian distribution.90 .. r (1 + nlm) [5J The second and third moments about the mean are /l. The method proposed by Profeesor McClintock to use the first three moments is quite good if the distribution is simple and the population not too 1II1lII1l.. for the theoretical distribution and the experimental data coincide. By plotting . but admits that it may sometimes have its advantages. e) on this paper. W. I 001 ""Ix: OJ """ itWl r I I 1. M. published in the December.' Aotually.. The proposal of Professor Tau to take any three sets of the values P and x is quite correct but does not use the data efficiently. the relation p._ -\-fl- ..JOURNAL this beeomes OF APPLIED MECHA.. For example. Shaffe.· Mem.xo" ~ r( 1 can be solved for x.A8ME... Ill.'[r(l + 21m) + 3/m) - 1"0 + l/m)J aI'( I + 2/JIl)r(l + 11m) + 21" (1 + lim)J [6J From these a measure of the skewness can be obtained £lo..0 oz 0. however.. -1 X 20 . vol. by putting z ~ (:tx)/v. . I Profeeaor of Mf'Chanieal }~ngineering...'/z.. Univeroity of IlIinoj. 10LO 5 0 1.." by W"loddi Weibull.. that is... it is easy to decide wbether the distribution is simple or complex and to estimate. the square of the standard deviation. 1. ASME. As to the question of a systematic procedure when the distrlbution is complex..i -""""'r---.". I By E. but a simple electronic computing machine. the parenthesea are an awkward misprint. [8J Finally. • "The Phenomenon of Rupture in Solids. 163.1'/X. the relation {!r-:r. This method IIl&J' be improved by laking the IICt from II smoothed curve. of course. 2 The writer would like to ask what procedure. not very satisfactory. IV A Handlina. Mugele is a valuable extension of the function. 6 and 7 of the paper. II. he has never app1ied thiA method.. the value of m can be chosen 80 that the values of a. H. 5 E 2 r- I +1I"looz cos tr1/ iJ where a ~ 11m.. I . . but the procedure is not entirely free of subjectiveness. Thus the value x. may be prepared. and ro. No./J1. [9J can be solved for 20 since the mean of the experimental data is known.'.. Plots of the quantities £lo. where is the mean and v the standard deviation and elirninating two of the parameters.0 1:5.x.. and p..) in Fig.•. 'Assistant Professor of MeehanicQl Engineering..'. But in other fields conditions may be quite different.'.. Urbana. recently completed.9.'rr(l and Il8 .. The introduction of a maximum value z.T(I + 1/m) .e.. it would be interesting to see wbether the other data on the ST-37 steel reported by Muller-Stock would result in the same division of the population as found from Figs.0 Of.. the author is BOrry to admit that so far he has found no better method than to cut and try. = Zo" Jam = ". 2 do not correspond to the &iven value :r.. The values for Jog (x . 23. Up to the past year the author's usunl method has been to plot the data as shown in the paper and to choose the value "". x.. preferably systematic.. ~ . The author haa been aware of this possibility of computingtheparamete1'8and has mentioned it {with BOrne different notation for the gamma fUnetian) in an earlier paper.. + 21m) - 1'. 1951. Untveretty of U1inoill. .5 X 201' but to "'. /l. pp. probably because the theoretical strength may be perhaps a hundred times higher than the technical strength. i. with a good approximation.. In this way it is easy to decide if the distribution is simple or complex.1952 /l.( 1 + I/m) . issue of the JouRNAr..30 I'is the correct one.he points (P. is a function of III only. as ."\ICS ArTHOR's CLOSURE JUNE. It should be mentioned that the e-valuea are mid-point values and should correctly have boon increased by 1/..x.'/r 171 The author appreciates the comments made by the djscussers. r-.. should be followed in the case of a "complex" distribution. of the experimental data is known.. FIG..1. Thie ie. and yet the writer would like to try applying the distribution in other cases... . Trans.. proposed by Mr. Lee and B.)/x. About a year ago the author decided that it would be better to start by standardizing the variable x. A curve paper for different values of a. to give the best straight line..n 1J"/~el{p(-'1)d'l: [4J This integral can be expressed in termsof the Gamma Iuuetlon "'. 405-418. OP ApPLIED M"~'1I"NlC8.2. . An extension of the foregoing procodure looks impreetleal./z. ~ l'-.0 1?1O. The di . 13.. p. are given in Figs. tribution function then takes the form x Since a.o ~~ 2 i--- ~{-. (or instance. 1 and 2 of this discuBBion. Then since the second moment about the mean. p.

Sign up to vote on this title

UsefulNot usefulAre you sure?

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

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.