Application of Extreme Value Statistics To Corrosion

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Volume 99, Number 4, July-Augusl 1994

Journal of Research of the National Institute of Standards and Technology


[J, Res. Nati. Inst, Stand. Tcchnol, 99, 327 (1994)]

Application of Extreme Value Statistics


to Corrosion

Volume 99 Number 4 July August 1994

Toshio Shibata Application of extreme value stati'iiics analysis using extreme value statistics,
to corrosion is rcvieweJ. It is cmpha which were reported mainly in Japan,
sized that the concept of corroiiion arc shown. Accumulated experiences
Department of Materials Science probability is important for a quanlita- suggest thai an appropriate classifica-
and Processing, livc evaluatioti Of corrosion failure and tion of data based on the corroiiion
-Faculty of Engineering, its prediction, especially for localized mechanism is reijuircd iKfore applying
Osaka University, corrosion. Extreme value statistics is extreme value analysis.
quite useful for assessing the maximum
2-1 Yamadaoka, Suita 565 pit depth and/or the minimum time for
JAPAN crack generation. The maximum pit Key worAs: corrosion probability; fail-
depth depending on the stirface area ure time; Cumhel distribution; localized
can be evaluated hy using the Gumbel corrosion; maximum pit depth; MVLUE
distribution with the concept of return method; return period; Weibull distri-
period. A standardized procedure i.? bution;
proposed for estimating the maximum
pit depth. Examples of corrosion failure Accepted: March 22, 1994

1. Introduction
Development of extreme value statistics and its samples. In Japan, early review papers on corrosion
application to various fields, including corrosion, probability and extreme value statistics by Masuko
has been described by Gumbel in Ref, [1] and [2]. [9] and Shibata [10, 11] contributed to the study of
Evans is one of the pioneers of modern corrosion the extreme value statistics as applied to corrosion
science, and first established the concept of corro- problems. Ishikawa [12,13,14] and Imagawa
sion probability [3, 4]. Eldredge [5] used extreme [15,16,17] applied extreme value statistics to ana-
value statistics to obtain the maximum value of pit lyze engineering data. Kase [18, 19] reviewed
depth on an oil well tube wait as a function of tube Lieblein's paper [20], introducing MVLUE (mini-
surface area. Scott [6J found a logarithmic depen- mum variance linear unbiased estimator) method
dence of the maximum pit depth on surface area, for estimating the distribution parameters. Lieblein
and explained that dependence by referring to Trip- had given the coefficient of MVLUE up io N -d.
pet [7]. Aziz [8] and Eledredge [5] discussed almost Recently, Tsuge [21] had calculated the coefficients
all important points to be considered for the analy- up to A^ = 45, and confirmed that the parameters es-
sis of corrosion pit data and made use of the return timated by the MVLUE method are unbiased and
period concept. This concept, originally introduced efficient and are consistent with values estimated by
in the fields of hydrology or meteorology, is now the method of moments or maximum likelihood
used to obtain a size factor which makes it possible when the sample size exceed.s more than 20. The
to estimate the maximum pit depth in a large sur- committee of Japan Society of Corrosion Engineer-
face area based on the distribution of a small num- ing (JSCE) proposed a standard procedure [22] to
ber of pit depth data from the small surface area estimate the maximum pit depth from the small

327
Volume 99, Number 4, July-August 1994
Journal of Research of the National Institute of Standards and Technology

sample size data by using the MVLUE method, and Table 2. Examples of the extreme value analysis for corrosion
a computer program named EVAN [23] was devel- usirg the Weibull (exponential) distribution
oped. Recently Laycock et al. [24] reported that a
generalized extreme value distribution is more con- Example Rcf.
venient for corrosion depth analysis, because no Failure life analysis of stress corrosion [56]
preliminary assumption on the type of distribution cracking of stainless steel heat exchanger tubes
is needed. An introductory book [25] by Kowaka et Failure life distribution of stainless steels [57]
al. helped to differentiate extreme value analysis in high temperature and high pressure water
among corrosion workers in Japan. General back- Effect of CaCb concentration on SCC life lime [58]
distribution? of stainless steels
ground on extreme value statistics is provided by Evaluation of SCC failure life of stainless steel [59]
Ang and Tang [26], and Kinnison [27]. in high teraperaiure water

2. Application of the Extreme Value


Analysis to Corrosion 2,1 The Gumbel Distribution
In the early 1980s, meetings and symposia The procedure is proposed mainly for analyzing
[28,29,30,31] were held in Japan for discussing the pit depth distribution by using the Gumbel distribu-
basic principles of extreme value statistics as well as tion and the return period in order to estimate the
difficulties and problems in their application to cor- maximum depth of the larger surface area from
rosion. In Table 1, several topics for which the which small area specimens are sampled. The Gum-
Gumbel distribution is applied are listed. Table 2 bel distribution is expressed as
includes cases analyzed using the Weibull distribu-
tion, including the exponential distribution. Before F (x)^ exp (-exp { - (x-\ )/a). (1)
discussing case histories, the standard procedure
[22] proposed by the committee is briefly explained; where F(x) is the cumu1ati%'e probability of pit
details are available elsewhere [22, 32]. depth, X, and A and a are the location and scale
parameters. The reduced variate, y.
Table 1. Examples of the extreme value analysis for corrosion
using the Gumbel distribution
y=(jc-A)/a (2)
Example Ref.
is introduced, and then
Life prediction of super heater tubes of the power [49]
plant y ^-\u {-In {F(y))) (3)
Application of the extreme value analysis to [37]
heating tubes of the boiler
is used for constructing the Gumbel probability pa-
Estimation of the maximum amotmt of impurity [42]
segregation in steel per. Plotting position for the cumulative probability
4. Failure life estimation of SCC for Ni base alien's[41] can be calculated simply by
5. Extreme value analysis of
uf the corrosion depth of [35]
the oil tank plate [36] F(y) = 1-*7(1+N). (4)
[50]
6. Life prediction of heat exchanger tubes [51]
7. Eddy current examination system for heat exchanger [43] where / is the jth of the ordered value, x, in de-
lubes with the extreme value analysis scending order and ^V is the total number of sample.
8. Extreme value analysis of pitting corrosion of heal [52] Plotting y as a function of x yields a best-fitting
exchanger tubes straight line; its slope provides \/a and its intercept
9. Methods for the parameter estimation of the pit [16]
(aty =0) yields A. Instead of this graphical estima-
distribution in plants
10. Ultrasonic method for heat exchanger tubes with [44] tion of the parameters, more reliable estimates of a
the extreme value analy.sis and A can be obtained by using the MVLUE (min-
11. Maintenance system for coated heat exchanger tubes [45] imum variance unbiased estimator) method, the
12. Corrosion of steels in sea water [S.'i] maximum likelihood and the method of moments.
13. Analysis of perforation of zinc plating steels by [54]
Among them the MVLUE method which is dis-
extreme value statistics
14. Fatigue crack behavior of high strength steel in [55] cussed by Lieblein [20] is found to be more efficient
artificial sea water and unbiased for smalt size samples. The MVLUE
estimator can be calculated by

328
Volume 99, Number 4, July-August 1994
Journal of Research of the National Institute of Standards and Technology

\=Xa, {N, n)xi where y, rj and m are the location, scale and shape
parameter, respectively. This third type of asymp-
a^Sb, {N, n)xi. (5) totic distribution for the smallest value can be trans-
formed to the first type for the largest value, that is,
where a, (N, n) and bi {N, n) are weights for each Eq. (1), by changing l-F{t) to F{z) and by intro-
sample depending on the sample size, N, and trun- ducing the fallowing reduced variate
cated number, n, which are tabulated in the table
given by Tsuge [21, 23] up to ^ = 45, The weights, X = \nit-y),z = (,X-X)/a. (11)
A, B, C, of variance, V,
The same MVLUE method used for Eq. (1) can be
V = a^{A iN,n)y^ + B(N,n)y + CiN,n)) (6)
utilized for parameter estimation, because the fol-
lowing relations exist between the parameters of
are also found in the table given by Tsuge [21, 23].
both distributions;
For the pit depth distribution, the return period, T,
is defined as
A =ln{T7), a~\lm. (12)
T = S/s, (7)
The above unified procedure for estimating
where 5 is the surface object (e.g., a tank plate) to
parameters of the Gumbel and Weibull distribution
be examined and s is the area of the small speci-
was coded in the computer program EVAN [23].
mens which are sampled randomly from the objec-
tive. The return period, T, is in effect a size factor.
The mode. A, of the pit depth distribution for the
small specimen is simply obtained by the MVLUE 3. Examples
estimators mentioned above, and the mode for the
T times larger surface, xma, is given by Several examples are provided to demonstrate
the usefulness of extreme value statistics for analyz-
XmB=A +aln (7). (8) ing corrosion problems.

The perforation probability, P, of the maximum pit 3.1 Maximum Pit Depth of Oil Tank Plate
through the wall thickness, d, is given by
Through the 1960s and 1970s a number of oil
tanks were built in Japan. In the late 1970s there
F=l exp{-exp(- [d - (\ + a\n (T)/a)). (9)
occurred frequent oil leakages from tanks due to
corrosion failure. Oil refinery or petrochemical in-
Finally, the procedure [22] requires reporting the
dustries were located along the seacoast and oil
surface area of the object, 5, with the small sample
leakage caused serious environmental damages. In
area, s, providing the return period, T ( = S/s), and
1976, the fire service law was revised to enforce in-
the number of samples, N, with data number, n, ac-
spection of the thickness of the base and annular
tually obtained. In addition, the original thickness
plates of oil tanks every time oil was evacuated. On
of the plate, d, and the perforation probability, P,
these occasions extreme value analysis was applied
if needed are to be stated. The above procedure
and found to be a powerful tool for estimating the
does not request to check a goodness of fit of the
maximum pit depth. It is emphasized that data for
distribution obtained to the Gumbel distribution,
the base plate and the annular plate should be con-
but recommends to examine the fitness by the Kol-
sidered separately because they are characterized
mogorov-Smirnov or chi-square test if needed.
by different corrosion damage and mechanisms.
The law requests that plate thickness has to be
22 The Weibull Distribution
measured at the corners of every 10 cm square on
The third type for the smallest value called the the whole surface of the plate. This inspection pro-
WeibuU distribution cedure contributed greatly to reducing corrosion
leakage, but was time-consuming and costly. The
F(r) = l-exp(-((r-y)/Tir) (10) extreme value analysis was then studied intensively
in this field [35] [36]. Pit depth distribution sampled
can be fitted to the failure life distribution of stress from the whole base plate was found to obey the
corrosion cracking [33, 34] as shown in Table 2, Poisson distribution.

329
Volume 99, Number 4, July-August 1994
Journal of Research of the National Institute of Standards and Technology

Araki et al. [36] found that the largest value from


each small square (5 = 1 m-) being randomly sam-
pled obeys the Gumbel distribution as shown in
Fig. 1 [36], The slope and intersect of the line
{cr = 0.694 and A =1.41) were estimated by the
MVLUE estimators of Eq. (5). In this case, the sur-
face area of the base plate, S, was 1535 m^ and the
return period or size factor was calculated to be
T-S/s = 1535. The maximum depth, Xma, was cal-
culated by Eq. (8):

jc™, = 1.41+0.694 X In (1535) = 6.50 mm

which is shown also in Fig. 1. These data were ob-


tained for a base plate which was exposed for 12.6
years. Data for both annular and base plate exposed
for 7.7 years were plotted in Fig. 2 [36], from which
1 2 3 4 E B T a 9 10
the first leak due to the maximum pit is to be ex-
Maximum pit depth x (mm)
pected after 17.6 years for the base plate and 23.5
years for the annular plate, respectively. The effect Fig. i. The Gumbel plot of tlie maximum pits on the bottom
of N, s, on the estimates was examined and it was plate of the oil tank.
concluded that the MVLUE method is optimal for
N <2JQ and the maximum likelihood method is reli- IB
able for A^ > 20.

3J Rupture of Heat Exchanger Tubes of the Nominal thicknaas ot annular plala


Boiler
Super heater and economizer tubes of boilers are
exposed to high temperature gases with salt de-
posits which cause severe corrosion attack. Corro-
sion attack is not uniform, but localized at several
sites, watt thinning at the localized site resulting in
burst. Regular inspection is needed to predict time
for replacement of the tube before burst. All tubes
have to be examined for predicting exact time with
high confidence, but cost of inspection being high,
that Fukuda et al. [37] introduced the use of
extreme value analysis to supplement the inspection
of a small number of tubes. In Fig. 3, the largest val-
ues of wall thinning observed for 14 tubes are plot-
ted on Gumbel probability paper. The distribution 5 IT.l) 10 (12.8) IS 20 ?9
of wall thinning at every inspection time is seen to Operation tinw (yi
obey the Gumbel distribution and the maximum
thickness determined by the return period (40 Fig. 2. Estimated depth of the maximum pit on the whole sur-
tubes) increases with operation time as shown in face as a functiun of opcratinn years, and predictJun uf failure
Fig, 4. A criterion for a proper replacement time life for the bottom and annular plate.
has been proposed, which requires replacement
when the wall thickness reaches half of the design 3.3 The Pit Depth Distribution of Steel Piles in
thickness, t,y. Risk of burst could be avoided by Sea Water
estimating the depth and noting the proposed crite-
rion. Since the 1970s, steel pipes and piles have been
used extensively in Japan for harbor construction,
because lead time for construction could be
reduced compared with using concrete. Recently,

330
Volume 99, Number 4, July-August 1994
Journal of Research of the National Institute of Standards and Technology

corrosion of stee! pipes and piles was found to cause


the collapse of harbor structures. Then corrosion
damage of steel structures exposed in sea water has
been inspected and analyzed by using extreme value
analysis. Itoh et al. [38] reported that three differ-
ent types of depth distribution were found for steel
piles and plates depending on exposure time and
exposure location such as water level and deep sea
(Fig. 5). The type A distribution which exhibits a
nearly straight line, was found for uniform corro-
sion loss, its mean value being below 1.0-1.2 mm
thickness. The type C distribution obeying the
Gumbel distribution was observed for heavily local-
ized specimens. The type B distribution is a mixed
type of A and C distributions. The estimated depth
using the return period was consistent with observa-
tions.

_i_ j_
1.0 2.0 3.0
TMckriAms lost (mm)

Fig. 3. The Gumbel plots of the maximum thickness loss of


boiler tubes used for different operation times.

/
3,0

-■
1/2 tsr r

2.0
^U5x10'?.^« /
^

2 3 4 5 6 7 8 9
1.0 - Thickness loss (mm)
Fie. 5. Various types of the distribution observed for steel piles
/ and pipes exposed in sea water.

^,-< , 3.4 Classification of Data Based on Corrosion


1.0 2.0 Knowledge
Operation tlnte (x 10*h)
In any of the cases mentioned earlier, measured
sets of data is fitted by two or three distributions
Fig. 4. Thickness loss of boiler tubes as a function of operation
lime and estimation of rupture time. and must be separated from each other before the
analysis in order to obtain the maximum value.

^K'^^
Volume 99, Number 4, July-August 1994
Journal of Research of the National Institute of Standards and Technology

Imagawa et al. [15,16,17,39] presented many cases rence which is a very specific and common feature
which require classification of data. For example, of materials fracture. The Weibull distribution has
data for the heat exchanger tubes had to be classi- been known to be quite useful to analyze the distri-
fied into the inlet and outlet side samples because bution of fracture strength of various materials [40]
corrosion form and its degree of damage were dif- and also has been found to be applicable for analyz-
ferent at the two locations owing to exposure to dif- ing failure life distribution due to stress corrosion
ferent temperatures. For the oil tank, Imagawa cracking [33, 34].
observed that more deep pits were formed on the An interesting application of the Gumbel distri-
welding line compared with other parts. He ob- bution for analyzing the crack depth distribution
tained the different estimated value of the pit depth has been reported by Tsuge [41]. The laboratory ex-
for each classified sample. At the present time, the periment for evaluating the susceptibility of stress
classification was done on corrosion knowledge and corrosion cracking of Type 304 stainless steel was
experience, but it is required to establish a proce- done by using a bent specimen of u-shape. Bending
dure based on a common criterion. gives stress to the specimen and the environmental
condition of high pressure water causes many
3.5 Crack Depth Distribution of Stress Corrosion cracks, which can be revealed by sectioning the
Cracking specimen after the test as shown in Fig. 6. Distribu-
tion of the crack depth plotted in the Gumbel prob-
Stress corrosion cracking is one of the most dan-
ability paper showed two hues with an inflection
gerous corrosion failure and shows random occur-

,♦•***♦••

too 200
U 2000

0 typ*

Fig. 6. The distribution of intergranular corrosion attacks and cracks observed for sensitized
type 304 stainless sicci exposed to the BWR simulated water (DO 8 ppm, 250 °C).

332
Volume 99, Number 4, July-August 1994
Journal of Research of the National Institute of Standards and Technology

point as can be seen in Fig, 7. This inflection point phenomena, so that extreme value analysis was ap-
was found to correspond just to a depth for initiat- plied for estimating the maximum amount of segre-
ing the intergranular crack. Thus the initiation of gation in steel plate from small area samples [42],
the intergranular crack growth could be separated the concentration of impurities being measured by
from the initial process of purely chemical inter- using EPMA. The maximum amount of segregation
granular corrosion. thus determined can be used to predict the suscep-
tibility to lamellar tear. It should be emphasized
3.6 Estimation of the Maximum Segregation of that the statistical procedure for the chemical anal-
Impurities in Steel ysis is mainly concerned with the mean and stan-
dard deviation which assesses the reliability of the
Continuous casting of steel is one of the innova-
measurement, but not with extreme values. In re-
tive technologies achieved by the steel industry.
cent years, highly sensitive analytical methods have
Segregation and its band which are formed during
been developed, but it is not clear how to correlate
sohdification at the center of slab remain after
the data of small area samples to that of the total or
rolling and work as initiation sites for fracture phe-
bulk specimen. The ratio of the analytical area to
nomena such as lamellar tear and hydrogen induced
the bulk specimen reaches almost to 10 ", and ex-
cracking (HIC). The maximum amount of segrega-
treme value statistics is expected to be useful.
tion was found to be related to the above fracture

m
I

1000 2000
Haximuin crack length (p)
Fig. 7. Distributions changing with exposure time, the initial distrlbuliun corre-
sponding to intergranular corrosion and the second lo intergranular cracking.

333
Volume W, Number 4, July-August 1994
Journal of Research of the National Institute of Standards and Technology

3.7 Non-Destructive Methods With the Extreme zero, negative or positive, the distribution changes
Value Analysis to type I, type 11, and type III, respectively. They
found that the pit distribution on stainless steels in
Various types of nondestructive methods are acidified chloride solution fits the GEV distribution
used for inspecting and examining corrosion dam- with k positive, indicating that the type III for the
age in order to prevent failure. High sensitivity and largest value could be fitted. The type 111 distribu-
resolution in time or in space are required for the tion has a bound or a limit with increasing area, as
measurement. In addition, a computer-aided opera- suggested by Evans.
tion becomes popular, because huge amounts of What sample size, or what size of specimen area
data must be evaluated. For the heat exchanger, a should be used are questions from non-specialists in
thousand tubes must be checked and the number of statistics. For this question, we proposed a proce-
data easily exceeds 10^. An eddy current sensor [43] dure or criterion for choosing 5, N and T based on
and ultrasonic sensor probe [44] to steel tubes, and the variance given by Eq. (6). The surface area, 5,
an impedance sensor probe [45] for coating tubes of the object is given, and the sampling area, s, is
have been developed with the data logger and the selected so as to include at least one pit. Then
extreme value analysis software. T{ = S/s) is obtained. Accumulated data of the
parameters of a and A suggest (32] that the ratio of
a/A for localized corrosion is below, or not much
4. Discussion larger than, 0.3. Kinnison [48] states that the asymp-
The size effect on the maximum pit depth is totic theory predicts a constant ratio of 0.313 for all
found to be estimated with confidence by introduc- extreme value distributions. Then it can be assumed
ing the concept of the return period. Theoretical that the ratio a/Xy is 0.3. If we wish to control vari-
bases of the procedure have been provided by ex- ance within (A/3)-, the following relation can be de-
treme value theory [1]. Our experience shows that duced from Eq. (6)
the pit depth distribution obeys the normal or expo-
nential distribution, which belong to the exponen-
tial distribution family. Thus the maximum values of (\/a3f=A{N,n)y"-\-BiN.n)y + C. (14)
pit depth extracted from the exponential family dis-
tribution may reasonably be expected to obey the
Gumbel distribution. Thus the size effect could be Equation (14) can be solved fory or 7" as a function
rationally predicted by using the concept of return of Af and ff/A, as plotted in Fig. 8. When the ratio
period. of Q/A can be equated to 0.3 as discussed before, a
Evans [46] pointed out, however, that some cases suitable number of samples can be found for a given
as observed by Wormwell et al. [47] does not obey return period, 7. From this figure, the required size
a normal or exponential type distribution, but that of samples is Af=30 for 7 = 1200, or A'=20 for
the tail of the distribution is limited at a certain 7 = 274 and so on. This figure is approximately the
depth, Evans emphasized that such a limit is rea- same as what was observed empirically.
sonable for the case of anodic reaction control situ-
ation and this limited depth gives a rough indication
of the greatest pit depth to be expected on a much 5. Conclusions
larger area. Evans, however, did not check another
Extreme value statistics has been found to be a
possibility using the type Hi distribution which has
powerful tool for estimating the maximum value of
an upper limit. Recently Laycock et al. [24] dis-
localized corrosion depending on the surface area.
cussed that usefulness of the generalized extreme
Accumulation of data and experience, however, re-
value (GEV) distribution:
veals that statistics is less important than corrosion
experience and knowledge for obtaining a reason-
F(jc)-exp {-(1 -Jt(x-«)/«)'* kx < =a+uk, able estimation; measured data must be classified
based on the form of corrosion damage and its de-
(13) gree before the analysis. Properly classified data is
found to provide a very reasonable value. Nonde-
structive methods for measuring wall thickness with
because the distribution subsumes all three types various types of sensors, combined with extreme
with the sign of a shape parameter, k. When k is value analysis, have been developed in recent years.

334
Volume 99, Number 4, July-Augusi 1994
Journal of Research of the National Institute of Standards and Technology

[12] Y. Ishlkawa, Boshoku Gijutsu 28 (5), 278-284 (1979) (in


Japanese.)
[13] Y. Ishikawa, T. Ozaki, N. Hosaka, and O. Nuhida, J. Soc.
Malcr. Sci. Jpn. 30 (337), 975-980 (1981) (in Japanese.)
[14] Y. Ishikawa, T. Ozaki, N. Hosaka and O. Nishida, Trans.
ISIJ 22 (12), 977-983 (1982).
[15] H. Imagaua and K. Matsuno, Boshoku Gijutsu 31 (6), 406^
411 (1982) (in Japanese.)
[16] H. Imagawa and K. Matsuno, J. Soc. Mater. Sci. Jpn. 36
(400), 76-81 (1987) (in Japanese.)
[17] H. Imagawa and K. Matsuno, J. SoC. Mater. Sci. Jpn. 38
(430), 823-828 (1989) (in Japanese.)
[18] S. Kase, Boshoku Gijutsu 31 (7), 481^187 (1982) (in
Japanese.)
[19] S. Kase, How to Analyze Reliability Data —Double Expo-
nential Distribution, Ohm Ltd., Tokyo (1983) (in
Japanese).
[20] J. Lieblein, A New Method of Analyzing Extreme-Value
Data, National Advisoty Committee For Aeronautics,
Tech. Note 3053, January 1954.
[21] H. Tsugc, J. Soc. Mater. Sci. Jpn. 36 (400), 35^6 (1987).
(in Japanese).
[22] JSCE 60-1 Technical Committee, Working Group, Corro-
sion Eng. 37 (12). 699-705 (1988).
[23] JSCE 60-1 Technical Committee, Working Group, Com-
10 20 30 AO 50 puter program "EVAN," Muruzcn Publ., Tokyo (1989).
Sarnf^e size N [24] P. J. Laycock, R. A. Cottis and P. A. Scarf, J. Electrochcm.
Soc. 137 (1), 64-69 (1990).
[25] M. Kowaka, H. Tsuge, M. Akashi, K. Masamura and H.
Fig. S. Calculated curves of the optimum condition In choose T Ishimoto, An Introduction lo the Life Prediction of Plant
and A^ at various value or a/A. Materials —Application of Extreme Value Statistical
Methods for Corrosion Analysis, Maruzcn Publ., Tokyo
(1984) (in Japanese).
[26] A. H-S. Ang and W. H. Tang, Probabilistic Concepts in
Acknowledgments Engineering Planning and Design, Vol. II —Decision,
The author would like to acknowledge the contri- Risk, and Reliability, John Wiley & Sons, New York
(1984) p. 186.
bution of the JSCE 60-1 Committee Members,
[27] R. R. Kiniiison, Applied Extreme Value Statistics, Batlclle
particularly H. Tsuge and M. Akashi for establish- Press, Macmillan Publishing Co, A Division of Macmillian,
ing the standardized procedure for estimation of Inc., New York (1985).
the maximum pit depth. [28] 4lst Corrosion and Protection Symposium, Life Prediction
by Extreme Value Statistics, JSCE, January 1982.
[29] 51st Qirrosinn and Protection S>'mposium, Life Prediction
6, References of Plant Materials, JSCE, September 1983.
[30] 58th Corrosion and Protection Symposium, Life Predic-
[1] E. I. Gumbcl, Statistics of Extremes, Columbia University tion of Plant Materials —Application of Extreme Value
Press, New Yorlt and London (1958). Statistics to Corrosion, JSCE, December 1984.
[2] E. J. Gumbe], Statistical Theory of Extreme Values and [31] 73rd Corrosion and Protection Symposium, Life Predic-
Some Practical Applications, Natl. Bur. Stand. (U.S.), tion of Plant Materials —Experience and Practice of the
Appl. Math. Scr. No. 33 (1954), Analysis, JSCE, March 1988.
[3] R. B. Mears and U. R. Evans, The "Probabiiiiy" of Corro- [32] T. Shibata, ISIJ International 31 (2), 115-121 (1991).
sion, Trans. Faraday Soc. 31, 527-542 (1935). [33] T- Shibata, Localized Corrosion, F. Hinc, K. Komui, and
[4] U. R. Evans, The Corrosion and Oxidation of Metals: Sci- K. Yamakawa, eds., Elcsvicr Applied Science, London and
entific Principles and Practical Appliculiunsi, Edward New York (1988) p. 197.
Arnold, London (1960). [34] T. Shibata, Proc. Life Prediction of Corrodible Structures,
[5] G. G. Eldredge, Corrosion 13 (13), 51-60 (1957). R. N. Parkins, cd., NACE, Hou.ston, to be published.
[6] G. N. Scolt, Proc. Anicr. Petroleum Inst. IS (IV), 18 [35] T. Shibata and K. Okamoto, Boshoku Gijutsu 34 (7), 404-
(1934). 408 (1981) (in Japanese.)
[71 L. H. C. Trippet, Biometrika 17, ,364-386 (1925). [36] R. Araki, A. Miura, M. Sakai, J. Yokoyama, and S.
[8] P. M. Aziz, Corrosion 12 (12), 495-5U6 (1956). Yokoya, Atsuryoku Gijutsu 24 (3), 124-132 (1986) (in
[9] N. Masuto, Boshoku Giiutsu 21 (8), 347-353 (1972) (in Japanese.)
Japanese.) [37] Y. Fukuda, H. Kawasaki, and M. Scki. in Rcf. [30], p. 24.
[10] T. Shibata, Boshoku Gijutsu 27 (1). 23-35 (1978). (in [38] S. Itoh, T. Murata, and H. Okada. Poc. Inter. Congr.
Japanese.) Metallic Corrosion, Vol. 2, National Research Council of
[11] T. Shibata, Kiuzoku Hyomen Gijutsu 31 (6), 336-344 Canada. Ottawa (1984) pp. 3.30-.333.
(1980) (in Japanese.)

335
Volume 99, Number 4, July-August 1994
Journal of Research of the National Institute of Standards and Technology

[39] H, Imagawa, Haikan Gijutsu (Piping Engineering), Specia 1


Vol. 91-96 (1987) (In Japanese.)
[40] W. Wcibull, J. Applied Mechanics. ASME IS, 293-297
(1951).
[41) H. Tsuge, in Ref. [29], pp. 16-21 (in Japanese.)
[42] H. Ogawa, in Ret [30], pp. 31-36.
[43] T. Anzai, H. Yamamoto, K. Wakebe. Boset Kanrt 34 (4),
135-139 (1990) (in Japanese.)
[44] M- Kimura, Haikan Gijutsu, Special Vol. 142-145 (1987)
(in Japanese.)
[45] S. Kurusu and T. Sato, Haikan Gijutsu. Special Vol. 156-
160 (1987) (in Japanese.)
[46] U. R. Evans, The Corrosion and Oxidation of Metals:
Scientific Principles and Practical Applications, Edward
Arnold, London (1960) p. 927.
[47] F. Wormwcll, G. Butler, and J. G. Beynon, Trans. Inst.
Marine Eng. 69, 109-120 (1957).
[48] R. R. Kinnison, Applied Extreme Value Statistics, Baitellc
Press (1985) p. 70.
[49] S. Kobayashi, in Ref. [30], pp. 8-13.
[50] K. Malsuno and H. Imagawa, in Ref. [31], pp. 8-15.
[51] H. Ishimoto, in Ref. [29], pp. 56-62.
[52] S. Takajtaki, T. Kontani, J. Soc. Mater, Sci, Jpn. 36 (400),
72-75 (1987).
[53] F. Blekkenhorst, G. M. Frerrari, C. J. Van Dcr Wcckcn,
and F. P, Ijsseling, Br. Corros. J. 21 (3), 16.^ 176 (1986).
[54] H. Sato, K. Shimogori, H. Nishimoto, K, Miki. K Ikeda,
M. Iwai, H. Sakai, and S. Nomura, Testu-to-Haganc 72
(8), 1098-1105 (1986) (in Japanese.)
[55] K. Komai, K. Minoshima, and K. Kim, J. Soc. Mater. Sci.
36 (401), 141-146 (1987) (in Japanese.)
[56] Committee of Materials for Chemical Industry Plants,
Society of Chemical Engineering, Japan, SCC of Heat Ex-
changer Tubes —2nd Report on Survey and Failure Life
Analysis, Kagakukogyosha, Tokyo (1984).
[57] W. L. Clarke and G. M. Gordon, Corrosion 29 (1), 1-12
(1973).
[58] T. Shibata, J. Nakata, and S. Fujimoto, Life Prediction of
Conxxiiblc Structures, R. N, Parkins, ed., NACE, Hous-
ton, to be published.
[59] M. Akashi, Localised Corrosion, F. Hine, K. Komai, and
K. Yamakawa, eds., Elesevier Applied Science, Ixindun
and New York (1988) p. 175.

About the author: Toshio Shibata is a professor at


Osaka University, Osaka, Japan, and chairs the
Environmental Materials and Surface Processing
Group in the Department of Materials Science and
Processing.

336

You might also like