Downloaded/printed by Universidad de Chile Pursuant To License Agreement. No Further Reproductions Authorized

Copyright by ASTM Int'l (all rights reserved); Thu May 8 10:42:12 EDT 2014
Downloaded/printed by
Universidad De Chile pursuant to License Agreement. No further reproductions authorized.

This standard is for EDUCATIONAL USE ONLY.
STP 1250
Case Studies for Fatigue
Education
Ralph I. Stephens, Editor
ASTM Publication Code Number (PCN):
04-012500-30
ASTM
1916 Race Street
Philadelphia, PA 19103
Printed in the U.S.A.

Library of Congress Cataloging-in-Publication Data
Case studies for fatigue education / Ralph I. Stephens, editor.
(STP ; 1250)
"ASTM publication code number (PCN) 04-012500-30."
Includes bibliographical references.
ISBN 0-8031-1997-6
1. Materi al s--Fati gue--Case studies. I. Stephens, R. I. (Ralph Ivan)
III. Series: ASTM special technical publication ; 1250.
TA418.38.C37 1994
620.1 ' 126- - dc20
II. Series.
94-41520
CIP
Copyright 9 AMERICAN SOCIETY FOR TESTING AND MATERIALS, Philadelphia, PA. All
rights reserved. This material may not be reproduced or copied, in whol e or in part, in any printed,
mechanical, electronic, film, or other distribution and storage media, wi thout the written consent of
the publisher.
Photocopy Rights
Authorization to photocopy items for internal or personal use, or the internal or personal use of
specific clients, is granted by the AMERICAN SOCIETY FOR TESTING AND MATERIALS for users
registered with the Copyright Clearance Center (CCC) Transactional Reporting Service, provided
that t he base fee of $2.50 per copy, plus $0.50 per page is paid directly t o CCC, 222 Rosewood
Dr., Danvers, MA 01923; Phone: (508) 750-8400; Fax: (508) 750-4744, For those organizations that
have been granted a photocopy license by CCC, a separate system of payment has been ar-
ranged. The fee code for users of the Transactional Reporting Service is 0-8031-1997-6/94 $2.50
+ .50.
Peer Review Policy
Each paper published in this volume was evaluated by three peer reviewers. The authors
addressed all of the reviewers' comments t o the satisfaction of both the technical editor and the
ASTM Commi ttee on Publications.
The quality of the papers in this publication reflects not only the obvious efforts of the authors
and the technical editor, but also the work of these peer reviewers. The ASTM Commi ttee on
Publications acknowl edges with appreciation their dedi cati on and contribution t o ti me and effort or,
behalf of ASTM.
Printed in Philadelphia, PA
December 1994

Foreword
This publication, Case Studies for Fatigue Education, contains papers presented at the
symposium of the same name, held in Atlanta, GA, 18 May 1993. The symposium was
sponsored by ASTM Committee E-8 on Fatigue and Fracture and its Subcommittee E-08.01 on
Research and Education. Symposium session chairpersons were: R. I. Stephens, The
University of Iowa; R. C. Rice, Battelle Memorial Institute; and N. C. Dowling, Virginia
Polytechnic Institute and State University. R. I. Stephens presided as symposium chairperson
and editor of this publication.

Contents
Overvi ew--n. 1. STEPHENS 1
1. Automotive Wheel Assembly: A Case Study in Durability Desi gn--
R. W. LANDGRAF, S. THANGJI THAM, AND R. L. RIDDER 5
2. The Use of Failure Analysis and Materials Testing in the Redesign of a
Boat Trailer Roller Arm----6. c. SALIVAR 23
3. Design of a Composite Hip Prosthesis for Long-Term Performance--i<.
L1AO AND K. L. REIFSNIDER 32
4. Fatigue of a Landing Gear Act uat or Beam in a Fighter Ai r c r a f t -
s. KANTI MATHI AND T. N. WHI TE 53
5. Fatigue Evaluation of Agitator Paddle Shafts--H. R. JHANSALE 67
6. Fatigue Cracking of a Welded Roll Used in a Paper-Mill Roll Pr ess- -
J. E. ZAPATA AND S. C. ANDERSON 77
7. Welded Pipeline Fatigue Analysis--D. SOCIE AND E. SEGAN 86
8. Thermal Fatigue Analysis: A Case Study of Recuperat ors--s. p. BHAT 101
9. Shell and Detail Fract ure Format i on in Railroad Rails--R. c. RICE 109
10. Equating Damped Vibration to Constant Amplitude Fatigue Loading for a
Thick-Walled Pressure Vessel--a. L STEPHENS, T. B. ADAMS, AND
S. L, CARLSON 139
11. Development of a Numerical Model for Predicting Fatigue Lives of
Tubular Threaded Connections--T. o. LIEBSTER AND G. GLINKA 156
12. Fatigue Life Prediction for Wind Turbines: A Case Study on Loading
Spectra and Paramet er Sensitivity--H. J. SUTHERLAND, P. S. VEERS, AND
T. D. ASHWI LL 174
13. Fatigue Cases Involving the Use of Wood and a Wood Composi t e--
G. H. KYANKA 208

STP1250- EB/ Dec. 1994
Overview
The Case Studies for Fatigue Education speci al t echni cal publ i cat i on (STP) was pl anned
to provi de engi neeri ng educat ors and students with a broad range of non-t ri vi al , r eal - wor l d
fatigue pr obl ems/ si t uat i ons and solutions for use in the cl assroom. Hopeful l y, t hese cases
wi l l provi de st i mul at i on for a bet t er underst andi ng of the maj or causes of mechani cal failure.
The 13 cases i ncl uded in this publ i cat i on i nvol ve new desi gns, r ewor k desi gns, failure anal-
ysi s, pr ot ot ype deci si ons, envi ronment al aspects, met al s, non-met al s, component s, structures,
and fasteners. As Ri ce poi nt s out in his case i nvol vi ng rai l road rails, the cases bri ng out the
need for students to integrate el ement s of engi neer i ng that commonl y enter into a fatigue
desi gn or fai l ure anal ysi s. These el ement s i ncl ude mechani cs of def or mabl e bodi es, mat eri al s
sci ence and charact eri zat i on, fract ography, nondest ruct i ve i nspect i on, desi gn of experi ment s,
performi ng and eval uat i ng experi ment s, dat a acqui si t i on and reduct i on, damage model i ng,
life predi ct i on, and rel i abi l i t y. Ri ce also poi nt s out that most fat i gue probl ems do not have
one unique solution, and in fact, the " bes t " sol ut i on can often be di ct at ed by financial
constraints, t i me l i mi t at i ons, avai l abi l i t y of pert i nent mat eri al and processi ng i nformat i on,
l i abi l i t y concerns, and publ i c percept i on. Based upon the above, the sol ut i ons for t hese cases
range from compl ex to simple.
In order to provi de real -l i fe cases rather than t echni cal research papers, authors were
request ed to use a format sui t abl e for educat i onal case studies. A vari et y of di fferent format s
coul d be successful in achi evi ng this end. Aut hors were given excerpt s from an Amer i can
Soci et y for Engi neeri ng Educat i on ( ASEE) paper, on wri t i ng engi neeri ng cases. It was sug-
gest ed that each case should have specific comment s, quest i ons, instructions, and so forth
for st udent / f acul t y readers to consider. The t hree referees of each paper were al so given
these i nst ruct i ons concerni ng format to ai d t hem in t hei r review deci si ons. Thus, the authors
and referees worked very hard to hopef ul l y bri ng t oget her qual i t y case studies on fatigue
that wi l l be benefi ci al in an educat i onal envi ronment . Thi s educat i onal envi ronment i ncl udes
undergraduat e and graduat e level courses and cont i nui ng educat i on such as short courses
and t el ecommuni cat i on medi a courses. The cases are al so appl i cabl e to pract i ci ng engi neers
i nvol ved wi t h fatigue pr obl ems either on a si ngl e i nvol vement basi s or as a group l earni ng
situation. Thus, the market or interest for t hese cases has act ual l y expanded from the ori gi nal
goal s of pr i nci pal l y uni ver si t y/ col l ege usage to i ncl ude the pr act i ci ng engineer.
Faculty, students, and pract i ci ng engi neers may have a difficult t i me in choosi ng cases for
specific goals. In order to si mpl i f y this choosi ng and to provi de a bet t er underst andi ng and
content of each case, the fol l owi ng t abl e is pr ovi ded in this overview. The t abl e i ncl udes
headi ngs that emphasi ze the pri nci pal aspect s of each case. The paper number agrees with
the number in the t abl e of contents. The second col umn, ent i t l ed Maj or Topic, i ncl udes one
to six words that best descri be the product involved. It is qui ckl y seen that a variety of
different product s are i nvol ved in the thirteen cases. The Aut hor col umn i ncl udes the names
of all authors for each case. In the Mat eri al col umn, it is seen that a variety of carbon steels
(1010, 1018, 1040, 1080), al l oy steels ( HSLA, 4340, D6AC), st ai nl ess steels (A312, 304),
al umi num al l oys (5454, 6063), wood and wood composi t e, and a pol ymer composi t e are
i nvol ved wi t h these cases. Both low- and hi gh-st rengt h mat eri al s are involved. The next three
col umns pr ovi de i nformat i on as to the t ype of fatigue model involved. An X in the E-N
col umn means that case i nvol ves the l ocal not ch strain met hodol ogy involving st rai n-l i fe
data. An X in the S-N col umn means the case involves the nomi nal stress met hodol ogy
i nvol vi ng st ress-l i fe data. An X in the LEFM col umn means that the case i nvol ves l i near
el ast i c fracture mechani cs usi ng fatigue crack growt h rate, da/dN, whi ch is a funct i on of the
1
Copyright* 1994 by ASTM International www.astm.org

2 CASE STUDIES FOR FATIGUE EDUCATION
stress intensity factor range, AK. Seven cases involve E-N, three cases involve S - N , and five
cases involve d a / d N - A K . Two cases have more than one fatigue methodology. The next
two columns, FEA and Experimental Stress Analysis, provide whether stresses/strains were
determined through finite element analysis and/ or through experimental means. In other
cases, stress calculations using a strength of materials approach were used if needed. Three
cases include FEA and four cases include experimental stress analysis. The next two col-
umns, Fatigue Life Predictions and Fatigue Tests are involved in every case; that is, every
case involved fatigue life predictions and/ or fatigue tests, which is probably expected. The
last column indicates six cases that involved some form of fractographic analysis. This
included both macro and micro analysis using optical and/ or scanning electron fractography.
Hopefully, this table will aid in making appropriate case selections for a given objective. It
is suggested that potential case users review this table before considering a specific case.
Case reproduction as class handouts will be a very important consideration for users.
ASTM offers quantity discounts for this STP, as well as for reprints of individual cases.
Please call ASTM customer service for more information at (215) 299-5585. As for photo-
copying, this authorization is addressed in a paragraph that appears in the front matter of
this, and all ASTM STPs. Please refer to this paragraph for phot ocopyi ng requirements.
The thirteen cases in this publication involve authors representing six universities, six
private companies, and two government agencies. The cases come from ten different states
within the United States and one province in Canada. They represent a broad spectrum of
engineering fatigue problems. Not the least of these problems is product liability litigation.
Two additional papers had to be withdrawn by the authors during the refereeing stage due
to lawyer requests, based upon active products liability litigation. This just points out addi-
tional difficulties in fatigue education and that hopefully this publication will contribute to
quality engineering education involving fatigue.
The editor would like to thank the authors, referees, symposi um session chairpersons, the
organizing committee, and the ASTM staff for making this publication possible. The organ-
izing commi t t ee included R. I. Stephens, The University of Iowa, Chairperson; R. C. Rice,
Battelle Memorial Institute; N. C. Dowling, Virginia Polytechnic Institute and State Univer-
sity; B. I. Sandor, The University of Wisconsin; and H. Sehitoglu, The University of Illinois.
Ralph I. Stephens
Mechanical Engineering Department
The University of Iowa,
Iowa City, I A 52242;
symposi um chairperson and STP editor.

C
a
s
e

S
u
m
m
a
r
y

T
a
b
l
e

P
a
p
e
r

E
x
p
e
r
i
m
e
n
t
a
l

F
a
t
i
g
u
e

L
i
f
e

F
a
t
i
g
u
e

N
u
m
b
e
r

M
a
j
o
r

T
o
p
i
c

A
u
t
h
o
r
s

M
a
t
e
r
i
a
l

~
-
N

S
-
N

L
E
F
M

F
E
A

S
t
r
e
s
s

A
n
a
l
y
s
i
s

P
r
e
d
i
c
t
i
o
n
s

T
e
s
t

F
r
a
c
t
o
g
r
a
p
h
y

1

a
u
t
o

w
h
e
e
l

L
a
n
d
g
r
a
f
,

T
h
a
n
g
j
i
t
h
a
m
,

1
0
1
0

a
n
d

H
S
L
A

X

X

X

X

a
s
s
e
m
b
l
y

a
n
d

R
i
d
d
e
r

s
t
e
e
l
,

5
4
5
4

a
l
u
m
i
n
u
m

2

b
o
a
t

t
r
a
i
l
e
r

r
o
l
l
e
r

S
a
l
i
v
a
r

1
0
4
0

s
t
e
e
l

X

X

a
r
m

3

h
i
p

p
r
o
s
t
h
e
s
i
s

L
i
a
o

a
n
d

R
e
i
f
s
n
i
d
e
r

p
o
l
y
m
e
r

c
o
m
p
o
s
i
t
e

X

X

X

X

X

4

l
a
n
d
i
n
g

g
e
a
r

K
a
n
t
i
m
a
t
h
i

a
n
d

W
h
i
t
e

D
6
A
C

s
t
e
e
l

X

X

X

X

X

X

a
c
t
u
a
t
o
r

b
e
a
m

5

p
a
d
d
l
e

s
h
a
f
t

J
h
a
n
s
a
l
e

1
0
1
8

s
t
e
e
l

X

X

X

X

6

p
a
p
e
r

m
i
l
l

r
o
l
l

Z
a
p
a
t
a

a
n
d

A
n
d
e
r
s
o
n

w
e
l
d
e
d

m
i
l
d

s
t
e
e
l

X

X

X

7

p
i
p
e
l
i
n
e

w
e
l
d
m
e
n
t

S
o
c
i
e

a
n
d

S
e
g
a
n

A
3
1
2

s
t
a
i
n
l
e
s
s

s
t
e
e
l

X

X

X

X

8

r
e
c
u
p
e
r
a
t
o
r

B
h
a
t

3
0
4

s
t
a
i
n
l
e
s
s

s
t
e
e
l

X

X

9

r
a
i
l
r
o
a
d

r
a
i
l
s

R
i
c
e

1
0
8
0

s
t
e
e
l

X

X

X

X

X

X

1
0

t
h
i
c
k
-
w
a
l
l
e
d

S
t
e
p
h
e
n
s
,

A
d
a
m
s
,

a
n
d

A
7
2
3

s
t
e
e
l

X

X

X

X

X

p
r
e
s
s
u
r
e

v
e
s
s
e
l

C
a
r
l
s
o
n

1
1

t
u
b
u
l
a
r

t
h
r
e
a
d
e
d

L
i
e
b
s
t
e
r

a
n
d

G
l
i
n
k
a

4
3
4
0

s
t
e
e
l

X

X

X

c
o
n
n
e
c
t
i
o
n

1
2

w
i
n
d

t
u
r
b
i
n
e

S
u
t
h
e
r
l
a
n
d
,

V
e
e
r
s
,

a
n
d

6
0
6
3

a
l
u
m
i
n
u
m

X

X

X

A
s
h
w
i
l
l

1
3

w
o
o
d

g
u
n

s
t
o
c
k

a
n
d

K
y
a
n
k
a

w
o
o
d

a
n
d

w
o
o
d

X

b
o
w
l
i
n
g

p
i
n

c
o
m
p
o
s
i
t
e

O

/
r
t

3
0

C
o
p
y
r
i
g
h
t

b
y

A
S
T
M

I
n
t
'
l

(
a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
)
;

T
h
u

M
a
y

8

1
0
:
4
2
:
1
2

E
D
T

2
0
1
4
D
o
w
n
l
o
a
d
e
d
/
p
r
i
n
t
e
d

b
y
U
n
i
v
e
r
s
i
d
a
d

D
e

C
h
i
l
e

p
u
r
s
u
a
n
t

t
o

L
i
c
e
n
s
e

A
g
r
e
e
m
e
n
t
.

N
o

f
u
r
t
h
e
r

r
e
p
r
o
d
u
c
t
i
o
n
s

a
u
t
h
o
r
i
z
e
d
.

Ronal d W. Landgr af ~ Surot Thangjitham, J and Ri chard L. Ri dde&
Automotive Wheel Assembly: A Case Study
in Durability Design
REFERENCE: Landgraf, R. W., Thangjitham, S., and Ridder, R. L., "Automotive Wheel
Assembly: A Case Study in Durability Design," Case Studies for Fatigue Education, ASTM
STP 1250, Ralph I. Stephens, Ed., American Society for Testing and Materials, Philadelphia,
1994, pp. 5-22.
ABSTRACT: A project to decrease the weight of a stamped metal automotive wheel assembly,
through material substitution and downgaging, is presented as a case study in durability design.
A coordinated analytical/experimental approach is used to assess wheel fatigue performance
under laboratory and simulated service conditions. Finite element modeling is employed to
develop relations between bending moments applied to the wheel during cornering maneuvers
and peak stress excursions in the wheel spider. Cyclic material properties for candidate ma-
terials (high-strength steel and aluminum), that include the effects of cold work resulting from
the wheel-forming operation, are used with strain-based fatigue methods to obtain estimates
of wheel performance under various cyclic loading situations, including a standard Society of
Automotive Engineers (SAE) laboratory fatigue test and service histories representative of
different drivers and customer routes. Finally, reliability design methods are employed to eval-
uate the effects of variations in wheel geometry, materials properties, and service loading on
the expected fatigue performance of a fleet of vehicles in service situations. This approach
provides failure probability information based on measured or estimated variations in design
parameters and is particularly relevant to quality and warranty issues.
KEY WORDS: fatigue analysis, wheel design, materials substitution, finite elements, service
histories, reliability, fatigue education
Aut omot i ve wheel s have evol ved over the decades from earl y spoke desi gns of wood and
steel, carry overs from wagon and bi cycl e t echnol ogy, to flat steel di scs and, finally, to the
st amped met al confi gurat i ons of moder n vehicles. Hi st ori cal l y, successful designs were ar-
ri ved at t hrough experi ence and ext ensi ve field testing. In recent t i mes, t hese procedures have
been suppl ant ed by a variety of experi ment al and anal yt i cal t echni ques for structural anal ysi s
(strain gages and finite el ement met hods), dur abi l i t y anal ysi s (fatigue life predi ct i on), and
rel i abi l i t y met hods for deal i ng wi t h the vari at i ons i nherent in engi neeri ng structures. Thi s
newer t echnol ogy provi des unique opport uni t i es to i mpr ove the product devel opment process
t hrough the appl i cat i on of more rat i onal and t i me savi ng procedures.
Wheel s are cl earl y safety rel at ed component s and, hence, fatigue performance has al ways
been a pri me concern. Further, wheel s cont i nue to recei ve consi derabl e attention as part of
i ndust ry efforts to reduce weight, in this case unsprung weight, t hrough mat eri al subst i t ut i on
and down-gagi ng. The disc, or spider, port i on of the wheel assembl y is part i cul arl y vul ner-
abl e to the hi gh bendi ng moment s generat ed duri ng corneri ng maneuvers. This situation is
Professor and associate professor, respectively, Virginia Polytechnic Institute, Blacksburg, VA 24061-
0219.
z Design engineer, B&W Fuel Company, Lynchburg, VA 24506-0935.
5

portrayed in Fig. I where the applied loading is seen to be a function of vehicle geometry
and weight and the lateral acceleration achieved in cornering. Shown in Fig. 2 is the con-
figuration of a standard Society of Automotive Engineers (SAE) laboratory test (SAE J328a,
Wheel-Passenger Cars: Pertbrmance Requirements and Test Procedures), designed to simu-
late cornering loads and which, in the United States, is used as an acceptance test by wheel
designers.
The case study presented here is based on a comprehensive project to reduce the weight
of a 14 by 5 in. stamped automotive wheel spider through material substitution. The student
is exposed to modern approaches for selecting materials and sizing components based on
specified performance objectives based either on a standard accelerated laboratory test or on
expected service history spectra. In a more general vein, the study provides a platform for
introducing students to modern design tools for structural and durability analysis and for
critically assessing their applicability in product development and evaluation.
After a statement of project objectives, the study is presented in three parts, each repre-
senting an increasing level of sophistication. The first part focuses on initial component
sizing, that is, selection of appropriate thicknesses for each candidate material, to successfully
meet the standard laboratory fatigue test requirement. Finite element analysis (FEA) is em-
ployed to determine stress distributions in the rather compl ex wheel spider; these results are
compared with experimental strain measurements on production wheels as a validation check.
Tests results on prototype wheel assemblies are provided to assess predictive accuracy.
In the second part, estimates of the performance of prototype wheels under simulated
service conditions are developed using field measurements representative of various percen-
tile drivers and various customer routes. Here attention is focused on the influence of vari-
ations in service usage and material properties on wheel life. Finally, a combi ned durability
and reliability analysis is presented to project the probability of wheel failure under service
conditions when expected variations in material properties and component geometry are
simultaneously considered.
t : t ~ l I
r V e h ic le : - ]
| g e o m e t r y ]
M = f / w e i g h t /
L _ l a t . accel._.J
FIG. l - - Wh e e l loading during cornering maneuver.

LANDGRAF ET AL. ON AUTOMOTIVE WHEEL ASSEMBLY 7
RiM
'1 1
I
I
TEST LOAO
F
~ O ~ E N T
I A R M
i
FIG. 2--SAE rotary fatigue test f or evaluating wheel performance.
I f
1
Pr oj ect Obj ect i ve
Determine, through the use of comput er simulation and analysis tools, the extent of weight
reduction possible by the substitution of high-strength, low-alloy (HSLA) steel or aluminum
for conventional hot-rolled, low-carbon steel in an automotive wheel spider.
S t a g e O n e : I n i t i a l C o m p o n e n t S i z i n g
Perf ormance Requi rement s
9 SAE standard rotary fatigue test (SAE J328a),
9 applied moment = 1 528 ft-lbs (2 072 N 9 m),
9 required fatigue life = 50 000 cycles (Weibull B , , 90% confidence level), and
9 for initial sizing, use 100 000 cycles.
Technical Issues
9 fatigue testing tO established standard procedures,
9 stress analysis using finite element methods,
9 incorporating plasticity considerations in elastic stress analysis,
9 correlating calculated stresses with experimental (strain gage) measurements,
9 fatigue analysis using strain-based methods, including material processing effects,
9 establishing component dimensions for least-weight design, and
9 material selection for fatigue resistance.

/ ' I
/
1
FIG. 3--Finite element model of wheel spider.
Technical Approach
The first step i n ar r i vi ng at an est i mat e of fat i gue per f or mance for the wheel spi der is to
l ocat e the hi ghl y st ressed areas under the pr escr i bed test l oadi ng. Because of the compl ex
geomet r y of the wheel spider, fi ni t e el ement model i ng is empl oyed for this purpose. Local
stresses and st rai ns at pot ent i al fai l ure-si t es are t hen det er mi ned as a f unct i on of l oad level
and t hi ckness. These in t urn are used i n conj unct i on wi t h mat er i al fat i gue pr oper t i es to
cal cul at e al l owabl e t hi cknesses for candi dat e mat eri al s.
Stress Analysis
Fi gur e 3 shows a fi ni t e el ement model of t he wheel spi der; i sost ress cont our s ( von Mi ses
stress), r esul t i ng f r om t he appl i cat i on of a be ndi ng mome nt t hr ough the wheel hub, are
di spl ayed i n Fi g. 4. Because it is the r ange of stress, and not the ma xi mum val ue, t hat is of
si gni f i cance for fat i gue cal cul at i ons, the stress excur si ons dur i ng one wheel r evol ut i on must
be det er mi ned. Thi s can be accompl i shed by i ncr ement al l y al t er i ng t he or i ent at i on of the
appl i ed moment . The resul t s of such an anal ysi s are shown for t hree cri t i cal l ocat i ons i n Fig.
5. By per f or mi ng addi t i onal anal yses at di f f er ent appl i ed moment s and mat eri al t hi cknesses,
a gener al stress r el at i onshi p can be devel oped
wher e
S, = stress ampl i t ude (ksi),
M = mome nt (i n. -ki ps), and
t = t hi ckness (in.). 3
S, = 0. 153 9 M" (t ,.4)
( 1)
For stress in MPa, moment in N 9 m, and thickness in mm, the constant in Eq I is 0.0344.

LANDGRAF ET AL. ON AUTOMOTIVE WHEEL ASSEMBLY 9
FIG. 4--Stress contour plot for applied bending moment.
When using computer simulation tools in engineering design, it is important to establish
the validity of design calculations by experimental confirmation. Only in this way can the
necessary confidence in computational methods be established. In the current study, produc-
tion wheels had been instrumented with single-leg strain gages at observed failure locations,
thus providing the necessary experimental check on finite element results. The location of
fatigue cracks and strain gages are shown in Fig. 6. These locations are seen to be consistent
with the finite element results in Fig. 5.
A problem arises, however, when, with reference to the material stress-strain curves in
Fig. 7, it is noted that the stresses achieved during the finite element simulation of the SAE
laboratory test exceed the yield strength of the production steel, thus violating the elastic
assumptions of the finite element model. An approach for resolving this dilemma entails
using a plasticity correction proposed by Neuber [1] for notched members. This method is
illustrated in Fig. 8 where the elastic finite element solution, when plotted on the elastic line,
allows determination of a rectangular hyperbola which, at the point of intersection with the
material stress-strain curve (cyclic), provides an estimate of the actual local stress and strain.
A check of the validity of this approximate, but easily applied, method is provided by strain
gage measurements on the production wheel. These results for the two highest stress points,
presented with the corrected finite element strains in Fig. 9, indicate that the method gives
quite reasonable results.
Thickness Determination
With an adequate stress-strain analysis in hand, appropriate material thicknesses, and hence
weight reductions, can be determined based on the relative fatigue resistances of candidate

W h e e l S p i d e r
~| <0 . .... ?
\ , " ,, /
, - : , , , ,
90 ," \ l eo 27o / 360
,.,~ ~ \ \ / Angl e
FIG. 5 - - S t r e s s excursi ons at cri t i cal l ocat i ons duri ng one- wheel revol ut i on.
(
/
FIG. 6 - - Fa t i g u e cracks in t est ed wheel and st rai n gage locations.

LANDGRAF ET AL. ON AUTOMOTI VE WHEEL ASSEMBLY 11
120 8 0 0
SAE 9 8 0 X
I 0 0
80 / / / ~ . . . . . . . . . . . . . 6 0 0
-~ / " SAE 9 5 0 X
~ 4o
200
20 - - - M o n o t o n i c
- - Cyctic
. J ~ = I
i 0 .0 1 0 4 t O 0 1 0 8 0 .0 1 2 0.016
S t r o i n
FI G. 7--St ress-st rai n curves f or three:wheel steels.
materials. Strain-life curves for three steel s are di spl ayed in Fig. 10. The strain-life relation
is o f the form
r
% ( 2Nf f ' + e.;-(2Nr)" E a ~ ( 2 )
where the relevant cycl i c properties, al ong with densi ty and cost information, for candidate
wheel materials are presented in Table 1. Properties are reported both for the as recei ved
and a 20% col d worked condi ti on for each material to account for metal formi ng effects on
wheel performance. Col d worki ng o f this magnitude, whi ch is typical o f wheel formi ng
processes, can measurably i mprove material fatigue performance.
An additional computati onal aid is provi ded when the strain-life curve for the candidate
materials are presented in the parametric form s hown in Fig. 11. By entering this curve at
the elastic stress sol uti on, the Neuber correction is automati cal l y applied and the estimated
life can be directly determined. This relation is a functi on o f the cycl i c material properties
and is o f the f ol l owi ng form
t ") ~h
S , , = [ ( % ) - ( 2 N f ) " + ~ r ' ~ E ( 2 N f f ' + " ] ' / 2
(3)
120
Ioo
8o
6o
4O
20
SAE 9 8 0 X ]800600
K t S - - - - ~ _ ~ ~ ~
/ / \ 950 X
7 :
L i I i
0 . 0 0 4 0 . 0 0 8 ODI 2
St r ai n
FI G. 8- - Ne ube r procedure f or applying plasticity correction to elastic analysis.

0 , 0 0 3
0 . 0 0 2
.c
0 . 0 0 1
0
4 0 0
Wh e e l S p i d e r
F E A / N e u b e r / /
/
/
o
~J
/
i i i i i i i
8 0 0 1 2 0 0 1 6 0 0 2 0 0 0
B e n d i n g Mo me n t ' , f t - l b =
0 . 0 0 3
0 0 0 2
. i
.E
~ 0. 001
W h e e l S p i d e r
/
o
F E A / N e u b e r / /
/
~ ~ ~
0 i i i i i i i L
4 0 0 8 0 0 1 2 0 0 1 6 0 0 2 0 0 0
B e n d i n g M o m e n t , f t - l b s
FIG. Q--Comparison of corrected FEA results with strain-gage measurements f or Locations 1 and 2
in Figs. 5 and 6.
E " % .
r 9 8 o x
~ , o o o , ~ ~' - - 9 s o x
[. ~ 2 N t H R L C
E
i
O 0 0 0 t L , , I , , J , , I ; , I ~ ~ I , ~ L ~ ~ I
I I O ~ 0 2 I O s 104 IO s I 0 s 107
R e v e r s a l s t o Fai l ur e, 2 N f
FIG. IO--Strain-life curves for candidate steels.

LANDGRAF ET AL. ON AUTOMOTIVE WHEEL ASSEMBLY
TABLE l --Wheel material data.
13
Material E, ksi(MPa) ~ , ksi(MPa) b ~ c p, lbs/in.3(g/cm 3) $/ l b
SAE 1015 29 500 (203) 90 (621) 0.10 0.16 -0. 44 0.284 (7.87) 0.20
20% CW 80 (552) -0. 07 0.17 -0. 45
SAE 980X 29 500 (203) 190 (1310) -0. 10 0.25 -0. 56 0.284 (7.87) 0.26
20% CW 190 (1310) 0.07 0.25 -0. 56
5454 Aluminum 10 000 (69) 80 (552) -0. 12 1 . 8 0 -0. 85 0.096 (2.66) 1.00
20% CW 80 (552) -0. 10 1.75 - 0. 80
Mat eri al t hi cknesses requi red to sat i sfy the SAE l aborat ory test (100 000 cycl es) can be
det ermi ned using Eqs 1 and 3. Resul t s of t hese cal cul at i ons are summar i zed in Table 2 where
it is seen that the SAE 980X steel provi des a 21% wei ght savi ngs wi t h no cost penalty,
whi l e the 5454 al umi num al l oy provi des a substantial 47% wei ght savi ngs but at a l arge cost
penalty. Finally, evi dence of t he l egi t i macy of the descr i bed met hod is provi ded in Fig. 12
where predi ct ed fat i gue lives are compar ed wi t h exper i ment al results f r om rotary wheel tests.
The agreement is found to be qui t e reasonabl e, part i cul arl y in light of the rather large scat t er
evi dent in the test results; the symbol s represent average lives, whi l e the bands i ndi cat e the
range of the test data.
Summari zi ng St age 1 Tasks, finite el ement results, correct ed for, pl ast i ci t y effects and
val i dat ed by experi ment al measurement s, are used to accompl i sh i ni t i al component sizing
based on the st andard SAE rotary fat i gue test. Mat eri al processi ng effects (col d worki ng)
are i ncl uded t hrough al t erat i ons in cycl i c mat eri al propert i es. An initial eval uat i on of wei ght
saving pot ent i al and rel at i ve mat eri al cost s is thus accompl i she& In the next part, these results
will be ext ended to predi ct i ons of ser vi ce performance.
St a g e 2: Pr e d i c t i o n o f S e r v i c e P e r f o r ma n c e
Obj ect i ve
Usi ng measured wheel servi ce l oadi ng spectra, est i mat e the expect ed servi ce life, in miles,
of the wheel spiders.
2 0 0 0
I 0 0 0
"7~ 5 0 0
2 0 0
<~ I 0 0
b
< 50
2 0
HRLC
i i I i , i f i [ q q i , , ~ , i ~ , ~ I
I 0 10 2 10 3 10 4 10 5 10 6 10 7
Rever sol s 1o Foi l ur e, 2 N f
F[G. 1 I--Parametric fatigue curves f or finite element life predictions.

14 CASE STUDI ES FOR FATIGUE EDUCATION
TABLE 2- - Wei ght reduction results.
Weight Thickness, mm (in.) Relative Weight Relative Cost (material)
SAE 1010 3.56 (1.14) 1.0 1.0
SAE 980X 2.79 (0.11 ) 0.79 1.0
5454 Aluminum 5.72 (0.22) 0.53 2.6
Addi t i onal Techni cal I s s ues
9 servi ce l oads measurement ,
9 det ermi nat i on of driver vari abi l i t y,
9 account i ng for mat eri al var i abi l i t y in fatigue cal cul at i ons,
9 est i mat i ng servi ce performance, and
9 use of damage di st ri but i on hi st ograms.
Techni cal Ap p r o a c h
Wi t h hi gh vol ume product s such as aut omot i ve wheel s, it is i mport ant to anticipate, as
accurat el y as possi bl e, the ext remes of field usage. Audi engi neers [2] have report ed wheel
servi ce spect ra devel oped usi ng an i nst rument ed vehi cl e that was dri ven over a st andard
rout e by a cr oss- sect i on of cust omers. Resul t s of this sampl i ng are t abul at ed in Table 3 and
di spl ayed in Fi g. 13 as a series of bendi ng moment - f r equency spect ra for various percent i l e
drivers. Wheel bendi ng moment s are nor mal i zed to the levels obt ai ned by a professi onal t est
dri ver (100%). These usage profiles, when combi ned wi t h the previ ous finite el ement results
and mat eri al pr oper t y i nformat i on, pr ovi de predi ct i ons of expect ed l i fet i mes for the vari ous
percent i l e drivers. The procedure i nvol ves per f or mi ng a l i near cumul at i ve damage anal ysi s
I 0 6
Wheel Spider
- = /
' ! - - - - Y -17Y .....
s ~ / ~ - - 7 ~ : . ' 2 . ! - - - - o H R L C - - -
i = t / ~ / 4 ', : , Q H S L A ' '
F . . . . . / ' ~ . ; / I i A H S L A 2 -
/ / . ~ H S L A :5
10 4 10 5 1 0 6
Experimental Fatigue L i f e , cycles
FIG. 12 Comparison of predicted f at i gue lives with wheel test results.

L A N D G R A F E T A L . O N A U T O M O T I V E W H E E L A S S E M B L Y
TABLE 3- - Audi wheel spectra.
15
Cycle Count per Driver
Moment, % FS Test 1% 3% 50% 97%
100 3
95 17
90 35
85 55 2
80 60 5 2
75 50 13 4
70 100 15 9
65 50 35 20
6O 100 5O 35 3
55 100 80 80 12
50 200 100 150 35
45 150 200 200 70
40 300 300 300 180
35 800 400 400 400
30 2 000 800 800 1 300
25 6 000 4 000 4 000 2 000
20 10 000 9 000 9 000 11 000
15 15 000 15 000 15 000 15 000
10 20 000 25 000 25 000 25 000
2
6
22
120
450
1 400
6 000
17 000
30 000
Total 55 020 55 000 55 000 55 000 55 000
Miles 64.6
NOTE--100% = 1900 ft-lbs (2576 N 9 m) for calculations.
using appropri at el y scaled bendi ng moment distributions for various drivers from Table 3.
Damage at each level is obt ai ned by iteratively sol vi ng Eq 3. The total damage for one
sequence is then det ermi ned and a lifetime, in miles, proj ect ed.
Re s ul ~
Such results, shown in the form of probabi l i t y plots in Fig. 14, compare t wo candi dat e
steels on the basis of average- and l ower-bound properties. Lower - bound est i mat es were
IOO
8 0 ver
~ 6 o
'? 4 o
2 o p
I '
0 ~
I I 0 102 I 0 5 104 105
Cumulative Frequency
FIG. 13--Audi service loading spectra f or various percentile drivers [2].

FIG. 14--Service life estimates for two steels.
obt ai ned using the l ower 95% confi dence interval from a statistical anal ysi s of strain-life
mat eri al dat a sets to adjust the mat eri al constants in Eq 2. In this way, both material and
usage vari abi l i t y can be conveni ent l y i ncorporat ed in desi gn analysis. Not e the significant
di fference in expect ed l i fet i me bet ween an average (50 percentile) and severe ( l percent i l e)
driver. Al so, the performance i mpr ovement achi eved by material sel ect i on becomes less at
the shorter lives. Anot her useful output from such an analysis is shown in Fig. 15 in the
form of damage distribution hi st ograms for the various drivers. These profiles are seen to be
quite dri ver dependent and can provi de val uabl e i nsi ght s concerni ng whi ch events are most
i mport ant in a spectrum. This i nformat i on is part i cul arl y relevant to the est abl i shment of
more realistic l aborat ory si mul at i on tests. It can be argued that this approach gives a more
rel i abl e i ndi cat i on of wheel performance than the const ant ampl i t ude rotary test used for
initial component sizing; it is used rout i nel y by European manufacturers.
In the next section, a more ri gorous treatment of the various sources of vari abi l i t y is
present ed using combi ned durabi l i t y and rel i abi l i t y concepts.
Stage 3: Combined Durability/Reliability Analysis
Object i ve
Through the use of combi ned durabi l i t y and rel i abi l i t y methods, quantify variations in
component geomet ry, material propert i es, and loading, in order to project wheel lifetimes in
terms of failure probabilities.

LANDGRAF ET AL. ON AUTOMOTI VE WHEEL ASSEMBLY 17
0
c
E
0
=E
,.n
I 0 0
8 0
6 0
4 0
2 0
I 0 0
8 0
6 0
4 0
2 0
I
I
t
I %
I I I
J
7
I
I
I
: 5 %
I
I
5 0 %
I [ I
l
I
I
I
J
T e s t D r i v e r
9 7 %
lO 2 0 3 0 0 lO 2 0 3 0
% T o t a l D a m a g e
I
I Dr i v e r % Re l a t i v e Da ma g e
l I
3 0 . 5 8
5 0 0 . 0 5
9 7 0.00:5
TD 9.1
FIG. 15--Damage distribution profiles fi)r different percentile drivers.
Addi t i onal Techni cal I ssues
9 deal i ng wi t h var i abi l i t y and uncer t ai nt y usi ng rel i abi l i t y met hods ,
9 i nt egr at ed appl i cat i on o f rel i abi l i t y and dur abi l i t y met hods,
9 f at i gue l i fe in t e r ms o f pr obabi l i t y o f f ai l ur e,
9 model i ng dr i ver usage by rout e synt hesi s, and
9 sensi t i vi t y anal ysi s in f at i gue desi gn.
Backgr ound
An addi t i onal i ssue i nvol ved wi t h hi gh vol ume manuf act ur e i s t he necessi t y t o cons i der
in a quant i t at i ve wa y how t he myr i ad o f of t en subt l e c ompone nt - t o- c ompone nt var i at i ons
can i nfl uence the ser vi ce pe r f or ma nc e o f l ar ge fl eet s o f vehi cl es. Rel i abi l i t y met hods [3]
pr ovi de a r i gor ous appr oach f or c ombi ni ng t he speci f i ed or expect ed uncert ai nt i es in desi gn
and pe r f or ma nc e pa r a me t e r s i nt o pr obabi l i st i c des i gn met hods t hat can pr ovi de es t i mat es o f
t he l i kel i hood o f f at i gue f ai l ur e at pr es cr i bed l i f et i mes. Thi s i nf or mat i on pr ovi des the de-
si gner wi t h val uabl e gui dance f or opt i mi zi ng desi gns and f or sel ect i ng and cont r ol l i ng ma -
t eri al and pr oces s i ng pa r a me t e r s to assur e a gi ven saf e l evel o f per f or mance.
Techni cal Appr oach
The basi c st eps i nvol ved in c ombi ne d dur abi l i t y and rel i abi l i t y desi gn are out l i ned in Fig.
t 6 [4]. Her e, mat er i al pr oper t i es, geomet r y, and l oads dat a, al ong wi t h t hei r di st ri but i ons,
are acces s ed and di vi ded i nt o st ress and st r engt h el ement s. Th e s e st ress and st r engt h f unc-
Copyright by ASTM Int' l (all rights reserved); Thu May 8 10:42:12 EDT 2014

Material Properties ]
Finite Element Analysis ~
and Validation J [ Stmin-Life P . . . . . ]
r . o o m v a b l e l
I I
' @ '
f ~
I I
L_ 1
Design Point, ]
Component Re ab ty
FIG. 16--Flow chart for combined durability~reliability analysis.
tions are then combi ned into the performance function, G(R,S), devel oped f l om Eqs 1 and
3, whi ch is cal l ed r epeat edl y by the rel i abi l i t y anal ysi s pr ogr am in order to arrive at a design
poi nt (most probabl e failure point) by an iterative scheme [3]. Det ai l s of the devel opment
and appl i cat i on of this procedure are present ed in Ref 4. The results and i mpl i cat i ons of the
approach wi l l be emphasi zed here. For the wheel study, vari at i ons in geomet r y (thickness)
and mat eri al fatigue propert i es a r e combi ned and a pr obabi l i st i c anal ysi s carri ed out for
vari ous servi ce spectra. Incl uded are the Audi percent i l e dri ver dat a and addi t i onal dat a
generat ed by MI RA in Engl and for a vari et y of cont i nent al and British routes [5]. These
routes are t abul at ed in Table 4 in the form of rel at i ve bendi ng moment - cycl e count
distributions,
Results
Probabi l i t y of fai l ure-fat i gue life pl ot s for various percent i l e drivers are shown in Fig. 17.
The steel and al umi num di spl ay qui t e si mi l ar trends, with the 1% dri ver produci ng a failure
pr obabi l i t y of about one in 100 at 200 000 mi l es for steel wheel s and 7 in I 000 for alu-
minum. An average dri ver produces failure probabi l i t i es of 3 in 100 and 2 in 1 000, respec-
tively, for the steel and al umi num at si mi l ar mileage.
Usi ng the dat a in Table 4, it is possi bl e to synt hesi ze the various route segment s into a
real i st i c servi ce hi st ory for different t ypes of driver usage. Three such combi ned routes were
devel oped: an " aver age" route, composed of 40% freeway, 30% highway, 15% city, 10%
rural, and 5% mount ai nous routes: a "t r avel er " route, with a respect i ve breakdown of 60,

T
A
B
L
E

4
-
-
M
I
R
A

w
h
e
e
l

,
s
p
e
c
t
r
a
.

z

C
o
n
t
i
n
e
n
t
a
l

M
o
m
e
n
t
,

%

F
S

T
u
r
i
n
i

F
a
s
t

T
u
r
i
n
i

S
l
o
w

C
o
l

d
e

B
r
a
u
s

C
o
l

d
e

B
l
e
i
n
e

L
o
n
d
o
n

B
r
i
t
i
s
h

D
e
r
b
y
,

P
a
r
t

I
I

L
e
i
c
e
s
t
e
r

A
5

U
7

7
7

m

1
0
0

1
4
6

8
5

5
4

6
9

1
1
7
2

5
4

3

5
3
4

3
8

3

0
0
5

2
3

3

7
9
0

8

4

9
3
3

T
o
t
a
l

M
i
l
e
s

3
1

6
6
5

3
8

.
.
.

2
1

,
.
,

3
0
5

8

5
1

6
3
3

2
2
5

3
6
8

[
0
5
4

7
8
4

1
5
4
0

9
3
5

1
8
6
1

3

2
9
2

2
4
1
6

3

9
2
1

4

6
9
0

3

4
6
5

6

7
[
3

3
1

6
8
7

2
6

0
3
5

3
7

5
2
4

3
8

3
2

4
4

8

5

1
7

8
0

"
"
6
2

4
6

2
9
9

4
6

5
4

9
1
2

1
9
6

1
5
2

2

4
1
4

8
9
9

2
8
1

4

5
2
4

2

7
8
9

5

4
8
2

1
6

8
8
8

2
3

0
1
4

1
5

3
5
1

4
6

7
8
0

4
1

6
2
4

1
8

5
4

4
9

9

6
6

1
8
0

1
8
0

6
5
7

1
0

3
4
5

4
9

4
2
0

5
9

[
-
-
-

O

z

c

-
I

0

0

-
I

m

N
O
T
k
-
-
1
0
0
%

=

1
2
1
0

f
t
-
l
b
s

(
1
6
4
0

N

9
m
)

f
o
r

c
a
l
c
u
l
a
t
i
o
n
s
.

-
7
-

m

m

F

m

c
o

r

C
o
p
y
r
i
g
h
t

b
y

A
S
T
M

I
n
t
'l

(
a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
)
;

T
h
u

M
a
y

8

1
0
:
4
2
:
1
2

E
D
T

2
0
1
4
D
o
w
n
l
o
a
d
e
d
/
p
r
i
n
t
e
d

b
y
U
n
i
v
e
r
s
i
d
a
d

D
e

C
h
i
l
e

p
u
r
s
u
a
n
t

t
o

L
i
c
e
n
s
e

A
g
r
e
e
m
e
n
t
.

N
o

f
u
r
t
h
e
r

r
e
p
r
o
d
u
c
t
i
o
n
s

a
u
t
h
o
r
i
z
e
d
.

20 CASE STUDI ES FOR FATIGUE EDUCATI ON
l f f 1
10-2
l f f 3
I f f 4
[ 0- 5
/
l f f 6
/
1 0 4
1 , 0 0 0
(a)
S A E I 0 1 0 S t e e l
/
. . . - -
1 % - - ' "
. . . . o - ' "
1
/
j f J / - '
/ / ' "
/
/ /
/ /
/ /
/
9 97%
, t / . . . . . I . . . . . . . , I
1 0 , 0 ( 3 0 1 0 0 , 0 ( 3 0 1,000,000
Fatigue Life(Miles)
5 4 5 4 A l u m i n u m
l i f t . . . . . . . . , . . . . . . . . ,
l f f 2
. . . . . . . - " "
10"3 1 % .-" "" " " " ~ "
u~ 1 0 ,4 ..- / .
>. /
/ -
o / 5o~ /
I O s
/ /
J " "97 %
10 -6 / /
10-7 , , / , , , , i . . . . . . . . ~ . . . . . . . .
1 , 0 0 0 1 0 , 0 0 0 100, 000 1,000,000
(b) Fatigue Life (Miles)
FIG. 1 7 - - Fa i l u r e probability estimates f or different drivers.
20, 15, 5, and 0%; and a "l ocal " route ent ai l i ng a 10, 20, 35, 20, and 15% rel ati ve route
di stri buti on. Resul t s for the st eel and al umi num are s hown in Fig. 18. Agai n, si mi l ar trends
are noted bet ween the mat eri al s wi t h about a f act or- of - f i ve di f f erence in fai l ure probabi l i ty
not ed be t we e n the travel er and l ocal driver at 100 000 mi l e s .
Exe r c i s e s o f this t ype can provi de val uabl e gui dance to desi gners in c ompone nt opt i mi -
zat i on by relating probabl e s e r vi c e perf ormance t o vari ati ons in des i gn, materi al , and man-
ufacturing parameters. Any number o f "what i f . . . ?" s cenari os can be readi l y perf ormed
and s ens i t i vi t y anal ys es carried out t o generat e quantitative rel at i onshi ps be t we e n c ompone nt
des i gn, producti on parameters, and structural durability. The s e cons i derat i ons hol d speci al
rel evance to product qual i ty and warranty i s s ues .

LANDGRAF ET AL. ON AUTOMOTI VE WHEEL ASSEMBLY 2 1
1 0 1
1 0 4
1 0 ,3
10 4
10-5
1 0 6
100
SAE 1010 Steel (Cold Worked)
......................................... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . : ~ . . . . . . .
/ -
y . . . . . . . . . .
................................ ............... s : i ........ : . . . . . . . . . : : ; : 71 . . . . . . . . . . . . . . . .
/ , " ..... i-- .....
............................. i i i ...................
1,000 10,000 100,000 1,000,000
( a ) F a t i g u e L i f e ( M ile s )
5454 Al umi num (Cold Worked)
10 -I . . . . . . .
i . . . . . . . . . . . . . .
i / - . . . . . . .
~ 3 - / . . . . . . . . . '
1 0 .............................. i . . . .
10" 6
100 1,000 10,000 1130,000 1,000,000
( b ) Fatigue Life (Miles)
F I G . 1 8--Failure probability estimates.fi~r synthesized service routes.
Co nc l us i o ns
Taken together, the three phases o f wheel devel opment presented provide a range o f ed-
ucational opportunities for engi neeri ng students. In addition to gai ni ng a rather broad vi ew
o f product devel opment and evaluation met hods currently applied in industry, the study
provi des a framework for presenting and devel opi ng treatments o f several technical areas o f
i mportance in engi neeri ng practice. Exampl es include:
9 materials sel ecti on in desi gn,
9 material processi ng effects,
9 strain-based fatigue methods,
9 finite el ement methods (plasticity corrections),

9 correlation of analysis with experimental (strain gage) measurements,
9 laboratory testing,
9 service loads measurement and usage,
9 estimating service performance (cumulative damage), and
9 deterministic versus probabilistic design methods (dealing with variability).
In a capstone design course, the study can be used to demonstrate open-ended problem
solving in a mode that emphasizes the importance of integrating and synthesizirig the knowl-
edge base and skills acquired throughout a student' s undergraduate educational experience.
Opportunities for computer-based exercises are many. Here students can gain experience in
developing and manipulating data bases for material properties, service histories, and so
forth, performing damage analyses using standard spreadsheets, and presenting technical
information in clear and concise ways using computer graphics. With regard to computer
usage, it is vitally important that students gain an appreciation of both the power and limi-
tations of computational methods and the numerical "accuracies" (real or believed) obtain-
able from them.
Finally, the wheel study can be extended to incorporate broader design issues such as the
need to control the unsprung mass in ground vehicles in order to meet ride and handling
objectives. Simple dynamics models of a vehicle system can be used to demonstrate how
vehicle sprung and unsprung weights and suspension characteristics interact in influencing
overall system performance. In this way, another impetus for weight reduction, in addition
to fuel economy, can be introduced.
Ref erences
[1] Neuber, H., "'Theory of Stress Concentration for Shear Strained Prismatical Bodies with Arbitrary
Non-Linear Stress-Strain Law," Journal qf Appl i ed Mechanics. Vol. 28. 1961, p. 544.
[2] Wimmer, A. and Peterson. J., "'Road Stress Resistance and Lightweight Construction of Automobile
Road Wheels," SAE Paper 790713, Society of Automotive Engineers. Warrendale. PA, 1979.
[3] Ang, A. H-S. and Tang, W. H., Probability Concepts in Engineering Planning and Design. Vol. 1:
Basle Prineiples and Vol. 11: Deeision. Risk. and Reliabilit3', John Wiley & Sons, New York. 1975
and 1984, respectively.
[4] Ridder, R. L., Landgraf, R. W.. and Thangjitham. S.. "Reliability Analysis of an Automotive Wheel
Assembly," SAE Paper 930406, Society of Automotive Engineers. Warrendale. PA. 1993.
[5] Bingham. D. C. H. and Crane, C., "'Road Wheel Stresses," MIRA Research Report 1977/1. The
Motor Industry Research Association, Warwickshire, United Kingdom, 1.977.

Gary C. Saliva?
The Use of Failure Analysis and Materials
Testing in the Redesign of a Boat Trailer
Roller Arm
RE F E RE NCE : Salivar, G. C., " Th e Us e o f Fa i l ur e An a l y s i s a n d Ma t e r i a l s Te s t i ng i n t h e
Re d e s i g n o f a Bo a t Tr a i l e r Rol l er Ar m, " Case Studies for Fatigue Education, ASTM STP
1250, Ralph I. Stephens, Ed., American Society for Testing and Materials, Philadelphia, 1994,
pp. 23-31.
ABS TRACT: A successful design for a small-capacity boat trailer roller arm was extended to
a new larger-capacity boat trailer and placed into service. After operation of the trailers for a
period of time, several failures of the roller arms were reported. The failures occurred while
the trailers were loaded in service with resulting damage to the boat hulls. There was no prior
warning of the failuresl
This paper discusses the background of the components and the methods used for manufac-
turing the roller arms. It will discuss the analysis procedures used to determine the cause of
the failures. Alternatives for design changes to the roller arms will be presented, along with
the results of a test program that confirms the improvement in component life, based on the
recommended change.
KEYWORDS: beachmark, failure analysis, fatigue, fracture surface, hardness, plasticity, stress
concentration
H i s t o r i c a l I n f o r m a t i o n
A boat t r ai l er manuf act ur er had successf ul l y been usi ng a par t i cul ar hull support desi gn
for a numbe r of years. An e xa mpl e of this t ype o f t rai l er and t he suppor t st ruct ure is shown
in Fi g. 1. The desi gn uses r ol l er s to suppor t t he hul l and al l ows t he boat to be move d easi l y
on and of f t he trailer. The r ol l er s mi ni mi z e t he sl i di ng da ma ge to t he hull that may occur
wi t h t rai l ers t hat use padded f i xed- beam support s.
The t r ai l er in Fig. 1 has f our r ol l er as s embl i es at t ached to t he frame. Each as s embl y
consi st s of a pi vot beam t hat is at t ached to t he f r ame c r os s - me mbe r : t wo r ol l er ar ms ( af f ec-
t i onat el y cal l ed " b i r d s " due to t hei r V- shape (see Fig. 2)), one at t ached to each end of t he
pi vot beam, and t wo r ubber r ol l er s, one at t ached to each end of t he r ol l er arm. Thes e as-
sembl i es pr ovi de t wo axi s mo v e me n t to a c c ommoda t e t he hull as it passes over t he trailer.
The cri t i cal pi ece in this des i gn is the r ol l er arm, as shown in Fi g. 2. The cent ral hol e
pr ovi des f or at t achment to t he pi vot beam and al l ows for mo v e me n t of t he arm.
The manuf act ur er of f er ed t rai l ers wi t h wei ght capaci t i es of 2000 to 28 910 N. Ini t i al l y,
t he r ol l er suppor t desi gn was l i mi t ed to t he l ower capaci t y t rai l ers, wi t h the l arge capaci t y
t rai l ers usi ng padded f i xed- beam support s. A deci s i on was made to ext end this successf ul
~ Professor, Florida Atlantic University, Department of Mechanical Engineering, Boca Raton, FL
33431-0991.
23
Copyright 9 1994 by ASTM International www.astm.org

FIG. l - - Boat trailer with roller supports.
roller-support design to the larger-capacity trailers, with no change in the geometry or size
of the roller arms and the parts of the roller assemblies. After introduction of the large
capacity trailers with these roller assemblies, a number of failures of roller arms occurred in
relatively short order, resulting in damage to the hulls of several boats and subsequent liti-
gation. At this point, the manufacturer contacted a private consultant to review the design
of the roller arms and determine the cause of the failures.
Student Questions
1. What information would you request from the trailer manufacturer in order to be able
to analyze the failures?
FIG. 2- - Boat trailer roller ann (original design).

SALIVAR ON FAILURE ANALYSIS AND MATERIALS TESTING 2 5
2, What procedures would you use to initially observe and handle any failed parts that
you receive?
While it is always desirable to do a complete analysis of the failure, in terms of materials
characterization and fracture analysis, often it is not possible due to financial or informational
constraints placed on the consultant by the customer. In this case, the manufacturer specified
their interest in determining the cause of the failures, if possible, but within a very limited
budget. Also, the manufacturer did not have good knowledge of the loads experienced by
the roller arms in service. This many times is the "typical" failure analysis scenario presented
to the consultant. At no time does the consultant have an unlimited budget; however, the
budget is often such that complete chemical analyses, moderate testing programs, or fracture
surface analysis procedures (that is, scanning electron microscope (SEM)) arc not possible.
In addition, the consultant almost never has all of the necessary information, and many times
important information about the service loads is unavailable. The problem becomes one of
determining as much information about the failure as possible within the cost and infor-
mational constraints. In these cases, hopefully, the limited analysis will be sufficient to de-
termine the cause of the failure and recommend design and material changes, or both to
alleviate the problem. The consultant needs to inform the client that the financial and infor-
mational constraints, or both may not allow a complete determination of the cause of failure.
This case study is submitted for discussion as an example of a failure analysis that has severe
financial restrictions and is lacking basic information on the service loads experienced by
the components.
M a t e r i a l I n f o r m a t i o n a n d M a n u f a c t u r i n g o f t h e R o l l e r A r m
Several new and failed roller arms were forwarded by the manufacturer for investigation.
A plan drawing of the roller arm was supplied, a schematic of which is shown in Fig. 3.
An estimate of the operational loads experienced by the roller arms in service was not
provided, due to the manufacturers lack of this information, as mentioned previously.
The material was hot rolled 1040 steel bar stock with a 1.91-cm diameter. The center
section of the roller arm was flattened in a press to the final 0.95 cm thickness to fit in the
pivot beam clevis. The bar was then hot-formed into the V configuration by bending around
a square-corner die in a press to the desired angle and dimensions. This operation was
performed at the center of the roller arm and 5.72 cm in from each end of the arm (Fig. 3).
The roller retention ears were also pressed into the ends of the arm. A 0.48-cm diameter
hole was drilled in each end of the bar for a cotter key to secure the roller. The center
attachment hole was punched into the flattened center section of the arm and the surface
was left in the as-punched condition. Finally, the entire roller arm was galvanized for cor-
rosion resistance.
i !
FIG. 3- - Schemat i c of a boat trailer roller arm.

Student Questions
1. Are you comfortable with the material and manufacturing information that has been
supplied, or would you ask for more specific information in certain areas? If so, where?
What if this information is not available?
2. Which tests might you suggest to verify the materials composition and properties?
3. Which tests would you specify if the manufacturer has placed you on a very limited
budget? Are tests always necessary?
Investigation and Failure Analysis
Initially in the investigation, the fracture surfaces of the failed roller arms were observed
visually with a 10X eyepiece. A typical surface (Fig. 4) will exhibit the classic fatigue-
fracture surface appearance [1,2]. Upon further investigation with a low-power stereomicro-
scope, the crack origin was located at the upper surface of the roller arm in the plane of the
center of the clevis attachment hole. Beachmarks were evident on the fracture surface that
were indicative of fatigue. The arrows in Fig. 4 showed the beachmark that represented the
final profile of the fatigue crack at the point of instability (final failure). The fracture region
between the upper surface of the roller arm and the final beachmark exhibited very little
macroscopic plasticity, which is characteristic of fatigue. The remaining fracture surface
above and below the attachment hole exhibited ductile overload failure as was evidenced by
the significant macroscopic plasticity and shear-lip formation.
Further examination under the stereomicroscope revealed that the origin of the cracking
was coincident with a notch on the upper surface of the roller arm (Fig. 5). This notch
provided a stress concentration at the upper surface in the vicinity of the attachment hole,
which under the applied service loads resulted in fatigue cracking and failure of the com-
ponent. This behavior was consistent for all of the roller arms. Again it should be noted that
the manufacturer did not have knowledge of the magnitude of the loads experienced by the
roller arms, nor did he have a cycle count for the failed parts.
Observing the schematic of the design drawing of the roller arm (Fig. 3), it is evident that
a sharp notch was designed into the component. In production, the notch was much less
severe; however, it was still significant (Fig. 5). As mentioned earlier, the V configuration
FIG. 4- - Fract ure smf ace of a roller arm.

SALIVAR ON FAILURE ANALYSIS AND MATERIALS TESTING 27
FIG. 5--Not ch on upper surfitce ~" roller arm.
was obtained by hot-forming the bar stock around a square-cornered die which resulted in
the notch in the upper surface of the roller arm.
Hardness measurements were taken on the flattened center section of the roller arms in
the vicinity of the notches and resulted in Rockwell B scale readings of 78 to 80. This was
consistent with the strength levels of hot rolled 1040 steel [3]. Since the problem was not
believed to be material related, and to minimize expenses, chemical analysis was not per-
formed on the material. Due to the convincing evidence of a fatigue failure under microscopic
evaluation, the fracture surfaces were not observed under an SEM. Whereas many times
SEM analysis is necessary to determine the mode of fracture, and striation counting is
performed to determine a cycle count for the failure, they were not required in this case and
represented unnecessary expense.
Student Question
I. Given the new information in this section, are there further comments that you can
make in regard to the necessity of the tests that you discussed previously for deter-
mining the material composition and properties?
Cause of the Failure
The cause of the failure is fatigue cracking originating in a notch in the upper surface of
the roller arm that is the result of the manufacturing operation used to deform the bar stock.
When this design was originally used for low capacity trailers, there were no failures of
the roller arms. The lack of failures suggests that the combination of the service loads and
the stress concentration due to the notch was not severe enough to cause fatigue cracking
within the current lives of the low-capacity trailers. Only when the design was extended to
the large-capacity trailers with higher loads did failures begin to occur. It would be desirable
to perform simple stress calculations for this geometry and compare them to stress versus
cycles-to-failure data for the material to estimate the life of the roller arms. However, in this
case, no load or accumulated cycles information was available for the parts. The question
then arises as to how to resolve the problem in the large-capacity trailers?

Student Questions
1. How would you estimate the stresses in the roller arms at the critical location? Discuss
the procedures that could be used to estimate the stress concentration factor for the
notch.
2. What possibilities for changes to the roller arms can you suggest to eliminate the failure
problem?
The local stress at the failure location in the roller arm in the large-capacity trailers needs
to be reduced in order to prevent the failures. This can be accomplished in two ways. First,
the bar diameter can be increased to lower the service stress. Changing the bar diameter
would require a significant amount of change in the design because it would affect the size
of the roller arm, the attachment clevis, and the rollers themselves. The second alternative
is to eliminate the stress concentration in the roller arm that would lower the local stress
and possibly eliminate the failures. This change is the easiest to investigate with minimum
reconfiguration of the roller assembly.
Recommendat i on
The recommendation was made to the manufacturer to eliminate the notch on the upper
surface of the roller arms by deforming the arms to shape around a die with a large radius
flastead of a square-corner. The reduction of the stress concentration was believed to be
significant enough to lower the stress and eliminate the early failures. This belief is based
on the notches being inadvertent and not part of the design, in addition to the fact that all
of the cracks originated at these notches.
The manufacturer agreed to the production change to the roller arms, as shown in Fig. 6.
This new procedure resulted in the complete elimination of the V configuration above the
attachment hole in the roller arm (Fig. 6). The manufacturer requested that some limited
fatigue testing be done to quantify the difference in life as a result of the change.
Student Questions
1. What type of testing would you suggest to investigate the difference in life, if any,
between the new and the old design?
FIG. 6--Boat trailer roller arm (new design).

2. What test conditions would you specify based on the information you have from the
manufacturer?
3. What statistical procedures are available to you to determine the number of tests that
need to be performed to statistically conclude that any differences in life between the
new and old designs is significant? Is it necessary or cost effective to go this far in
every failure case?
A three-point bend apparatus was constructed to load the roller arms in the manner in
which they were loaded in service (Fig. 7). Rollers were not attached to the arm during
testing. As stated before, the manufacturer did not know the magnitude of the in-service
loads, and therefore, specified what he thought would be a reasonable maximum test load
(6670 N) for comparison purposes between the old and new designs. Constant amplitude
fatigue tests were performed with a stress ratio of +0.1 (667 N minimum load) at a frequency
of 10 Hz.
Four roller arms of the original configuration (Fig. 2) were tested to failure, with the
results shown in Table 1. The lives ranged from approximately 63 000 to 223 000 cycles.
All of the cracks in these tests originated from the notch in the upper surface of the roller
arms, identical to the in-service components. As is seen in the results, there is significant
variability in the lives of these parts, showing the lack of uniformity in the manufacturing
process, in terms of the bar size and the notch configuration.
FIG. 7--Three-point bend test apparatus.

TABLE 1--Fatigue tests o f roller arms (ol d design).
Specimen Number Cycles to Failure
2 ~ 118 890
4 223 420
1 62 820
3 92 710
Norz--P .... 6670 N; Prom = 667 N; frequency = 10 Hz, square wave-
form, three-point bending.
"Specimens were tested in random order mean life = 111 525 cycles.
Four roller arms of the new configuration (Fig. 6) were also tested. It was decided to
consider the test a run-out when the life reached one million cycles (four times the life of
the longest lasting old design roller arm). Each of the four new configuration specimens
reached this run-out life without failure and were overloaded at the completion of the tests
to check for crack initiation. The four specimens yielded between 14 720 and 17 125 N,
without any cracks found in the upper surface of the arm. As a comparison, a new config-
uration roller arm that was not fatigue cycled, yielded at 14 767 N when it was overloaded.
These tests confirm that the manufacturing change to remove the notch and the resulting
stress concentration in the roller arm significantly improves the fatigue life of the component.
A reduction in stress concentration will lower the local stress and extend the life. The new
design was placed into service without any further failures being reported to date.
Summary and Discussion
Analysis of the boat trailer roller arms indicated that the failures were the result of fatigue
cracking, and subsequent fracture. The origin of the failures was determined to be an inad-
vertent notch in the roller arm that was the result of the deformation process used during
manufacture. Constant amplitude fatigue testing under three-point bending in the laboratory
showed that this notch was responsible for the initiation of fatigue cracks and the very early
failure of the arms within a range of 63 000 to 223 000 cycles. A design change in the
manufacturing operation was suggested, which eliminated the notch and the V configuration
of the roller arm. Fatigue tests of the new design showed no failures of the roller arms within
a runout period of one million cycles. The roller arms were subsequently overloaded with
no evidence of cracking found. The new design was placed into service with no failures
reported.
This case study presents the important example that design and manufacturing planning
need to be performed together, with an appropriate understanding of the influence of stress
concentration. All too often, stress concentrations are overlooked in the design, even by
experienced engineers, which subsequently provide the ~ource for the origin of the failures.
Care must also be exercised that inadvertent stress concentrations are not built into the
components during manufacturing operations, as was the case in this example. These simple
considerations need to be addressed to prevent what all too often results in early catastrophic
failures.
This case study also presents an example of a situation where there is little money for
analysis and important service-load information is unavailable. At this point, the experience
of the failure analyst must come into play. Basic solutions can be presented that will help
the situation, but may not eliminate the failures. A small test program can help to confirm
an improvement in part life but the actual change will only be known by the accumulation

of i n-servi ce history. Many t i mes, this is all t hat the failure anal yst can do under t he vari ous
const rai nt s that he is pl aced.
St ude nt Que s t i ons
1. The failure anal yst is oft en const rai ned by budget, in t erms of the anal ysi s procedures
that are avai l abl e for a part i cul ar proj ect . Di scuss the l i mi t at i ons this pl aces on the
anal ysi s and the abi l i t y to det er mi ne the cause of fai l ure concl usi vel y.
2. What are your vi ews as to how t he fai l ure anal yst shoul d deal with budget si t uat i ons
wi t h the cl i ent ?
Re f e r e nc e s
[1] Wulpi, D. J., Understanding How Components Fail, American Society for Metals, Metals Park, OH,
1985, p. 124.
[2] American Society for Metals, "Failure Analysis and Prevention," Metals Handbook, Ninth Edition,
Vol. I 1, Metals Park, OH, 1986, p. 104.
[3] American Society for Metals, "Properties and Selection of Irons and Steels," Metals Handbook,
Ninth Edition, Vol. 1, Metals Park, OH, 1978, p. 204.

Ki n Li ao ~ and Kennet h L. Rei f sni der'
Design of a Composite Hip
Long-Term Performance
Prosthesis for
REFERENCE: Liao, K. and Reifsnider, K. L., "Desi gn of a Composite Hip Prosthesis for
Long-Term Performance, " Case Studies.[br Fatigue Education, ASTM STP 1250, Ralph I.
Stephens. Ed., American Society for Testing and Materials, Philadelphia, 1994, pp. 32-52.
ABSTRACT: In this paper, a pert'omlance simulation model is described for the design and
simulation of the long-term behavior of a composite hip prosthesis (an engineering device for
ilnplant in the human body), that is subjected to repeated mechanical and enviromnental loads
during service, The life prediction method for the composite hip prothesis is developed from
a critical element concept, a cunmlative, mechanistic approach for predicting the residual
strength and life of composite material systems subjected to cyclic loads. Results from the
experimental characterization of prototype specimens are also presented, and how that infor-
mation can be used for the life-prediction scheme is discussed.
KEYWORDS: polymeric composite materials, hip prosthesis, compression-compression fa-
tigue, fatigue damage, life prediction
Ost eopor osi s and ost eomal aci a r ound in the el derl y popul at i on are t wo maj or causes of
i nt r acapsul ar fract ure of the neck of the femur. The porot i c and soft er bone st ruct ure r esul t i ng
tio111 these di seases suggest a decrease in l oad- car r yi ng and ener gy- abs or bi ng capaci t i es;
therefore, the bone is mor e suscept i bl e to fracture. A numbe r of surgi cal t r eat ment s can be
empl oyed to repai r such a fi' acture. However, it is gener al l y bel i eved that el derl y pat i ent s are
mor e l i kel y to benefi t f r om total hi p art hropl ast y [/ ]. Such t r eat ment is al so empl oyed for
pat i ent s wi t h ost eoart hri t i s. Thi s is i l l ust rat ed schemat i cal l y in Fig. 1. There are over 200
000 such oper at i ons a year i n the Uni t ed States al one, and that numbe r is expect ed to i ncr ease
wi t h rapi d agi ng of the popul at i on st ruct ure [2,3]. For this reason, i mpr oved and mor e dur abl e
hi p prost heses are i n demand for pat i ent s who need such t reat ment .
Met al l i c hi p prost heses now used in hi p art hropl ast y have a numbe r of di sadvant ages. The
st i ffness of the i mpl ant mat er i al is rel at i vel y hi gh compar ed to the cort i cal bone, the l oad-
car r yi ng t i ssue of the human femur. As a resul t of this mi s mat ch in mechani cal propert i es,
the i mpl ant carri es most of the appl i ed l oads and the suppor t i ng bone is not properl y stressed.
Thi s so-cal l ed st r ess- shi el di ng effect is bel i eved to be the r eason for bone resorpt i on, whi ch
may l ead to a hi gher i nci dence of i mpl ant l oos eni ng and fract ure of the prosthesis. In ad-
di t i on, pat i ent s may exper i ence hyper sensi t i vi t y as a resul t of rel ease of met al l i c i ons f r om
the prost hesi s, caused by wear [4,5].
Wi t h advances i n mat er i al s sci ence, f i ber - r ei nf or ced pol ymer i c composi t e mat eri al s are
bei ng consi der ed as candi dat e mat er i al s for maki ng hi p prot heses. The st ruct ure of a l ami -
nat ed composi t e mat eri al is i l l ust rat ed in Fig. 2. The st ruct ure of a si ngl e layer, a lamina, is
Research associate and Alexander Giacco professor, respectively, Virginia Polytechnic Institute and
State University, Engineering Science and Mechanics Department, Materials Response Group, Blacks-
burg, VA 24061-0219.
32

LIAO AND REIFSNIDER ON DESIGN OF A COMPOSITE HIP PROSTHESIS 33
FIG. l - - Schemat i c representation of total hip arthroplasty.
composed of continuous fibers embedded in a polymeric matrix. Typical values of the di-
ameter of an AS graphite fiber and lamina thickness are 7 i~m and 0.13 mm, respectively.
These single layers (laminae) can be put together in different ways (that is, with fibers in
different directions) and cured under heat and pressure to form a laminate. The mechanical
properties (such as stiffness and Poisson' s ratio) and strength of a laminate with a specific
arrangement of laminae differs from one another. For instance, the stiffness of a unidirec-
tional (all the laminae are in the same direction) and quasi-isotropic (laminae arranged in
the - 45, 45, 0, and 90 degrees directions such that its in-plane extensional stiffness are the
same in all directions) composite laminate made from T300/5028 graphite (fiber)/epoxy
(matrix) are about 180 and 70 GPa, respectively [6].
Prostheses made from fiber-reinforced composite materials possess some advantageous
characteristics that their metallic counterparts do not: the stiffness of a polymeric composite
material can be tailored to provide a state of stress in the proximal femur closer to physio-
logic level, compared to materials such as alloys (for example, cobal t / chrome/ mol ybdenum
zJ
,eatand p,esso,e i ' ' / / "
laminate
FIG. 2--Structure r a composite laminate.
- X
f

alloys) and stainless steel, which are widely used for making hip prostheses; the strength a
composite prosthesis can also be tailored according to service loads, and many polymeric
composites are resistant to harsh human body environments. In addition, they are superior
in damage tolerance and fatigue performance, as well as being more biocompatible [4,5].
When designing an engineering structure, three basic sequential questions are frequently
asked by engineers: (1) How should the structure be made? (2) How strong is it (and/or,
for structures where stiffness is a basic concern, how stiff is it?)? and, (3) How long can it
last? If these three inter-related questions are properly addressed, a structure will serve its
purpose properly, at least in theory, if not in practice. The last question is of particular
importance for structures subjected to long-term repeated loads (which may be a combination
of mechanical, thermal, or chemical loads during service). If one can tell in advance when
the structure should be retired from service, undesired events associated with structural failure
can be prevented. Moreover, if a comprehensive engineering scheme can be established to
address these questions, more product competitiveness can be achieved in terms of cost and
performance. In this paper, we attempt to construct such an engineering scheme for designing
and simulating the performance of a composite hip prosthesis under repeated mechanical
and environmental loads in the human-body environment.
How Shoul d it be Made?
Before answering the first question, the loading environment for a hip prosthesis must be
well understood. The biomechanics of hip prosthesis has been studied in detail by many
investigators [7]. A comprehensive review on the subject can also be found in Ref 8. Briefly,
under static conditions, (for instance, in one-legged stance), the prosthesis is subjected to a
resultant mechanical force (of body and muscle forces) acting at an angle to the anterior and
medial plane of the prosthesis, on the head of the prosthesis. This is illustrated in Fig. 3.
The loading angle may vary with activities. The resultant force on the head of the prosthesis
can be resolved into a force acting normal to (hereafter referred to as the out-of-plane force)
and one parallel to the anterior plane (hereafter referred to as the in-plane force). The angl es
between the resultant joint force and the in- and out-of-plane forces are in- and out-of-plane
FIG. 3--Nomenclatures and joint forces for a hip prosthesis.

load angles, respectively. The in- and out-of-plane forces acting on the head of the prosthesis
result in in- and out-of-plane bending moments, torsional, shear, and compressive forces.
These moment s and forces vary from point to point along the prosthesis, resulting in varying
stress states in the structure. In addition to the mechanical loads, boundary conditions for
the prosthesis must also be considered, as the prosthesis is either cement ed or press-fitted in
the femoral canal. During its service, a prosthesis is subjected to long-term repeated me-
chanical loading, moisture, and temperature. The influence of these factors on the mechanical
behavior of the composite hip prosthesis will be discussed in a later section.
Several preliminary design considerations for composite hip prosthesis are shown in Fig.
4. Configuration A is a "flat-plate" design, Configuration B is a "curved-pl at e" design, and
Configuration C is somewhat in between, a "rot at ed-pl at e" design. Each of these three
configurations may have its own advantages and disadvantages. For instance, the flat-plate
design is more resistant to in-plane bending, but is stiffer; the curved-plate design may be
less stiff in response to in-plane bending moment, but at the same time is weaker; and the
rotated-plate design may possess additional freedom in balancing stiffness and strength, but
this may also imply additional complexity, from the standpoint of manufacturing.
We choose the flat-plate configuration for our investigation. A prot ot ype specimen is ob-
tained by cutting from a cured laminate, and its cont our is machined, according to the design,
based on the geomet ry of the femoral canal. The basic design approach is to first determine
the stress distributions in prostheses made from laminates of various stacking sequence and
the associated stiffness and strength. Stiffness and strength of various laminate configurations
are compared and an opt i mum design is selected. Such a design is one with a balanced
stiffness and strength such that it is more compatible to the stiffness of the femoral bone
structure and possesses the required strength to sustain service loads. Specifically, the lam-
inates picked for consideration in this study were selected from a study of dozens of pos-
sibilities, with the criteria that the fatigue life at four times average body weight should be
more than one hundred million cycles and that the stiffness should be as low as possible,
given that constraint.
In order to design a composi t e hip prosthesis effectively, a stress-analysis model was
developed using a strength-of-materials approach. The structure of the composite prosthesis,
which consists of more than 100 individual plies with different ply orientations, is highly
~
A. flat plate
~
B. curved plate
U ~ C. rotated plate
FIG. 4--Three basic design considerations: (a) flat plate, (b) curved plate, and (c) rotated plate.

3 6 CASE STUDIES FOR FATIGUE EDUCATION
anisotropic. Considerable computation time would be required if a linite element method
(FF, M) was used to evaluate the stress state of each individual ply along the prosthesis. This
is a major obstacle with the use of the FF.M as an effective design tool. For this reason, a
strength-of-materials approach was chosen.
The general approach for the stress-analysis model is shown schematically in Fig. 5. The
prosthesis is assumed to be embedded in a layer of elastic foundation. The geometry of the
structure is tirst detined. This includes description of the longitudinal profile of the prosthesis
by polynomial equations and calculation of cross-sectional geometric properties, such as the
area moment of inertia. In- and out-of-plane applied loads are delined next. Under the three-
dimensional applied load, global stress and moment distributions on each arbitrary chosen
section along the prosthesis are determined. Curved-beam theory is used in this model to
account lbr curved geometry of the prosthesis. Beam on elastic foundation approach is used
to simulate the supporting bone structure for the prosthesis. Global stress and moment com-
ponents determined are then transformed into the laminate coordinate system at three loca-
tions of a section, namely, an element on the lateral, center, and medial curve. Laminate
analysis is performed at each of these three locations to determine stress and strain com-
ponents of each individual ply. A number of phenomenological strength theories (such as
Tsai-Hill, maximum stress, and maximum strain) are computed in association with the stress
state of each individual ply to predict the strength of the structure.
Ho w St r ong and Ho w St i f f is it?
Typical strain distributions along the medial and lateral surfaces of a prototype hip pros-
thesis, as determined tYom the stress-analysis model, are shown in Fig. 6. Note that the
decreasing trend in both distributions beyond 3.8 cm, as the prosthesis is supported by the
elastic foundation from that location to its distal end. It is obvious from Fig. 6 that a local
high-stressed area (represented by the local high strain) is tound at the mid-neck region (at
about 2.6 cm from the head). The results were verified experimentally by scanning the
surface of a cyclically stressed hip prosthesis using a thermographic stress imaging system
moments / ~ laxialforce
shear forces
appiied load
tors on
x
% . " , , I
% ,,, 14
, /
FIG. 5--Schematic representation of the stress-analysis model.

LIAO AND REIFSNIDER ON DESIGN OF A COMPOSITE HIP PROSTHESIS 3 7
% strain
0 . 4
0 .2
(0.2)
(0.4)
(0.6)
I
2
U 0 p O e 0 0 ' e ' a s ' ' c ' O U i 0 a i ' O /
[ la te ra l~ 3 u rf . . . . e d ia l~ . u r f a c e ]
I I I I I I
4 6 6 1 0 12 1 4
d i s t a n c e f r om prosthesis h e a d ( cm)
FIG. 6-..-Strain distributions along a composite hip prosthesis.
16
known as stress pat t ern anal ysi s by t hermal emi ssi on (SPATE). The SPATE system is de-
vel oped based on a t hermoel ast i c t heory by Thomson 19], whi ch states that when a homo-
geneous, i sot ropi c, l i near-el ast i c mat eri al is cycl i cal l y l oaded under adi abat i c condi t i ons
within its el ast i c range, the reversi bl e change in the t emperat ure of the mat eri al is propor-
tional to the sum of its pri nci pal stresses. The SPATE syst em uses a hi ghl y sensitive, infrared
phot on det ect or to sense the i nfrared emi ssi on from the speci men surface, and those signals
can be convert ed to stress values by a conversi on fact or [10,11]. A result of the SPATE i mage
is shown in Fig. 7, whi ch represent s the qual i t at i ve change in sum of pri nci pal stresses in
the mat eri al surface, as i ndi cat ed by the scale on the right. Not e that the local stress peak is
at mi d neck (i ndi cat ed by the arrow), whi c h is in agr eement with the resul t of stress anal ysi s.
To choose an opt i mum desi gn with bal anced st rengt h and stiffness, one may compare,
using the st ress-anal ysi s model, the strength of the cri t i cal region (that is, the neck region)
and the associ at ed l ami nat e stiffness for various l ami nat es. A strength versus stiffness pl ot
for a number of l ami nat es at the neck regi on is shown in Fig. 8. The stiffness and strength
(det ermi ned using t he maxi mum strain criterion, whi ch states that the mat eri al is said to
have fai l ed i f the appl i ed strain exceeds its failure strain) for all dat a poi nt s that appear on
the pl ot are nor mal i zed to the stiffness and strength of a quasi -i sot ropi c laminate. The stack-
ing sequence for these l ami nat es are as fol l ows (the pl y angl e is defi ned with respect to the
prost hesi s axis, see Fig. 9):
9 Design l - - o p t i mu m desi gn
9 Design 2 - - I - 4 5 / 0 / 4 5 / 9 0 1 ~
9 Design 3 - - [ 8 0 / 6 5 / 4 5 / 2 5 / 1 0 / - 2 5 ] . ~
9 Desi gn 4- - - 180/ 45/ 10/ - 251. ,
9 Desi gn 5 - - [ 6 5 / 4 5 / 2 5 / - 2 5 1 ,
9 Design 6 - - I 4 5 / 6 0 / - 3 0 / 9 0 ] . ,
9 Design 7- - 1- 45/ 0/ 45/ 9012~[ 80/ 60/ 10/ - 251~165/ 45/ 25/ - 251,
The subscri pt s denot es "s ymmet r i c. " For exampl e, the actual l ami nae arrangement for
Desi gn 2 is [ - 4 5 / 0 / 4 5 / 9 0 / 9 0 / 4 5 / 0 / - 4 5 ] . As can be seen from Fig. 9, a desi gn with

38 CASE STUDI ES FOR FATIGUE EDUCATI ON
FIG. 7--Stress distribution pattern in the neck region from the SPATE" system.
relatively high strength is also one with high stiffness (such as Design 5), and vice versa.
However, an ideal design would be one with high strength and low stiffness, a data point at
the upper left hand corner of Fig. 8. From the Fig. 8, Designs 1, 6, and 7, which are closer
to the upper left-hand comer of the plot, may be more desirable than those of Designs 2, 3,
4, and 5. Note that Design 7 is an unsymmetric laminate, where the stronger [65/ 45/ 25/
-25]s sublaminate is designedly chosen to support higher compressive stress in the medial-
Normal i zed st rengt h
1. 7
16 ....................................................................................................... E35 ....
1.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . 6" [ ~ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
~0 13a
1,3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 [ ] ~at a taken from
1 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i - / - . . medi at neck . . . . . . . . . . .
11
,_ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . u . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
09 I I I I I I I
0.9 ~ ~,~ ~.2 ~.3 ~. . ~ s ~.G ~.7
Normal i zed l ami nat e st i f f ness
FIG. 8--Strength versus stiffness plot for several laminate designs.

k_Y l a , e r a l L o s t h e s i s a x i s
L ~ positive angle of incidence
prosthesis axis
medial
FIG. 9- - Pl y identification scheme.
posterior corner of the prosthesis, which arises as a result of out-of-plane moment. The less
stiff quasi-isotropic sublaminate [ - 45/ 0/ 45190] 2, has the effect of lowering the structural
stiffness as a whole.
We chose Design 1, the "opt i mum design, " for further investigation. Due to the proprietary
nature of the stacking sequence, it will not be discussed in this paper. However, to facilitate
further discussion, plies in this design are identified as follows (Fig. 9):
9 Ply X- - pl i es with the lowest positive angle of incident (PAl) to the prosthesis axis (first
quadrant of the angular coordinate plane).
9 Ply Y- - pl i es with the second lowest PAI to the prosthesis axis (first quadrant).
9 Ply Z- - pl i es with the highest PAI to the prosthesis axis (second quadrant).
The prototype specimens were manufactured from C/ PEEK (carbon fiber-reinforced
poly(ether ether ketone)). The laminate consisted of approximately 100 individual plies at
its thickest section (shoulder).
How Long Can it Last?
Once a specific design has been chosen, one Would like to know what factors and to what
extent these factors may influence its long-term behavior in the human-body environment.
I f one can anticipate the long-term behavi or of the implant device under compl ex loading
conditions (mechanical and environmental), in terms of its damage mode, remaining stiffness
and strength, and failure mode, not onl y unwanted events associated with structural failure
can be prevented, but further i mprovement for the structure may also arise. To this end,
laboratory test procedures to obtain necessary information, as well as life prediction meth-
odologies to use and interpret that information, should be established [12]. Conventional life-
prediction schemes (such as equivalent damage concept) is denied in this case due to the
fact that it is almost impossible to produce a complete phenomenological data base of stress
and life associated with such factors as compl ex loading conditions, required testing time,
as well as cost. Instead, the life-prediction scheme should represent, mechanistically and
physically, the property change in a composi t e hip prosthesis and be able to predict its long-
term behavior under cyclic loading.
A life-prediction model is developed based on the concept of a critical-element model
[13], a mechanistic, cumulative damage model for predicting the residual strength and life
of composite laminates subjected to arbitrary cyclic loading. A conceptual chart of the model
is shown in Fig. 10. The failure mode of a component or structure is first established.
Generally, there are only a fairly small number that apply to specific engineering situations.

I Representative
Volume
I
l
Co n s t i t u t i v e
Behav or
I S t r , ~ , h I
F I G . 1 O--Schematic representation of the critical-element concept.
Based on t he failure mode and the damage mechani sm, a "r epr esent at i ve vol ume" in the
l ami nat ed component is identified. It is the process of cont i nuous propert y degradat i on in
this represent at i ve vol ume (as a funct i on of appl i ed cycl es) that causes reduct i on of strength,
and eventually, failure of the structure, that is to be descr i bed in the model . For instance, in
our case, the composi t e prost hesi s, that is a represent at i ve volume, is i dent i fi ed exper i men-
t al l y to be within ei t her the medi al neck or the medi al mi d-st em regi on where structural
fai l ure usual l y occurred [14,15]. Anal yt i cal l y, the l ocat i on of the represent at i ve volume, a
l ocal l y hi gh-st ressed point, can be det er mi ned from t he stress anal ysi s model .
Once the represent at i ve vol ume is est abl i shed, it is subdi vi ded into t wo t ypes of el ement s,
"cr i t i cal " and "subcr i t i cal . " Cri t i cal el ement s are defi ned as t hose part s of the represent at i ve
vol ume whose fai l ure causes fai l ure of the component . Si nce cri t i cal el ement s are the l ast
part of a component to fail (after the failure of subcri t i cal el ement s), the strength of a
component is domi nat ed by the response of the cri t i cal el ement to appl i ed load. For t ensi l e
l oadi ng of a quasi -i sot ropi c l ami nat e, for exampl e, t he critical el ement may be the zero-
degr ee plies. Subcri t i cal el ement s are defi ned as t hose parts of a component or laminate that
sustain damage duri ng the fatigue l oadi ng process, but cause redi st ri but i on of internal stress
because of fat i gue damage, such as mat ri x cracki ng. For the exampl e j ust ment i oned, the
subcri t i cal el ement s may be the off-axi s pl i es where mat ri x cracki ng usual l y occurs. The
response of the subcri t i cal el ement s is det er mi ned by mechani cs anaJysis, whi l e the response
of the critical el ement s is represent ed by phenomenol ogi cal i nformat i on and const i t ut i ve
equat i ons. Once the subcritical el ement s are defined, and i nformat i on regardi ng the damage
modes and the ext ent of damage is obt ai ned, then a damage anal ysi s can be conduct ed to
est abl i sh the state of stress in the i nt eri or of the component , that is, the stress state in bot h
the critical and the subcri t i cal el ement s, as funct i ons of the number of appl i ed l oad cycl es.
Damage modes and damage mechani sms can be assessed using some nondest ruct i ve (NDE)
and dest ruct i ve techniques. The l ocal state of stress in the critical el ement , the "el ement l oad
hi st ory, " obt ai ned at a given number of appl i ed cycl es, may be used to det ermi ne the in-
st ant aneous el ement response, that is, the "st at e of the mat eri al . " For instance, such a re-

sponse may be represent ed by an S - N curve, from whi ch the r emai ni ng life (cycl es) of the
critical el ement at the current stress level is det ermi ned. If the st at e of stress and the state
of the mat eri al have been det er mi ned t br the cri t i cal el ement s, then one should he abl e to
use some strength phi l osophy to det ermi ne the resi dual strength of the critical el ement s and,
t here/ ore, of the component .
To illustrate how one can act ual l y appl y the model concept , we begi n with Fig. 11, a
schemat i c represent at i on of the basi c rel at i onshi ps for l ami nat e fat i gue behavior. Fi gure 11
can be consi dered as a one- di mensi onal case where the resi dual strength curve of the critical
el ement (and, therefore, the l ami nat e) and the S - N curve of the l ami nat e are det ermi ned from
uni di rect i onal l oadi ng. Under an appl i ed load, S,,, the reduction in strength, A S , , of the critical
el ement at any poi nt p on the resi dual strength curve can be wri t t en as
:XS,.= ( l - ~ ) ( N) / (1)
where
S,, = strength of the uni di rect i onal coupon,
n - appl i ed cycl es,
N = cri t i cal el ement response (remai ni ng life) at current stress, S,,, and
i = nonl i near parameter.
The value of i is gener al l y about 1.2 for pol ymer i c composi t es. N is det ermi ned from
phenomenol ogi cal dat a ( S - N curve) of the cri t i cal el ement . Si nce the stress state of the cri t i cal
el ement is const ant l y changi ng as a result of stress redi st ri but i on caused by damage devel -
opment in the critical and subcri t i cal el ement s, the appl i ed stress, S,,, in the critical el ement
is a function of the number of appl i ed cycles, n. The critical el ement response, N, therefore,
is also a function of n because of the changi ng S,,. The strength of the critical el ement , S,,,
may also be cycl e dependent , due, for instance, to envi ronment al l oadi ngs.
To det ermi ne the accumul at i ve reduction in resi dual strength over a normal i zed peri od, "r,
Eq 1 is first expressed in di fferent i al form and i nt egrat ed over "r
Sa/Su
residual strength curve
of the critical element
p~ ~sl n)
S- Nc~"~ \
of the laminate ~ \
l i f e ( c y c l e s )
FIG. l 1 - - S c h e m a t i c d i a g r a m o f c r i t i c a l - e l e me n t s t r engt h- l i f e relationship.

42 C A S E S T U D I E S F O R F A T I G U E E D U C A T I O N
- S,,(n)J i \ N( n) ] d
( 2 )
Up to this point, Eq 2 represent s a one- di mensi onal resi dual strength reduction as a func
tion of appl i ed cycles. A local failure function, F~.(n), that represent s the t hree-di mensi onal
stress state (such as the Tsai - Hi l l cri t eri on) in the critical el ement may be i nt roduced to
repl ace the l ocal appl i ed st ress-t o-st rengt h ratio, S,(n)/S,(n)
s : ) 2 ~ S , . ( n ) = , [ 1 FL(n)] i - ~ d ( 3 )
The resi dual strength, S,(n), can be eval uat ed at any appl i ed cycl e by numeri cal l y i nt egrat i ng
Eq 3.
To appl y the critical el ement model concept to the composi t e hi p prosthesis, we use the
scheme shown in Fig. 12. We begi n the cal cul at i on by comput i ng the gl obal and l ocal stress
state in the prost hesi s as descr i bed in the st ress-anal ysi s model . Af t er the stress state in each
i ndi vi dual pl y at each chosen comput at i onal poi nt al ong the prost hesi s has been det ermi ned,
they are compared, so that the location of the cri t i cal el ement and the subcritical el ement s
can be identified: a pl y wi t h the hi ghest compr essi ve stress in the fiber di rect i on, ~L~, is
i dent i fi ed as the critical el ement ; all ot her pl i es are i dent i fi ed as subcri t i cal el ement s.
As the number of appl i ed cycl es increases, damage devel opment in the subcri t i cal el ement s
will cause degradat i on of mat eri al propert i es as a result of mechani cal and envi ronment al
l oadi ngs. For instance, mat ri x cracki ng due to mechani cal l oadi ng will cause reduct i on in
the pl y t ransverse and shear modul us [16-20]. Creep and moi st ure effects will pr oduce the
same consequences [21 31]. Evol ut i on of mat eri al propert i es is represent ed as a function of
IMechanical l o a d s ]
, i
4 I "
Ply S t r e s s e s
T
I I
T
Strength Reduction
A s r (~)
~
Envtronmental I
L o a d s I
Internal Stress
Redistdbutlon
I
L i f e
Residual Strength
= A p p led S t r e s s
L i f e Prediction Model for Composite Prosthesis
F I G . 12--Schematic representation of the life-prediction model f or composite hip.

applied cycles. The effect of mechanical loading, creep, and moisture on property degradation
will be briefly discussed as follows.
Mechanical Loading
Characterization of damage in laminated fibrous composite materials due to mechanical
fatigue has been the subject of numerous investigations [16-20]. In general, fatigue damage
in composite materials consists of various combinations of matrix cracking, fiber-matrix
debonding, delamination, void growth, and local fiber breakage. These damage mechanisms
are complex and very difficult to describe in a general way. From previous studies, it has
been found that matrix cracking due to compressive load is a dominant damage mode in the
medial neck area of a composite prosthesis subjected to compression-compression fatigue
loading [14,15].
These studies were carried out in a specially designed device to simulate in vivo loading
conditions for the hip prosthesis, as shown in Fig. 13. 0-degrees in-plane and 10-degrees
out-of-plane load angles were used for all the cyclic tests, according to a previous study by
Gavens and Locke [7]. The applied (compression-compression) cyclic load level ranged from
47 to 84% of the quasi-static strength of the prot ot ype specimens at R = 10, where R is the
ratio of the mi ni mum applied load to the maxi mum applied load. Penetrant enhanced X-ray
radiography and a surface replication technique were used to monitor damage development
in the hip prosthesis at selected intervals.
The medial-posterior corner of the mid-neck region is under highest local compressive
stress, as a result of a combination of in- and out-of-plane bending moment and axial com-
pressive force normal to the neck axis. Consequently, neck damage initiated from that lo-
cation. Damage occurred mainly in the form of matrix cracking along fiber directions. This
type of damage first occurred in X plies, plies with the closest orientation to the neck axis.
Matrix cracking also occurred in Y and Z plies, but usually after mid-life or in late-life.
Selected results of X-ray radiography for a specimen tested under a maxi mum cyclic load
of 75% static strength are shown schematically in Fig. 14. These drawings represent damage
development in a portion of the medial neck region, and they are typical of the prot ot ype
specimens under investigation. Thin penetrant lines in fiber directions were observed in the
neck region and were the consequence of matrix cracking along fiber directions. A micro-
graph of a sectioned neck near the medial surface is shown in Fig. 15, where a through-the-
[ . I
acetabular cup
B~.~),, metallic head
, . ~ ~ hip prosthesis
I ~ ~ _ ~ i ~ pro~ le~l er .cl ieneseer t
~
metallic bar and knobs
I
I
[ ~ ball and socket support
FIG. 13--Schematic diagram of the test fixture.

/
80% life
F]G. 14--Schematic representation of X-ray radiographs.
ply matrix crack is evident. The penetrant lines all initiated from the medial surface and
extended to the lateral side. From X-ray radiographs, the density and area covered by these
thin penetrant lines both increased with an increase in applied cycles (that is, more and
longer penetrant lines were seen), indicating damage progression. Most of the neck damage
occurred in X plies, some in Y plies. Only a few penetrant lines were seen in Z plies. A few
tiny delaminations were also observed from the radiographs.
From surface replication images, damaged plies protruded from the specimen surface as
a result of compression and shear, as shown in Fig. 16. Aided by X-ray radiographs, these
FIG. 15--Micrograph showing a matrix crack in the medial neck region.

FIG. 16--Surface replications images showing damage development in the neck region: (a) 53% life
and (b) 88% life.
damaged plies were identified to be the X plies. They first appeared in the medial-posterior
corner, extending their distribution toward the anterior side and along the medial surface,
with an increase of applied cycles. In general, results obtained from X-ray radiography and
surface replication suggested the overall damage development pattern, shown in Fig. 17.
Although the results from the nondestructive and destructive tests were qualitative, they
provided useful information for subsequent modeling. To represent transverse cracking in
the matrix material during fatigue, it is required that the transverse ply modulus, E2, and
shear modulus, G~2 , both decrease as transverse crack density increases. Such a stiffness
reduction relation also depends on stress and stacking sequence
~
pplied load
t damaged region
me osterior
plane K ~ p l a n e
arrows represent
direction of damage
development
FIG. 17--Schematic representation of the damage development pattern in the neck region.

where
= E , O'tl
(4)
E', ~ = initial t ransverse pl y modul us,
E, t ransverse pl y modul us at n cycl es, and
% = stress component s of an i ndi vi dual ply.
The same kind of equation al so appl i es for the pl y shear modul us, G,_,. Equation 4 appl i es
to all plies at all times.
C , ' e e p E f f e c t
The effect of creep on stiffness change of the composi t e hip prost hesi s has been studied
in some detail [2]. To model the t i me-dependent effect, anal ysi s si mi l ar to Yeow et al. [21]
was followed. The rel at i onshi p for the stifl' ness ratios (the ratio of the pl y t ransverse and pl y
shear modulus to t hei r initial val ues) is expressed by a master curve
4
stiffness rat i o = ~ a,,,i'" (5)
m=O
where i is creep t i me, defi ned by
= log(t) + l og( at ) (6)
and
where
a,,,, b,,, = coefficients,
t = t i me in minutes,
a r = shift factor, and
T = t emperat ure.
l og(a r) = ~ b , , , T " (7)
m=O
In the life predi ct i on model , cr eep t i me is convert ed into the number of appl i ed cycl es by
mul t i pl yi ng by the l oadi ng frequency. The stiffness ratio is mul t i pl i ed by the initial values
of the pl y t ransverse modulus, ~ , and the pl y shear modul us, G~ 2, for each i ncrement of
life fraction, n / N , to obt ai n the instantaneous stiffness values.
M o i s t u r e E f f e c t
It has been known that the presence of moisture in composi t e mat eri al s causes degradat i on
of mechani cal propert i es [ 2 1 - 3 1 ] . For exampl e, the t ransverse pl y modul us, E2, of a unidi-
rect i onal pl y is est i mat ed by the rel at i onshi p [25]

LIAO AND REIFSNIDER ON DESIGN OF A COMPOSITE HIP PROSTHESIS 4 7
wher e
l 0 f Um
- + - ( 8 )
E2 E2 s E,,(m,., 1)
v,, = mat ri x vol ume fract i on,
v~ = fiber vol ume fract i on,
E_2r = t r ansver se modul us of t he fiber,
E,,,(m,, T) = modul us of t he mat ri x at cer t ai n moi st ur e cont ent , mr, and
~_L -
wher e
T~, = gl ass t r ansi t i on t emper at ur e of the mat r i x at m,,
r~! = gl ass t r ansi t i on t emper at ur e of the mat r i x at dry ai r condi t i on, and
AT = t emper at ur e above nor mal t emper at ur e.
The shear modul us of a uni di r ect i onal ply, Gi2, can be f ound by a si mi l ar appr oach
1 v I v , , ,
- + - - ( 9 )
G,2 G f G,,,(m~, T)
wher e
G_f = shear modul us of the fi ber at dry ai r and
G(m, , T) = shear modul us of mat ri x at mr and T.
In order to appl y the af or ement i oned anal ysi s to the l i fe pr edi ct i on model , moi st ur e con-
tent, mr, must be r epr esent ed as a f unct i on of appl i ed cycl es, so that changes in E2 and G~2
due to moi st ur e can be eval uat ed at any appl i ed cycl es. At suffi ci ent l y large t i me, t, the
sol ut i on for a one- di mens i onal moi st ur e di f f usi on pr obl em is [6]
wher e
m-z~ = 1 - -~8 e ~-'K",/h~ (10)
m~ ar ~
ms = sat urat ed moi st ur e cont ent ,
K" = moi st ur e di f f usi on coeffi ci ent , and
h = t hi ckness of the speci men.
It is real i zed that the di f f usi on pr obl em for t he composi t e prost hesi s is a t wo- di mens i onal
one rat her t han one- di mens i onal . The sol ut i on for a t wo- di mens i onal moi st ur e di f f usi on prob-
l em in a composi t e l ami nat e was gi ven by Far l ey and Her akovi ch usi ng a fi ni t e di fference
met hod [28]. For si mpl i ci t y, however, t he af or ement i oned one- di mens i onal sol ut i on is used.
Upon est abl i shi ng the moi st ur e- t i me r el at i onshi p, moi st ur e cont ent at any appl i ed cycl e can
be eval uat ed, and the degr adat i on of mechani cal propert i es can al so be eval uat ed. Thi s deg-
r adat i on of propert i es wi l l al so cont r i but e to i nt er nal stress r edi st r i but i on. Al t hough t he afore
ment i oned anal yses are est abl i shed on a physi cal basi s, it is by no means a compl et e rep-

resent at i on of the real situation. For instance, Gi l l at and Brout man [23] have shown that
moi st ure gai n in gr aphi t e/ epoxy l ami nat es was i ncreased with an i ncrease in the presence
of ext ernal stress. In their anal ysi s, the moi st ure di ffusi vi t i es at di fferent l oad levels were
det er mi ned exper i ment al l y from i ni t i al sl opes of the curve of the moi st ure gain as a function
of square root of exposure t i me and the percent equi l i br i um wei ght gai ns; no expl i ci t rel at i on
was given rel at i ng the appl i ed l oad level and diffusivity. Moreover, all t he aforement i oned
anal yses are based on static condi t i ons. The i nt eract i on of moi st ure di ffusi on and fatigue is
not taken into consi derat i on. Invest i gat i ons of the effects of envi ronment on fatigue of com-
posi t e mat eri al s has i ndi cat ed that the presence of moi st ure will accel erat e matrix cracki ng,
thus affect i ng the stiffness and strength of the mat eri al [30,31].
Havi ng di scussed propert y degradat i on due to mechani cal and envi ronment al l oadi ngs, we
now return to Fi g. 12. As di scussed earlier, damage devel opment in the subcri t i cal el ement s
wi l l cause i nt ernal stress redi st ri but i on. Anal yt i cal l y, from l ami nat e theory, this reduct i on in
t ransverse stiffness wi l l cause change in the stress state of all the pl i es, that is, in the cri t i cal
and subcfi t i cal el ement s.
We now shift our attention to the cri t i cal el ement . At 0 appl i ed cycl es, no propert y deg-
radat i on in the cri t i cal el ement has t aken pl ace. At this moment , the failure function
F~.(n) = FL(0). The cri t i cal el ement response is det er mi ned by phenomenol ogi cal dat a
~..__! = a + b log(N) (11)
x,
or
where
cq~ = compr essi ve stress in the fiber di rect i on and
a, b = const ant s.
(12)
Not e that ~j~ and X,. are bot h funct i ons of the number of appl i ed cycles, n, so that the
el ement response, N, is al so a funct i on of appl i ed cycl es. Equat i on 12 is onl y used to de-
t ermi ne N(n).
Havi ng cal cul at ed the failure function FL(0), and el ement response N(0), we now ski p the
"st r engt h r educt i on" step in Fig. 12 and go back t o " pl y stress, " as it is assumed that no
damage is i nt roduced to the el ement s initially. Now, assume that the number of appl i ed
cycl es is i ncreased to n. In this instance, stress redi st ri but i on occurs in the i nt eri or of the
l ami nat e, due to damage devel opment and accumul at i on. Usual l y this woul d mean that the
cri t i cal el ement s are under hi gher stress than before. Moreover, envi ronment al effects may
cause a reduct i on of the compr essi ve strength, X,, of the critical el ement . An i ncrease in the
appl i ed stress, ~ , and reduct i on in strength X,., of t he cri t i cal el ement wi l l cause an i ncrease
in the val ue of the failure function, so that FL(n) > FL(0). The el ement response (remai ni ng
life), N(n), is al so changed by the stress redi st ri but i on. Si nce the rat i o ~/ X~. is no l onger
the same at n cycl es when compar ed to 0 cycle, N(n) is different from N(0). Usual l y N(n) <
N(0). Now we have det er mi ned FL(0), FL(n), N(O), N(n), and n. These quant i t i es wi l l ent er
the strength reduct i on equat i on (Eq 3) and the change in resi dual st rengt h is eval uat ed by
i nt egrat i ng that equat i on numeri cal l y.

RESIDUAL STRENGTH
1.1
1
0.7 DATA I ~ 1
o.s - - pred~chon ~ []
[ ] [ ] data DATA 2
0.5
1 2 3 4
LOG (CYCLES)
FIG. 18--Residual strength predictions.~r prototype specimens.
Examples
Results of two sampl e cal cul at i ons from the l i fe predi ct i on model are shown in Fi gs. 18
and 19, where the resi dual st rengt h of a ki nd of composi t e prost hesi s (flat-plate desi gn)
under normal i zed l oad is pl ot t ed agai nst appl i ed cycl es. The i nput paramet ers for the cal -
cul at i ons are summar i zed in Table 1. The critical el ement , i dent i fi ed as a poi nt under hi ghest
l ocal compr essi ve stress in the fiber direction, is found to be l ocat ed on the medi al - neck
regi on, in the pl y wi t h the l owest angl e of i nci dence to the neck axis. The uni di rect i onal S-
N curve used is
o-,__1 = 1.0 - 0.12 l og N (13)
X c
The coefficient 0.12 is obt ai ned by adj ust i ng Curve 1 to Dat a 1 (Fig. 18). Damage due to
RE~ ;IDUAL STRENGTH
1.1
1
0.~ ~ E 4
0.8
0.7
0 . 6
o.s - - prediction DATA3 ~ATA4
O.4 ,~. Z~ data
0.3
1 2 3 4 5 6
LOG(CYCLES)
FIG. 19--Residual strength predictions f or prototype specimens.

T A B L E l - - I n p u t par ame t e r s f o r exampl es.
Curve 1 Curve 2 Curve 3 Curve 4
% Ul t i ma t e l o a d 84 63 52 47
Lo a d a n g l e " 0, l 0 0, 10 0, 10 0, 10
Number of plies 32 32 32 32
Stiffness degradation Eq 14 Eq 14 Eq 14 Eq 14
Strength degradation . . . . . . Eq 15 Eq 15
" In-plane, out-of-plane.
mechani cal l oadi ng is represent ed by l i near degradat i on functions in the pl y t ransverse and
shear modul us
where
G~ 2 = i ni t i al shear modul us and
G~2 = shear modul us at n.
: 1 - ( 1 4 a )
G~ = 1 - (14b)
For si mpl i ci t y, effect of appl i ed l oad level is not consi dered here. The ordi nat es of the
exper i ment al dat a poi nt s shown in Fi gs. 18 and 19 are taken to be the maxi mum strain rat i o
of the cri t i cal el ement at t hei r respect i ve appl i ed l oad level as det er mi ned by the stress-
anal ysi s model . It should be ment i oned that the linear degradat i on functions for the pl y
t ransverse and shear modul us were chosen for t hei r si mpl i ci t y. Here the residual strength is
cal cul at ed based on the maxi mum strain ratios, si nce the l ocal fai l ure function, FL, of the
critical el ement is represent ed by that ratio. Hence, failure of the prost hesi s occurs when the
(i nst ant aneous) maxi mum strain rat i o exceeds the resi dual strength of the critical el ement ,
expressed in t erms of such a ratio.
The second exampl e is to i l l ust rat e the fat i gue behavi or of the same ki nd of composi t e
prost hesi s in the presence of moi st ure. The cri t i cal el ement was found to be at the same
l ocat i on as in the last exampl e. In addi t i on to t he l i near stiffness degradat i on equation, a
strength degr adat i on equat i on is al so used
(15)
The results are shown in Fig. 19. Not e that Eq 15 is chosen so that Curve 3 can be adj ust ed
to Dat a 3.
Conclusions
A per f or mance si mul at i on model for desi gni ng and predi ct i ng the remai ni ng strength and
life of a composi t e hi p prost hesi s subj ect ed to cycl i c mechani cal and envi ronment al l oads is
descri bed. Labor at or y tests necessary to charact eri ze fatigue damage in the composi t e pros-
thesis are al so hi ghl i ght ed. It is real i zed that some of the anal yses present ed here are still in

i ncompl et e form. For f ut ur e devel opment , it is suggest ed t hat mor e refi ned t reat ment s, such
as mi cr omechani cs , be used to r epr esent pr oper t y and st r engt h change i n t he composi t e
syst em, Nevert hel ess, it is demonst r at ed t hat t he cri t i cal el ement concept is a r easonabl e
approach, bot h physi cal l y and mechani st i cal l y, for pr edi ct i ng t he l ong- t er m behavi or of en-
gi neer i ng st ruct ures made of composi t e mat er i al syst ems.
Acknowledgment
The aut hors gr at ef ul l y acknowl edge t he suppor t of t hi s wor k by Smi t h & Nephew Ri ch-
ards, Inc.
References
[1] Marvin H. Meyers, Ed., Fractures of the Hip, Year Book Medical Publishers, Inc., Chicago, 1985.
[2] Maharaj, G. R. and Jamison, R. D., "Creep Testing of a Composite Material Human Hip Pros-
thesis," Composite Materials for Implant Applications in the Human Body: Characterization and
Testing, ASTM STP 1178, Russell D. Jamison and Leslie N. Gilbertson, Eds., American Society
for Testing and Materials, Philadelphia, 1993, pp. 86-97.
[3] Olshansky, S. J., Carnes, B. A., and Cassel, C. K., "The Aging of the Human Species," Scientific
American, April 1993, pp. 46-52.
[4] Davidson, J. A., "The Challenge and Opportunity for Composites in Structural Orthopedic Appli-
cations," Journal of Composites Technology & Research, Vol. 9, No. 4, 1987, pp. 151-161.
[5] Reifsnider, K. L., Jamison, R. D., Gavens, A. J., and Maharaj, G. R., "Long Term Bahavior of
Biomedical Composites," Proceedings, American Society for Composites Fourth Technical Con-
ference, Technomic Publishing Co., Inc., Lancaster, PA, 1989, pp. 886-895.
[6] Tsai, S. W. and Hahn, H. T., Introduction to Composite Materials, Technomic Publishing Company,
Inc., Lancaster, PA, 1980.
[7] Gavens, A. J. and Locke, L. L., "Influence of the Abductor Muscle Force on Femoral Prosthesis
Strain," Proceedings, American Society of Biomechanics 13th Annual Meeting, Burlington, VT,
23-25 Aug. 1989.
[8] Callaghan, J. J., "The Clinical Results and Basic Science of Total Hip Arthroplasty with Porous-
Coated Prostheses," The Journal of Bone and Joint Surgery, Vol. 75-A, No. 2, Feb. 1993, pp. 299-
310.
[9] Thomson, W., "On the Dynamic Theory of Heat," Transactions of the Royal Society of Edinburgh,
Vol. 20, Royal Society of Edinburgh, Edinburgh, 1853, pp. 261-283.
[10] Bakis, C. E. and Reifsnider, K. L., "Nondestructive Evaluation of Fiber Composite Laminates by
Thermoelastic Emission," Review of Progress in Quantitative Nondestructive Evaluation, Vol. 7B,
D. O. Thompson and D. E. Chimenti, Eds., Plenum Publishing Corp., NY, 1988, pp. 1109-1116.
[11] Zhang, D., Enke, N. E, and Sandor, B. I., "Thermographic Stress Analysis of Composite Materials,"
Experimental Mechanics, Vol. 30, March 1990, pp. 68-73.
[12] Reifsnider, K. L., "Interpretation of Laboratory Test Information for Residual Strength and Life
Prediction of Composite Systems," Cyclic Deformation, Fracture, and Nondestructive Evaluation
of Advanced Materials, ASTM STP 1157, M. R. Mitchell and O. Buck, Eds., American Society for
Testing and Materials, Philadelphia, 1992, pp. 205-223.
[13] Reifsnider, K. L. and Stinchcomb, W. W., "A Critical-Element Model of the Residual Strength and
Life of Fatigue Loaded Composite Coupons," Composite Materials: Fatigue and Fracture, ASTM
STP 907, H. T. Hahn, Ed., American Society for Testing and Materials, Philadelphia, 1986, pp.
298-313.
[14] Liao, K. and Reifsnider, K. L., "Multi-axial Fatigue and Life Prediction of a Composite Hip
Prosthesis," Advances in Multiaxial Fatigue, ASTM STP 1191, D. L. McDowell and R. Ellis, Eds.,
American Society for Testing and Materials, Philadelphia, 1993, pp. 429-449.
[15] Liao, K. and Reifsnider, K. L., "Compression Response of a Fatigue-Loaded Composite Hip Pros-
thesis," Compression Response of Composite Structures, ASTM STP 1185, S. E. Groves and A. L.
Highsmith, Eds., American Society for Testing and Materials, Philadelphia, 1994.
[16] Stinchcomb, W. W. and Reifsnider, K. L., "Fatigue Damage Mechanisms in Composite Materials:
A Review," Fatigue Mechanisms, ASTM STP 675, J. T. Fong, Ed., American Society for Testing
and Materials, Philadelphia, 1979, pp. 762-787.

[17] Charewicz, A. and Daniel, I. M., "Damage Mechanisms and Accumulation in Graphite/Epoxy
Laminates," Composite Materials: Fatigue and Fracture, ASTM STP 907, H. T. Hahn, Ed., Amer-
ican Society for Testing and Materials, Philadelphia, 1986, pp. 274-297.
[18] Reifsnider, K. L., Schulte, K., and Duke, J. C., "Long-Term Fatigue Behavior of Composite Ma-
terials," Long-Term Behavior of Composites, ASTM STP 813, T. K. O' Brien, Ed., American Society
[19] Highsmith, A. L. and Reifsnider, K. L., "Internal Load Distribution Effects During Fatigue Loading
of Composite Laminates," Composite Materials: Fatigue and Fracture, ASTM STP 907, H. T. Hahn,
Ed., American Society for Testing and Materials, Philadelphia, 1986, pp. 233-251.
[20] Bakis, C. E. and Stinchcomb, W. W., "Response of Thick-Notched Laminates Subjected to Tension-
Compression Cyclic Loads," Composite Materials: Fatigue and Fracture, ASTM STP 907, H. T.
Hahn, Ed., American Society for Testing and Materials, Philadelphia, 1986, pp. 314-334.
[21] Yeow, Y. T., Morris, D. H., and Brinson, H. E, "Time-Temperature Behavior of a Unidirectional
Graphite/Epoxy Composite," Composite Materials: Testing and Design (Fifth Conference), ASTM
STP 674, S. W. Tsai, Ed., American Society for Testing and Materials, Philadelphia, 1979, pp.
263-281.
[22] Shen, C. H. and Springer, G. S., "Moisture Absorption and Desorption of Composite Materials,"
Journal of Composite Materials, Vol. 10, Jan. 1976, p. 2.
[23] Gillat, O. and Broutman, L. J., "Effect of an External Stress on Moisture Diffusion and Degradation
in a Graphite-Reinforced Epoxy Laminate," Advanced Composite Materials--Environmental Ef-
fects, ASTM STP 658, L R. Vinson, Ed., American Society for Testing and Materials, Philadelphia,
1978, pp. 61-83.
[24] Whitney, J. M. and Browning, C. E., "Some Anomalies Associated with Moisture Diffusion in
Epoxy Matrix Composite Materials," Advanced Composite Materials--Environmental Effects,
ASTM STP 658, J. R. Vinson, Ed., American Society for Testing and Materials, Philadelphia, 1978,
pp. 43-60.
[25] Chang, E, Shahid, I., and Engdahl, R. A., "Predicting Moduli and Strengths Reduction of Unidi-
rectional Graphite/Epoxy Composites Due to Hygrothermal Effects," Journal of Reinforced Plastics
and Composites, Vol. 8, March 1989, pp. 106-132.
[26] Chamis, C. C., Lark, R. F., and Sinclair, J. H., "Integrated Theory for Predicting the Hygrother-
momechanical Response of Advanced Composite Structural Components," Advanced Composite
Materials--Environmental Effects, ASTM STP 658, J. R. Vinson, Ed., American Society for Testing
[27] Loos, A. C. and Springer, G. S., "Moisture Absorption of Graphite-Epoxy Composites Immersed
in Liquids and in Humid Air," Journal of Composite Materials, Vol. 13, April 1979, p. 131.
[28] Farley, G. L and Herakovich, C. T., "Two-Dimensional Hygrothermal Diffusions Into a Finite
Width Composite Laminate," VPI-E-77-20, Virginia Technical Institute, College of Engineering,
Blacksburg, 1977.
[29] Shirrell, C. D., "Diffusion of Water Vapor in Graphite/Epoxy Composites," Advanced Composite
Materials--Environmental Effects, ASTM STP 658, J. R. Vinson, Ed., American Society for Testing
[30] Hofer, K. E., Skaper, G. N., Bennet, L. C., and Rao, N., "Effect of Moisture on Fatigue and Residual
Strength Losses for Various Composites," Journal of Reinforced Plastics and Composites, Vol. 6,
Jan. 1987, pp. 53-65.
[31] Sumsion, H. T. and Williams, D. P., "Effects of Environment on the Fatigue of Graphite-Epoxy
Composites," Fatigue of Composite Materials, ASTM STP 569, J. R. Hancock, Ed., American
Society for Testing and Materials, Philadelphia, 1975, pp. 226-247.

Sam Kanti mathi I and Tommy N. Whi te 2
Fatigue of a Landing Gear Act uat or Beam
in a Fighter Aircraft
REF ERENCE: Kantimathi, S. and White, T. N., " Fa t i g u e o f a La n d i n g Ge a r Ac t u a t o r
Be a m i n a Fi g h t e r Ai r c r af t , " Case Studies f or Fatigue Education, ASTM STP 1250, Ralph I.
Stephens, Ed., American Society for Testing and Materials, Philadelphia, 1994, pp. 53-66.
ABS TRACT: In 1988, inspectors discovered widespread cracking in the main landing gear
actuator support beam in a U.S. Navy fighter aircraft. Fractography confirmed metal fatigue as
the cause. The beam was made of D6AC steel, a high-strength material.
A flange with a small radius behind the attachment lug was the location of crack origination.
A second radius had introduced a multiplication of stress concentrations. Some fixes were
evaluated and ruled inviable due to structural constraints or economic considerations.
The solution was to "beef up" (reinforce) the backup structure with a 17-7 stainless-steel
plate. Interference fit fasteners were used to stiffen the surrounding structure. A finite element
analysis was performed to analyze the rework design of the support structure. This helped in
fracture mechanics calculations, which are included. Critical crack size and crack growth life
determinations were made. This paper describes the investigations, analyses, and development
of the solution to the cracking. Discussion of the causes of the cracking was given an emphasis.
Questions for the students have been added.
KE YWORDS : fatigue, crack initiation or nucleation, crack propagation, fracture mechanics,
repairs, stress concentration, aircraft
H i s t o r y
In 1988, an i ns pect or per f or mi ng a r out i ne vi sual i nspect i on of a U. S. Navy f i ght er j e t
di s cover ed a cr ack in t he shoul der o f t he mai n l andi ng gear nacel l e r i ng f r ame (Fi gs. 1 and
2) l ocat ed wi t hi n t he l ef t - hand sponson. The cr ack l ocat i on was in t he l ower fl ange be t we e n
t he mai n l andi ng gear t r unni on l ugs, adj acent to t he r et r act cyl i nder at t achment l ug (Fig. 3).
The beam is a f or gi ng whos e mat er i al is D6AC st eel wi t h c a dmi um pl at i ng. The mai n l andi ng
gear support beam in this mode l ai r cr af t is a cr uci al st ruct ural e l e me nt in that many fl i ght
and l andi ng cr i t i cal f unct i ons ar e as s i gned to it. It is al so an expens i ve i t em. Consequent l y,
exhaus t i ve i nvest i gat i ons wer e ma de to many fl eet ai r cr af t bel ongi ng to this model . Thes e
i nvest i gat i ons r eveal ed wi des pr ead cr acki ng in many fl eet ai r cr af t in t he s ame r egi on o f t he
st ruct ure (Fig. 4).
The magni t ude o f t he pr obl e m and its wi des pr ead nat ure, af f ect i ng combat readi nesss,
necessi t at ed i mme di a t e at t ent i on. The r emoval of t hi s pr i mar y st r uct ur e f r om t he ai r cr af t
woul d i nvol ve " f i xt ur i ng" t he ai r cr af t ( suppor t i ng it on st ands and l oc ki ng it in posi t i on) .
Af t e r this t ask is accompl i s hed, it woul d t ake sever al days to r e move t he pr i mar y st ruct ure
by dr i l l i ng out sever al hundr ed fast eners. Gi ven t he sense of ur gency di ct at ed by t he t hen
vol at i l e geopol i t i cal si t uat i on, t hi s appr oach was d e e me d t oo slow. Thi s f or ced t he devel -
President, Fatigue Concepts, E1 Dorado Hills, CA 95762-9734.
2 Senior structures engineer, U.S. Navy, Norfolk, VA 23611-5899.
53

c
t
~

m

o
o

-
-
4

c

c
~

m

o
o

-
1
"
1

0

c

m

m

c

0

z

F
I
G
.

1
-
-
F
i
g
h
t
e
r

a
i
r
c
r
a
f
t

m
a
i
n

l
a
n
d
i
n
g

g
e
a
r

a
c
t
u
a
t
o
r

s
u
p
p
o
r
t

s
t
r
u
c
t
u
r
e

N
a
c
e
l
l
e

r
i
n
g

f
r
a
m
e
.

C
o
p
y
r
i
g
h
t

b
y

A
S
T
M

I
n
t
'
l

(
a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
)
;

T
h
u

M
a
y

8

1
0
:
4
2
:
1
2

E
D
T

2
0
1
4
D
o
w
n
l
o
a
d
e
d
/
p
r
i
n
t
e
d

b
y
U
n
i
v
e
r
s
i
d
a
d

D
e

C
h
i
l
e

p
u
r
s
u
a
n
t

t
o

L
i
c
e
n
s
e

A
g
r
e
e
m
e
n
t
.

N
o

f
u
r
t
h
e
r

r
e
p
r
o
d
u
c
t
i
o
n
s

a
u
t
h
o
r
i
z
e
d
.

KANTIMATHI AND WHITE ON A LANDING GEAR ACTUATOR BEAM 55
I NTERSECTI NC RADII ~ ~
CRACK ORI CI R
FIG. 2--Multiple stress concentrations and crack origin (Detail A of Fig. 1).
opment of structural modifications and reinforcements to remove existing damage and lower
the high stresses without dismantling the primary structure. Several parts were sectioned and
subjected to fractographic examination. Later, fract ography of cracked parts pointed to metal
fatigue as the cause of failure (Fig. 5).
The interim repair involved shaving of f material at the flange radii to remove small cracks
that had initiated and grown in that region. Such a crack is highlighted by fluorescent dye
penetrants (Fig. 4). Additionally, overall magnetic particle inspection (Fig. 3) was carried
out to quantify crack lengths. This was followed by later partial disassembly and in-depth
inspection using a technique called "magnet i c rubber." This fix would have been sufficient
i f it was an isolated problem resulting from qual i ~ control at the manufact urer' s site, and
i f the cracking was restricted to onl y one airplane in the fleet. But, since the cracking was
discovered in numerous aircraft, the probl em pointed to one of a structural design in nature,
and hence, it was necessary to find a more elaborate "f i x" and apply it to all aircraft oper-
ational in the fleet, as well as to aircraft on the production line.
Small intersecting radii (Fig. 2) inadvertently introduced by weight reduction efforts in a
flange adjacent to the main landing gear actuator attachment lug were the location of crack
origination. Al t hough statically the structure was still adequate, these intersecting radii pro-
vided a multiplication of stress concentrations, produci ng the magnitude of stresses sufficient
for fatigue cracki ng to develop substantially sooner than the "design life" of the aircraft.
An instrumented flight test revealed that landing gear actuator loading was the only service
condition that subjected the structure to subsequent fatigue damage. Due to an inadequate
backup structure, these actuator loads were inducing beam bending into an adjacent flange
containing the intersecting radii. Finite element analysis (Fig. 6) was performed to analyze
the structure. The stresses from the analysis were confirmed by the instrumented flight test
results (Fig. 7), both of which substantiated prior analysis findings. These stresses were then
used to establish critical crack size and remaining life expectancy of in-service cracks and
the residual strength of the part in the presence of these cracks. These were useful pieces of
information in defining the subsequent usage of fleet aircraft.
Since flight loads, as determined by strain-gaged flight, did not play a significant role in
the cracking of the structure, further flight testing was deemed unnecessary. The prototype

FIG. 3--Photograph showing main landing gear trunnion lug and retract cylinder attachment lug.

FIG. 4--Typical crack locations.
design was thus installed and tested on an aircraft supported with jacks to permit gear
cycling. The appropriate gear actuator fatigue load spectrum was applied for an equivalent
of one service tour (approximately 88 of the service life, or approximately 2000 landing gear
cycles). Strain gages were again used to monitor the stress profile in the reworked radii.
Later fatigue analysis using the same stresses for crack initiation life and crack growth life
were performed using strain-life curves and linear-elastic fracture mechanics, respectively;
This indicated that adequate fatigue life and strength conditions were met for the aircraft's
intended usage.
The discovery of this cracking problem was quite timely since the critical flaw size for
the flange was just slightly larger than the size of most of the flaws detected in the fleet
aircraft. Although this critical flaw size was for the flange alone and not for the entire
structure, continued growth of the crack would have threatened the entire main landing gear
support structure. Early detection of the cracks in the flange enabled salvage of the structure
before irreversible damage occurred.
FIG. 5--Fractography of cracked part exhibiting features of metal fatigue.

FIG. 6--An example of finite element analysis.
Investigations and Discussion
Cracking was found to originate at a compounded stress-raiser (radius) in the integral
flange. It is well known that a structural discontinuity often called a stress concentration or
stress-raiser results in elevation of local stresses in the immediate vicinity of the discontinuity.
Multiple stress concentrations make the situation even worse. For example, for a central hole
in a wide plate loaded in uniform remote tensile stress, the stress concentration factor is
three at two diametrically opposite points at the edge of the hole along the minimum section.
If another hole is drilled with axis normal to that of the first hole and normal to the load
axis, then the extent of local maxi mum stresses at a point of intersection of the two holes
could be three times three or nine times the original stress! Such a design has occurred in
real life [1] where a small access hole for lubrication was thoughtlessly drilled almost per-
pendicular to the load axis, intersecting a pre-existing hole (Fig. 8).
Similarly, while a radius is unavoidable in stepped sections, existence of compound radii
spells doom for most structures, especially highly stressed aircraft structures where stresses
can be high in order to avoid weight penalties. That was precisely what happened in the
landing gear support structure beam under discussion (Fig. 2). These high stresses at the
notch can result in local plastic deformation t hat hastens the formation of cracks that can
then grow due to time-variant in-service loadings (fatigue) or stress corrosion. The flange
with multiple stress-raisers was also normal to the load path provided by the main landing
gear actuator attachment lug. Also, the backup structure was found to be weak with low
stiffness so the flange acted as a beam with fixed ends carrying a concentrated load from

GAGE TYPE:
A/C STRAI N EA- 06- 125BZ- 350W
/ OC; ATI ON. q r - l o EA REQ. ON STRUCTURE
GAGE . . . . . . . . . . / 30 EA REQ. ON COUPONS SF 782
/ / " - p F 781
SF 779" / I ISF 780
SF 775 ~ - -
!1 + / I
l i + 1 t s , , , , _
+ . . - . . - + /
F.S. 569
GAGE TYPE:
EA - 06 - 062AQ - 350ES
5 EA REQ. ON STRUCTURE
15 EA REQ. ON COUPONS
L/H TOP OF BULKHEAD
1474M LEG DOWN M.T.S.
LOCATIONS INDICATED ON
NC BY STRESS
FIG. 7- - St r ai n gage locations in i nst rument ed f l i ght test. FWD = f or war d o f the aircraft and
I NBD = inboard o f the aircraft.
the actuator. Moreover, the maxi mum bending moment happened to occur close to the lo-
cation of the flange with a compound radii. In addition, out-of-plane deflections due to
secondary bending aggravated flexural stresses, hastening crack initiation along with the other
factors outlined in this paragraph.
Metallurgical examinations ruled out material and processing defects to be contributors of
the cracking. Improper heat treating was eliminated as a causal factor. The percentage content
of each constitutent alloy was well within its specified range as verified by X-ray spectrog-
raphy. Tensile strength, modulus of elasticity, and surface hardness were tested to be normal.
The grain size and structure was also verified. Since the aircraft operated extensively in sea
atmosphere, environmentally assisted cracking becomes a major concern. Stress corrosion
cracking is a leading cause of structural failures in naval aircraft of this type [2]. Careful
examination of the fracture regions showed no evidence of stress corrosi on damage at this
location. In addition, other environmental degradation, such as hydrogen embrittlement, can
damage plated structural components. Electrolytic cadmi um plating makes the forging more
susceptible to hydrogen embrittlement. Vacuum deposited cadmi um plating is a preferred
process, due to the absence of hydrogen-i nduci ng influences [3]. Examination of this forging
revealed that it was indeed plated with cadmi um using the vacuum-deposited process. Sub-
sequently, hydrogen embrittlement from the plating process was ruled out to be a cause for
the initiation of the cracks.
Examination of the fracture surface also revealed beach marks indicative of fatigue crack
growth over a period of time. Micro-fissures were observed on closer examination. Striations

TIPS O N FATIGUE
5.52 INTERSECTING
HOLES--Here is a c a s e
w h e r e a h o l e was t a p p e d
i n t o t h e h i g t d y s t r e s s e d
region of another h o l e
in the t e n s i o n f l a n g e of
a w i n g beam. T h i s is
s i m i l a r to the case of a
g r e a s e f i t t i n g hole for
a b e a r i n g .
If y o u m u s t h a v e a
hole, the idea is to move
it t o t h e l e a s t d a m a g i n g
p o s i t i o n . A l i t t l e local
r e i n f o r c i n g ( b e e f up) is
a l s o h e l p f u l .
P O O R
b ~ e f u p
A L S O A C C E P T A B L E
FIG. 8- - The dangers of intersecting holes. (This rough sketch is a true reproduction from Ref 1.)
wer e observed, but not in such a way that quant i t at i ve i nf or mat i on coul d be gl eaned from
t hem. Typi cal l y, in st eel surfaces, it is mor e di ffi cul t to fi nd quant i t at i vel y useful st ri at i ons
[4]. Thi s is in cont rast wi t h t he mor e preval ent st ruct ural mat er i al s in aircraft, namel y, alu-
mi num al l oys, whi ch usual l y exhi bi t ni ce striations. The mai n reason is that steels l ack the
prot ect i ve nature of the oxi de fi l m that forms qui ckl y in t he fract ure surface in al umi num
al l oys. It is vi rt ual l y i mpossi bl e to i dent i fy cl ear l y defi ned areas of striations in steels~ t hereby
maki ng fract ographi c exami nat i on most di ffi cul t [4]. The si t uat i on is even more difficult for
hi gh-st rengt h steels, and t he beam was made of a hi gh-st rengt h st eel (Rc 46-49).
The final sol ut i on became a combi nat i on of several modi fi cat i ons i nt ended to i mpr ove the
fat i gue l i fe of t he cracki ng structure. These i ncl uded:
(1) reduct i ons in stress concent rat i ons t hrough modi fi ed geomet ry,

(2) r ei nf or cement of adequat e backup structure, and
(3) use of i nt erference fit fast eners to i ncrease j oi nt stiffness.
Analysis
Fr act ogr aphy of t he avai l abl e cr ack surfaces reveal ed t ypi cal beach mar ks found in fat i gue
fracture surfaces (Fi g. 5). The ori gi n for this fracture was l ocat ed near t wo i nt ersect i ng radii,
ori gi nal l y descr i bed usi ng Fig. 2. The initial slow crack growt h r egi on near the ori gi n was
appr oxi mat el y 1. 5-mm in length. Under mi cr oscopi c exami nat i on, this region reveal ed
mi cro-fi ssure format i ons of fatigue cr ack growth. The adj acent sect i on of fracture surface
was an area about 25- mm l ong With ei ght or ni ne t humbnai l - shaped areas exhi bi t i ng over-
stress di mpl i ng and shear lips, wi t h subsequent crack arrest (Fig. 5). The crack arrest areas
al so had mi cro-fi ssure format i ons t ypi cal of fatigue. Each t humbnai l - shaped area, whi ch was
appr oxi mat el y 2. 5-ram l ong, char act er i zed a crack "j ump, " due i ni t i al l y to the overstress,
wi t h crack arrest occurri ng upon di ssi pat i on of the energy appl i ed. Fast crack growt h
(overst ress) was obser ved beyond this region.
Initially, t he frame flange cont ai ni ng the i nt ersect i ng radi i was model ed as a beam wi t h
fi xed ends, i ncl udi ng an el ast i c spri ng aft of the act uat or l ug represent i ng t he backup structure
and a concent rat ed l oad appl i ed at the act uat or lug. The maxi mum bendi ng stress was deri ved
from beam anal ysi s to be 1079 MPa. Thi s val ue compar ed favorabl y wi t h the stresses mea-
sured during i nst rument ed flight test in l ocat i ons out l i ned in Fi g 7. Then a Neuber correct ed
notch factor, K = 2.35, was obt ai ned from the t heoret i cal stress concent rat i on factor, K =
2.4, by usi ng the rel at i on
K, - 1
K, , = 1 + 1 + P / r (1)
f r om Pet er son' s S t r e s s C o n c e n t r a t i o n F a c t o r s [5]. Thi s compounded wi t h the maxi mum
bendi ng stress pr oduced a stress val ue of 2535 MPa at t he i nt ersect i ng radii.
Subsequent l y, an el ement ar y finite el ement l oad anal ysi s was empl oyed to study the l oads
in the regi on of failure. A t wo- di mensi onal coarse gri d mesh was pr epar ed of this " be a m"
structure. Load val ues resul t i ng from this anal ysi s were i ndi cat i ve of an ineffective backup
structure. The backup structure l acked sufficient stiffness to carry the act uat or l oads away
from the frame flange, whi ch in turn was left to deflect t he l oad i nt o adj acent , integral parts
of the frame.
Later, a mor e devel oped finite el ement stress anal ysi s was empl oyed t o better underst and
the stresses in the cri t i cal structural l ocat i on. A t wo- di mensi onal fi ne-gri d mesh was pr epar ed
wi t h a part i cul arl y cl ose mesh in and around the site of crack i ni t i at i on (the i nt ersect i ng
radii). This pr ovi ded stress values at the i nt ersect i ng radi i that were compar abl e to t hose
measured by i nst rument ed flight test and convent i onal not ch stress cal cul at i ons from Neuber
correct ed t heoret i cal stress concent rat i on factors for appl i ed bendi ng.
Si nce the not ch-correct ed stress cal cul at i on exceeded the yi el d st rengt h of the mat eri al ,
and is not represent at i ve of the true stress state, a not ch strain anal ysi s was performed. In
l i eu of notch stress, t he notch strain, e, is used as a funct i on of nomi nal strain, e, namel y
9 = Ke . This notch strain then rel at es to the correct not ch stress usi ng the cycl i c stress strain
curve for the mat eri al . First, a not ch strain was cal cul at ed by gr aphi cal l y sol vi ng the cycl i c
stress-strain curves si mul t aneousl y wi t h Neuber ' s rul e [6]. Neuber ' s rul e appl i es to thin struc-
tures and was used si nce the part bei ng anal yzed is thin, appr oxi mat el y 6 mm, and t he l i near

rule that appl i es to t hi cker structures is t oo l i beral in this case. The cycl i c stress-strain curve
is gi ven by the rel at i on
where ~E e = ~ and
( ~ ~ l / n '
The Neuber ' s rul e is gi ven by t he rel at i on E~r = K2eS. Stress val ues cal cul at ed compar e
cl osel y to t hose obser ved earl i er f r om i nst rument ed flight tests as wel l as t hose deri ved from
finite el ement consi derat i ons. An i ni t i at i on fatigue life, appr oxi mat el y 2000 cycl es, was then
obt ai ned f r om cont ract or st ress-l i fe curves speci fi cal l y generat ed for this material. It was
i nt erest i ng to note that the cr ack i ni t i at i on l i fe cor r esponded roughl y to the number of in-
servi ce l andi ng gear cycl es, i ncl udi ng flight and mai nt enance cycl es, accumul at ed for t hose
ai rcraft that exhi bi t ed the smal l er cracks.
A resi dual st rengt h di agr am and a cri t i cal cr ack size, a, were t hen devel oped from l i near
el ast i c fracture mechani cs in the f ol l owi ng manner. Fi r st a reference stress, or, was chosen
to be the flange bendi ng stress pr evi ousl y arri ved at by convent i onal anal ysi s for the affect ed
geomet ry. As ment i oned earlier, an i nst rument ed flight t est had r eveal ed that l andi ng gear
act uat or l oadi ng was the onl y servi ce condi t i on that subj ect ed the structure to subsequent
fat i gue damage. Therefore, a si mpl e const ant ampl i t ude l oadi ng spect r um was adequat e in
this case, wi t h the stress fl uct uat i ng bet ween a maxi mum of 1079 MPa and a mi ni mum of
0 MPa (R = 0). Next , the appr opr i at e stress i nt ensi t y factor, [3, of appr oxi mat el y 0.9 for this
geomet r y and l oadi ng condi t i on was obt ai ned from a compendi um of such factors [7]. Now,
the mat er i al ' s pl ane-st rai n fracture t oughness was t aken to be 105 MPaX/-m [8]. By usi ng
the above i nformat i on and sol vi ng for stress levels for i ncr ement al l y i ncreasi ng crack sizes,
starting at an i ni t i al fl awsi ze a, of 0.125 mm, a resi dual strength di agr am (Fig. 9) was
generat ed f r om [cr c = ( K l / ~ - - - " t r d " t r a ) ] by pl ot t i ng ~r c versus a. Fr om t he resi dual strength di a-
gram, the cri t i cal cr ack si ze can be det ermi ned by finding the appl i ed reference stress on t he
ordi nat e and then l ocat i ng the cri t i cal flaw size, whi ch was found to be 3.7 mm on t he
abscissa.
Next, the Pari s crack growt h rate equat i on
da
- - = C ( ( 3 )
dN
where A stands for an i ncrement , was i nt egrat ed to pr oduce a cr ack growt h di agr am (Fig.
10) and to cal cul at e the crack growt h l i fe to failure. Usi ng the rel at i onshi p [5,6] for AK =
Ac r [ 3 ~ a we get
da C(
dn
in the above expressi on for t he Pari s equation, we find that for D6AC steel, C = 1.75 E-07,
and m = 2.5 [6]. However, the const ant C appl i es onl y to Engl i sh units specifically, when
the crack growt h rate is measured in inches per cycl e and AK is in ksiX/~m. An overwhel m-

(Y
GPa
4
3
2
i
I
I
I
1.0 2.0 3.0 4.0
a, mm
FIG. 9--Residual strength diagram,
cr ack l engt h (ram)
5
q - i . . . . . . . . . . . . . . . . . 9 . . . . . . . . . . . . . . . . . . . . . . . . . .
2
I . , .
J
J
0 P P I r
5 IFO15 ~0 2'5 30 315 410 415 50
L i f e ( c y c l e s )
FIG. lO--Crack growth diagram f or constant amplitude loading.

ing majority of the crack growth data available in the literature is still in English units. The
useful conversion factor given below [6] is left t o the student as an exercise.
C [ m / c y c ; MPaX/-mm] =
C[ i n. / cyc; ksi X/~ln.]
39.37 (1.1) m
(6)
to be used in the relation
d a [m/ cyc] = C (AK[MPa ~ / / m m ] ) m
d N
Using the conversion factor (m remains the same), in metric units, the Paris parameters are:
C = 3.51 E-09, and m = 2.5. Usi ng the crack growt h rates for a water environment is
appropriate, since the structure is periodically exposed to moisture intrusion from field
environments.
Integrating this, and assuming that ~r, [3, C, and m are all constants with respect to crack
size, we find that [5,6]
where
N = C~r,,,f3,,,~m/2 L 1 - m / 2 J (6)
ai = initial crack length and
a s = final crack length.
Now plotting a; - a s versus N produces a crack growth diagram for the constant amplitude
loading (Fig. 10). The crack growth life to failure can now be determined by finding the
critical crack size on the ordinate, then finding the corresponding crack growth life, which
was found to be fifty landing cycles on the abscissa. This very low number of cycles for
crack growth was also apparent in the fracture surfaces exhibited using metallurgical ex-
aminations (Fig. 5).
Conc l us i ons
Numerous examples of fatigue failures brought about by carelessly or inadvertently in-
duced high-stress concentrations are given in Ref 8. The author of that reference, Dr. Fred-
erich Karl Naumann, remarks in Cfipater 3 entitled, "Failures Caused by Planning Er r or s' "
It has been known for a long time that during construction of machine parts and tools,
sharp edges and cross-sectional changes should be avoided because stress concentrations
form in such places. These can result in stresses substantially above the nominal stress.
It is surprising how often this well-known design axi om is violated even today.
Let us examine some of the possible reasons for such a violation of a fundamental prin-
ciple. One reason could be a lack of communi cat i on among various engineers and other
professionals involved ' in the design and processing stages of the landing gear actuator sup-
port beam. It appears that a good design had fallen victim to the well-intentioned trimming
at the hands of a weight engineer. A double radius resulted, and while the static strength of
Copyri ght by ASTM Int ' l (al l ri ght s reserved); Thu May 8 10: 42: 12 EDT 2014
Downl oaded/ pri nt ed by
Uni versi dad De Chi l e pursuant t o Li cense Agreement . No furt her reproduct i ons aut hori zed.

the part remained acceptable, the fatigue performance clearly suffered. Another example of
an inadvertent introduction of a stress concentration came about in another fighter aircraft
due to accumulation of positive tolerances in one direction that left a tighter radius at one
end. Finally, in another manufacturing situation, a generous fillet radius stipulated by the
design engineer was destroyed by a machinist who decided to reduce the number of passes
using the numerical control machine, thereby saving the manufacturer some money! Though
he achieved his goal in the short run, the part did fail in fatigue later due to the higher stress
concentration he introduced.
In the current case, failure to follow through in original design of the actuator support
backup structure had resulted in an unintended and unwanted load path. Equally important,
a double-stress concentration was introduced in the actuator support beam during structural
optimization efforts for the least weight design. This necessitated an expensive and time-
consuming fix in hundreds of cracked fleet aircraft. However, this fix has proved successful
and has developed into a permanent design change. No additional cracking at the actuator
beam has been reported to date.
St udent Quest i ons
1. Why were striations not quantitatively useful in the failure analysis in this case?
2. If the backup structure resulted in reducing the stress level by 25% in the actuator
beam, from 1079 to 810 MPa, calculate the increase in its crack growth life.
3. If the backup structure resulted in halving the stress level in the actuator beam, from
1079 to 540 MPa, calculate the increase in its crack growth life.
4. Can you infer a rough idea about the relationship between the stress level in a part and
its crack growth life, using the result of the calculation done in the text of this case
study, as well as your answers to questions 2 and 3?
5. You have found a material with the same fracture toughness, but roughly half the Paris
constant C. Estimate the improvement in crack growth life, if you use this new material.
6. Your partner in this project suggests that choosing a material with twice the fracture
toughness and roughly the same C and m will result in a much better life for the part
than switching to a material where C is cut in half. Do you agree? Prove it!
7. What steps would you undertake to ensure that the doubled stress concentration does
not occur during weight optimization or other steps?
8. If the aircraft is continuously operated in a non-water environment, by how many cycles
will the crack growth life improve? Use C = 9.1 E-10, and m = 2.5. Repeat for 3.5%
sodium chloride environment (use C = 7.9, E-09, m = 2.5). Comment on the results.
References
[1] Smith, C. R., "Tips on Fatigue," NAYWEPS 00-25-559, U.S. Government Printing Office, Wash-
ington, D.C., 1966.
[2] White, T. N. and Kantimathi, S., "Stress Corrosion Cracking of a High-Strength Steel in a Landing
Gear Support Frame in a Fighter Aircraft," in Handbook of Case Histories in Failure Analysis, ASM
International, Materials Park, OH, 1992.
[3] Kamdar, M. H., "Embrittlement by Solid-Metal Environments," in Failure Analysis and Prevention,
Metals Handbook, Ninth Edition, Vol. 11, American Society for Metals, Materials Park, OH, 1986.
[4] Hertzberg, R. W., Deformation and Fracture Mechanics of Engineering Materials, Third Edition,
John Wiley & Sons, New York, 1989.
[5] Peterson, R. E., in Stress Concentration Factors, John Wiley & Sons, New York, 1974.
[6] Fuchs, H. O. and Stephens, R. I., in "Metal Fatigue in Engineering," John Wiley and Sons, New
York, 1980.

[7] Kantimathi, S., in Fracture Mechanics--Aerospace Applications, American Institute of Aeronautics
and Astronautics, Washington, D.C., 1995.
[8] Damage Tolerant Design Handbook, A Compilation of Fracture and Crack Growth Data for High-
Strength Alloys, MCIC-HB-01, Metals and Ceramics Information Center, Battelle, Columbus, OH.
[9] Naumann, E K., Failure Analysis Case Histories and Methodology Dr. Riederer-Verlag GmbH,
Stuttgart, Germany, English Translation by American Society for Metals (ASM), 1983.

H. R. J hansal e ~
Fatigue Evaluation of Agitator Paddle
Shafts
REFERENCE: Jhansale, H. R., "Fatigue Evaluation of Agitator Paddle Shafts," Case
Studies f or Fatigue Education, ASTM STP 1250, Ralph I. Stephens, Ed., American Society for
ABSTRACT: Existing agitator paddles in material feeders were failing in fatigue. An exper-
imental evaluation was undertaken to assess the fatigue strengths of the existing agitator paddles
and the newly designed replacement paddles. As in most typical industrial situations, the study
was constrained by budget and time limitations. After considering several options, it was
decided to strain gage and test the paddles in their own feeder frame and drive motor with
minimum mock-up fixturing to achieve several load levels. While using real-time test speeds,
the test durations were accelerated by using higher than normal operating loads. Results were
analyzed using strain-based cumulative fatigue damage-analysis techniques and component
fatigue-life curves were developed to project the long life fatigue strengths. The investigation
showed that while the newly designed replacement paddles would have significantly improved
fatigue strengths, the actual improvements may be limited by other built-in weld discontinuities
in the paddle stiffeners.
KEYWORDS: steel, torsional fatigue, cumulative damage analysis, fatigue test, rainflow his-
togram data analysis
Nomencl at ure
Ae/ 2
A~r/2
As~2
K'
rl r
lyf
lY~f
b
C
E
D
n
Us
Local st rai n ampl i t ude
Local stress ampl i t ude
Nomi na l stress ampl i t ude
Fat i gue not ch fact or
Cycl i c st rengt h coeffi ci ent
Cycl i c st r ai n har deni ng exponent
Tr ue t ensi l e fact or st rengt h
Fat i gue st r engt h coeffi ci ent
Fat i gue duct i l i t y coeffi ci ent
Fat i gue st r engt h exponent
Fat i gue duct i l i t y exponent
El ast i c modul us
Damage
Appl i ed cycl es
Cycl es to fai l ure
Principal engineer, Fatigue and Fracture Mechanics, Siemens Power Corporation, 1040 S. 70th Street,
Milwaukee, WI 53214.
67
Copyright 9 1994 by ASTM lntcrnational www.astm.org

The Problem
Figure 1 shows a view of a material feeder with two agitator paddles as part of the test
setup used in the study. The flexible hopper bag has been removed and replaced by a central
frame in which are mounted shoe pads to provide the load resistance to the paddles. In
service, these paddles oscillate and agitate the hopper bag from outside to keep the material
flowing freely down the hopper. These material feeders are made in several sizes and are
used in various industries including chemical, plastics, pharmaceutical, and food processing.
The materials are usually in dry powder or granular form and require constant agitation to
keep them flowing.
Each of the two existing paddles included a 1-in. diameter shaft made from a 1018 cold-
rolled steel with a reduced diameter splined section at one end to transmit the torque from
a driving motor arrangement. The shaft was welded to the backside of the paddle, along
with stiffener webs by using circumferential and longitudinal fillet and spot welds (Fig. 2).
These shafts were prematurely failing in the field at the splined section due to torsional
fatigue. Information was not available regarding either the load levels or life durations.
The manufacturer had developed a newly designed paddle with a 1.25-in. diameter shaft
that used friction coupling instead of a reduced spline section to replace the failed paddles.
Basically, the manufacturer wanted to evaluate the long-life fatigue strengths of both the
existing (old) and replacement (new) paddles and determine the degree of improvement in
fatigue strength, if any. As in most typical industrial situations, the scope of the investigation
was constrained by budget and time limitations.
FIG. 1--Test setup showing material f eeder box with a central frame and adjustable shoe pads f or
resisting the two paddles. Approximate scale - f/8.

JHANSALE ON AGITATOR PADDLE SHAFTS 69
FIG. 2--Back side of a failed paddle showing welds and failure through spline section. Approximate
scale = %.
The Approach
Student Question
9 What type of study should be considered?
Answer
9 Since there were several unknown factors such as material properties, stress concentra-
tion factor, notch sensitivity, and so forth, it was decided that a testing program was
needed in addition to analytical estimations to develop the required answers. This paper
describes the test method used and the subsequent analyses required to interpret the
less-than-adequate results.
Student Question
9 What testing and analytical approaches should be used?
Answer
9 The plan was to perform constant amplitude fully reversed torsional cycling tests at
several load levels on both the old and new design paddle shafts. The load levels were
to be selected to obtain finite lives in the range of about 10 3 t o 10 6 cycles. Using
estimated strain-life and cyclic stress-strain properties of the material, a Neuber rule-
based notch-fatigue analysis [1,2] was performed. By matching the results of this anal-
ysis with the test data, appropriate "component factors," which are essentially "fatigue
notch factors," would be determined. By extrapolating this fatigue curve, the expected
fatigue strengths, corresponding to the desired service-life durations, would be
determined.

The Test Me t hod
The first choi ce was to use the exi st i ng axi al cl osed- l oop test syst em to per f or m the tests.
However, the cost and t i me requi red for maki ng appr opr i at e fixturing to devel op t orsi onal
l oadi ng from l i near act uat or mot i on were not accept abl e. Therefore, it was deci ded to use
the exi st i ng f eeder arrangement as it was, and measure the torques and bendi ng moment s
by way of strain gages.
Each paddl e was i nst rument ed wi t h t wo- ar m rosette strain gages to measur e torque. These
gages were mount ed sl i ght l y i nward from the spline sect i on on the 1-in. di amet er section.
Li near strain gages were mount ed near the wel d to measure bendi ng strain. The actual
l ocat i ons of t hese strain gages were sel ect ed based on the results of a separat e st ress-anal ysi s
test using brittle coat i ng. Scal e factors for convert i ng the st rai n-gage out put to appropri at e
nomi nal stresses and torque were der i ved using bot h anal yt i cal cal cul at i ons and static dead-
wei ght cal i brat i ons.
Inst ead of usi ng the hopper and t he mat eri al for l oad resi st ance, a cent ral frame was
fabri cat ed wi t h adj ust abl e shoe pads to al l ow for the change in l oad levels (Fi g. l ). Port abl e
"r ai n fl ow" hi st ogr am recorders were used to col l ect the l oad hi st ory data, whi ch i ncl uded
strain range, mean strain, and cycl e counts. Test durat i ons in mi nut es were al so moni t ored
and recorded. Under - peak det ect i on and t ri ppi ng swi t ches were pr ovi ded to sense cracki ng
and st op the tests.
Test Experi ence
Each test consi st ed of one pai r of paddl es, desi gnat ed as left and ri ght paddl es. A total of
four pai rs of ol d paddl es and one pai r of new paddl es were tested. The tests were per f or med
at about 100 cycl es per rain, whi ch was within the range of operat i ng frequency in the field
(60 to 100 cpm).
Typi cal l y, each test was set up to appl y constant ful l y reversed t orque on the shaft. In
reality, due to l ack of preci se cl osed- l oop cont rol and syst em dynami cs, a spect rum of am-
pl i t udes were encount er ed in each test, as measured by the hi st ogram recorders.
In the case of the first pai r of ol d paddl es and the onl y pai r of new paddl es tested, the
i ni t i al l y sel ect ed l oad level was not sufficient to pr oduce failure in a reasonabl e number of
cycl es. Load levels were i ncreased several t i mes to i nduce failure. Whi l e cracki ng was even-
t ual l y achi eved in the first set of ol d paddl es, the new paddl es tested di d not experi ence any
cracki ng in the shaft or wel d l ocat i ons, even after several i ncreases in l oad level. However,
t hese tests had to be i nt errupt ed to r epai r wel d cracks that f or med in the paddl e bl ade edges
and out er frame. Si nce the new paddl e shafts were st ronger than the bl ade and the frame,
and since the al l ot t ed t i me and budget was exhaust ed, the test pr ogr am was t ermi nat ed at
this point.
Torsi onal Fat i gue Resul ts Anal ysi s
Student Question
9 How shoul d the mul t i pl e ampl i t ude test dat a and t he new paddl es data, whi ch di d not
fail, be anal yzed and i nt erpret ed to arrive at the requi red si mpl e answers?

J HANSAL E ON AGI TATOR PADDLE SHAFTS
TABLE 1 - - T o r s i o n a l f a t i g u e t e s t resul t s.
Specimen Equivalent Tor que Equivalent
Number Amplitude, N-m Cycles Failed
OLT 1 289.5 47 039 Yes
ORT1 289.5 47 130 Yes
OLT2 315.7 23 395 Yes
ORT2 315.7 28 187 Yes
OLT3 252.6 110 186 Yes
ORT3 252.6 155 286 Yes
OLT4 315.7 26 840 Yes
ORT4 336.8 18 215 Yes
NLT 1 369.9 422 027 No
NRT 1 369.9 338 693 No
NOTE--All failures represent cracks initiated in the spline section.
71
A n s w e r
9 The multiple amplitude data required a cumulative fatigue-damage analysis procedure
and a conversion to an equivalent constant-load amplitude for interpretation. The un-
failed new paddle data were used as conservative failure data. The details of the pro-
cedure used are described in Appendix A and summarized in the following section.
The equivalent torque versus the equivalent cycles data derived in Table 1 are plotted as
torque amplitude versus cycles to cracking for both old and new paddle shafts (Fig. 3).
Estimated torque amplitude versus cycles-to-failure curves, derived from the cumulative dam-
age analysis (Appendix A) are also plotted. The component calibration factors (or fatigue
notch factors) for each paddle, which was required for deriving the component fatigue curve,
was developed by trial and error, as described in Appendix A. These values ranged from
1.43 to 1.49 for the old paddles. They reflect not only the effective stress concentration due
to spline geometry, but also errors in the assumed material properties and specimen-to-
specimen variations. An average value of 1.46 was used for predicting the component fatigue
2;
w-
12b
12_
<z:
(:9
o
4 0 0
%
c~
300 ' b\
\
200
1 DO D
o
N E W P A D D L E T E S T S L
O L D P A D D L E T E S T S
J
_ _ _ _ J
0 4 , 2 . . . . . . . i - . . . . . . . . . . i , , , . . . . , . . . . . . . . , , ,
1 1 0 5 2 1 0 6 2 1 0 7 2 1 0 8 2 1 0 9
C Y C L E S T O C R A C K I N G
FIG. 3 - - T o r s i o n a l f a t i g u e l i f e c u r v e s f o r o l d a n d n e w p a d d l e s .

l i fe curve. In the case of t he new paddl es, t he requi red fact or was 2.76 to 2.81. Al t hough
the stress concent rat i on factor is vi rt ual l y el i mi nat ed in the new paddl e shafts, the cal cul at ed
hi gh fact or is due to the fact t hat t hese paddl es di d not fail but were assumed to be fai l ures
for conservat i ve anal ysi s.
In t he field, the paddl es are report ed to experi ence about 3 9 107 to 6 x 107 cycl es per
year. Therefore, t he t orsi onal fat i gue strength of t he ol d paddl es, cor r espondi ng to about 10
or mor e year s of servi ce (about 109 cycl es), is est i mat ed from t he pr oj ect ed curve to be
about 113 N.m. In the case of the new paddl es, the dat a poi nt s do not represent failure.
However, t he pr oj ect ed curve ext r apol at ed f r om t he dat a poi nt s is assumed to conservat i vel y
represent the t orsi onal fatigue strengths of the new paddl es. Fr om this curve, the est i mat ed
t orsi onal fat i gue strength of t he new paddl e shafts is about 186 N.m.
Fatigue Analysis of Welded Locations
Al t hough the manufact urer di d not speci fi cal l y express a concer n for the wel d l ocat i ons,
it was t hought appropri at e to eval uat e the on- goi ng fatigue damage at t hese l ocat i ons.
Student Question
9 What procedure shoul d be used to eval uat e t he fatigue damage at the wel d l ocat i ons?
Answer
9 Raw hi st ogr am dat a from bendi ng strain gages were anal yzed, si mi l ar to the t orsi onal
data, to cal cul at e equi val ent l oad l evel s for each of the paddl es. The cumul at i ve damage
anal ysi s was per f or med usi ng the fat i gue curves and pr ocedur e as per Bri t i sh St andard
BS5400 (Steel, Concret e and Composi t e Bri dges, Part 10, Code of Pract i ce for Fat i gue).
Thi s st andard, al t hough somewhat conservat i ve, is bel i eved to be the most compr ehen-
sive and up-t o-dat e pr ocedur e for wel d j oi nt s. A br i ef out l i ne of the pr ocedur e is in-
cl uded in Appendi x B.
Tabl e 2 summari zes the bendi ng and t orsi on l evel s exper i enced by each paddl e in the
vi ci ni t y of wel ds. The l ast col umn shows the cal cul at ed maxi mum pri nci pal stress, i deal l y,
the l oadi ng was supposed to be proport i onal ; however, the i ndependent l y col l ect ed dat a from
t he several hi st ograms showed some di scr epancy in the number of cycl e counts recorded
TABLE 2--Summary of weld location history.
Specimen Torsional Bending
Number Stress, MPa Torsional Cycles Stress, MPa Bending Cycles
OLT2 98.1 23 395 41.4 29 372
ORT2 98.1 28 187 41.4 23 818
OLT3 78.5 110 186 41.4 134 300
ORT3 78.5 155 286 35.5 252 700
OLT4 98.1 26 840 41.4 26 946
ORT4 104.7 18 215 35.5 33 223
NLT1 58.9 422 027 53.3 857 600
NRT1 58.9 338 697 82.9 317 700

JHANSALE ON AGITATOR PADDLE SHAFTS 73
bet ween bendi ng and t orsi onal l oadi ngs. Therefore, for t he pur pose of the anal ysi s, the l oad
hi st ory was assumed to consi st of a combi ned t orsi onal and bendi ng hi st ory and a pure
bendi ng hi st ory for cases where the bendi ng cycl es wer e hi gher in number compar ed to the
t orsi on cycl es (Tables 2 and 3).
None of the wel d l ocat i ons on any of the paddl es showed any cracki ng as tested by dye
penet rant and magnet i c part i cl e exami nat i ons. However, as shown in Col umn 6 of Tabl e 3,
t he cumul at i ve damage anal ysi s showed accumul at i on of si gni fi cant damage in the ol d pad-
di es. The t hi rd set of ol d paddl es (OLT3 and ORT3) and t he onl y set of new paddl es (NLT1
and NRT1) were pr edi ct ed to have cr acked (predi ct ed bl ocks is l ess than 1.0). Si nce none
of the paddl es showed any cracki ng, t he fatigue anal ysi s per f or med is consi der ed t o be t oo
conservat i ve.
Therefore, it appears that al t hough the t orsi onal fat i gue strength of the new paddl es has
been i ncreased, as di scussed earlier, t he fatigue st rengt hs in the wel d and ot her l ocat i ons
such as bl ade edges and out er frame may be the l i mi t i ng factors in the fat i gue per f or mance
of the mat eri al f eeder as a whol e. Si nce the degree of conser vat i sm in t he anal ysi s is un-
known, the onl y way to assess this si t uat i on is to gai n some field exper i ence wi t h the new
paddl es. I f wel d l ocat i on cracki ng is exper i enced in t he new paddl es, wi t hi n the desi gn
l i fet i mes, then fat i gue strength i mpr ovement s, such as post wel d gri ndi ng and shot peeni ng,
or both, can be consi dered.
Conclusions
It was concl uded that the l ong l i fe fat i gue strength of the ol d paddl e shafts was about 113
N. m of al t ernat i ng t orsi on wi t h the spl i ne section bei ng the cri t i cal l ocat i on. The l ong l i fe
t orsi onal fatigue st rengt h of the new paddl e shafts was est i mat ed t o be at l east 186 N.m.
However, the wel d regi ons may be rel at i vel y weaker in the new paddl es. Si nce the degr ee
of conservat i sm in t he wel d fat i gue anal ysi s pr ocedur e is not known, it was suggest ed that
t he new paddl es be fi el d tested, bef or e any fatigue i mpr ovement t echni ques are consi dered
for wel d regions.
This case st udy i l l ust rat es t he f ol l owi ng from an educat i onal poi nt of view. As found in
this study, most real -l i fe i ndust ri al pr obl ems requi re "s ol ut i ons " wi t h l ess than i deal or no
i nput i nformat i on avai l abl e, in addi t i on to const rai ned budget and t i me l i mi t s. Experi ment al
i nvest i gat i ons may or may not yi el d i deal l y expect ed results. Thi s forces the engi neeri ng
anal yst to make "r eas onabl e" assumpt i ons, pri ori t i ze needs, i mpr ovi se techniques, and cre-
at i vel y sol ve the pr obl em.
TABLE 3--Summary of weld location fatigue analysis.
Specimen Stress, MPa, Cycles, Predicted Blocks
Number Combined/Bending Combined/Bending to Failure ~ Cracked
OLT2 121/41 23 395/5977 1.28-3.65 No
ORT2 121 /... 28 187/... 1.08-2.98 No
OLT3 102/41 110 186/24 000 0.37-1.06 No
ORT3 96/36 155 286/97 400 0.45-1.29 No
OLT4 121/41 26 840/106 1.1-3.2 No
ORT4 124/36 18 215/15 008 1.5-4.3 No
NLTI 91/53 422 027/415 500 0.14-1.07 No
NRT1 113/... 338 693/... 0.10-0.83 No
~Lower bound and mean predictions.

Acknowledgments
Thi s i nvest i gat i on was p e r f o r me d f or ACCURATE, I NC. , in Whi t ewat er , WI , wi t h Mr.
J a ya nt Sha r ma as t he key pr oj ect engi neer. The exper i ment al par t o f this i nvest i gat i on was
pe r f or me d b y DATASYST, I NC. , in Del afi el d, WI , wi t h Messr s. Da ve Hoi s i ngt on and Tom
Ki nzel wer e t he key st aff.
APPENDIX A
Torsional Data Analysis
Ros et t e st ai n gages mount e d on t he 1-in. di a me t e r sect i on o f the shaf t j us t out si de o f t he
spl i ned sect i on wer e us ed t o moni t or st rai ns and eval uat e the appl i ed nomi nal t orgues. Por t -
abl e hi s t ogr a m r ecor der s [3] wer e used t o col l ect t he st rai n hi st or i es and pr oces s the dat a
usi ng the " r a i n f l ow" count i ng t echni que. St r ai n- bas ed cumul at i ve f at i gue d a ma g e anal ysi s
t echni ques [1,2] wer e us ed t o eval uat e t he data.
The f our bas i c equat i ons us ed f or pe r f or mi ng t he strain bas ed cumul at i ve f at i gue d a ma g e
anal ysi s ar e as f ol l ows
(1) The Cycl i c St r ess- St r ai n Cur ve
A~/2 = (A~r/2)/E + [(A~r/2/K')] ~"'
(2) The st r ai n ver sus Li f e Cu r v e
A~/ 2 = ( ~ ' s / E) (2Ns)b + ~'S<2NS ~c
(3) The nomi na l ver sus No t c h Root St r ess- St r ai n Rel at i ons hi p ( Ne ube r Rul e)
K,(As/2) = [(A~r/2)(A~/2)E] ' / 2
(4) The Li ne a r Da ma g e Summa t i on Rul e
D, = N~/N s
Dtota I = ~ Di
e=l
I n the a b o v e equat i ons, t he el ast i c modul us, E; t he cycl i c st r engt h coeffi ci ent , K i the cycl i c
st rai n har deni ng exponent , n; t he f at i gue st r engt h coef f i ci ent , ~r's; t he f at i gue st r engt h ex-
ponent , b; t he f at i gue duct i l i t y coeffi ci ent , ~'r and t he f at i gue duct i l i t y exponent , c, ar e the
mat er i al f at i gue pr oper t i es. Th e s e pr oper t i es wer e es t i mat ed f r om t he known t ensi l e pr op-
er t i es o f t he mat er i al usi ng est abl i shed empi r i cal equat i ons avai l abl e in t he l i t erat ure [1]. The
act ual val ues us ed in t he pr es ent anal ysi s are l i st ed in Tabl e 4,
For a gi ven nomi na l st ress, As~2, and a sel ect ed val ue o f t he f at i gue not ch fact or, K s, t he
da ma ge , D i, cor r es pondi ng t he n u mb e r o f appl i ed cycl es, Ni, i s cal cul at ed by s i mul t aneous
sol ut i on o f t he equat i ons s hown pr evi ousl y. Typi cal l y, t hese cal cul at i ons are a c c ompl i s he d
t hr ough a c o mp u t e r p r o g r a m [2]. Tabl e 5 i l l ust rat es a t ypi cal c umul a t i ve da ma ge cal cul at i on
f or t he t est ed paddl e OLT2. Th e ni ne l evel s o f t or si onal ampl i t udes and cor r es pondi ng cycl es
exper i enced by this paddl e ar e l i st ed in col umns 2 and 3, r espect i vel y. The absol ut e and

JHANSALE ON AGITATOR PADDLE SHAFTS
TABLE 4--Estimated fatique properties f or shaft material (1018
cold rolled steel).
9 Tensile Strength = 627 MPa, measured
9 Percent reduction of area = 56, measured
9 Elastic modulus = 206 000 MPa
9 Fatigue strength coefficient, e ' = 965 MPa
9 Fatigue strength exponent, b = - 0. 08
9 Fatigue ductility coefficient, e} = 0.425
9 Fatigue ductility exponent, c = - 0. 60
9 Cyclic strength coefficient, K' = 1083 MPa
9 Cyclic strength exponent, n' = 0.137
75
pe r c e nt val ues of t he d a ma g e s c a l c ul a t e d f or e a c h o f t he l oa d l evel s ar e al s o s hown. The s e
c a l c ul a t i ons we r e r e pe a t e d f or var i ous s el ect ed val ues o f K s unt i l t he t ot al d a ma g e of uni t y
wa s obt a i ne d. As s h o wn i n Ta bl e 5, t he K s va l ue f or t he pa ddl e OLT2 is 1.45. The a mpl i t ude
t hat has t he ma x i mu m pe r c e nt c ont r i but i on i s c h o s e n as t he e qui va l e nt s i ngl e a mpl i t ude . I n
t he e x a mp l e s h o wn i t i s t he e i g h t h l evel wi t h a 6 9 . 0 2 % c o n t r i b u t i o n t o da ma ge . Th e cor r e-
s p o n d i n g e qui va l e nt cycl es ar e obt a i ne d by di vi di ng t he act ual cycl es e xpe r i e nc e d at t hi s
a mpl i t ude by t he c o n t r i b u t i o n f r act i on. I n t he pr e s e nt cas e, i t i s 16 1 4 7 / 0 . 6 9 0 2 = 23 395.
Ta bl e 1 s u mma r i z e s t he s e r es ul t s f or al l t he t est s.
APPENDI X B
General Outline of BS5400, Part 10
The f at i gue l i f e of a we l d e d c o mp o n e n t i s g o v e r n e d b y s ever al f act or s i nc l udi ng t he
ma t e r i a l pr oper t i es , t he r e s i dua l st r esses, t he i nhe r e nt fl aws, s uc h as unde r c ut s a nd over l aps ,
a nd ot he r s t r es s r ai s er s due t o ge ome t r i c a l di s c ont i nui t i e s o f t he wel d. Taki ng al l o f t hes e
f act or s i nt o c ons i de r a t i on, t he Br i t i s h St a nda r d BS5400, Par t 10 ( 1980) , Co d e o f Pr a c t i c e f or
TABLE 5--Typical torsional fatigue data analysis f or specimen OLT2.
Component Factor, K s = 1.45
Equivalent Uniaxial Contribution
Nomi nal Stress, Cycles, Damage to Damage,
Event MPa, s/ 2 Ni Ni/Ni %
1 170.4 46 5.2E-6 0.00
2 191.7 879 3.8E-4 0.04
3 213.0 176 2.3E-4 0.02
4 234.3 2 216 7.7E-3 0.77
5 255.6 407 3.1E-3 0.31
6 276.9 11 551 1.7E- 1 17.23
7 298.3 4 697 1.2E- 1 12.47
8 319.5 16 147 6.9E-1 69.02
9 340.9 22 1.4E- 1 0.14
Totals 1.0 100.00
No'rE--equivalent nominal stress = 319.5 MPa (Event 8); equivalent torque amplitude = 315.7 N'm and equivalent
cycles = 16 147/0.69 = 23 395,

76 C A S E S T U D I E S F O R F A T I G U E E D U C A T I O N
3 0 0 ,-
, o o r I I I 1
, , o K a P I t l l l l l I
z 70 N r " q d I I " r - ~ X I - . , , L ~ ! ~ . I I I r t - . ? ' - - - , t : . ~ : 5 . _1
b~ 60
50
c t , O
o
' ~ 3 0
2 0 - -
1 5
1 0
10 s 2
I I I IIII
I I I I I I I I l , ~
! ! l
, , , I I I I I I I
I ~ r ~ t T ~ l ~ ~ r ~ F ~ l ~ l ' % ~ l l l ~ t ~ I - l ' ~ l J _ l _ 1 1 1 1 6 0
I I F I " N . C " ~ J - F z ' t - . . ~ I - - . L ' I ' $ E / " F . " 1 " - L I I I ] ' I T - ' < f - - - - 4.-...1_ I I I I
" - ~ - q ~ ~ - - 2 . - v - ! I T H - I I I N - - - L I F I T l t t - , o
- - ~ -i ~ ~ .
I I I / ~ - - - . - - ~ . - > : : F - I . I . I t l F k l H - I f l I - L L , o
, , I , , 1 , 1 , , , I , I
3 s 6 7 . ~ , ~ 6 s , 2 4 6 a , I T" ~ I - - I - I ' I - l ' 1 5 -
l o l o ~l I " ' 1 , ~ 1 " t ' l J I r
6 1 I I I " l - q . l l ' l - "
s l I I I I I l l ' t " s
E n d u r a n c e N ( r 2 t, 6 8 10 l
FIG. 4--Fatigue design curves, as per British Standard BS5400.
Fat i gue: St eel Concret e and Concret e and Composi t e Bri dges, " has devel oped a fami l y of
fat i gue desi gn curves (Fig. 4) for vari ous wel d confi gurat i ons cl assi fi ed with letter codes.
These curves are pl ot s of appl i ed stress range versus cycl es- t o- f ai l ur e at sel ect ed levels of
probabi l i t y. The stresses to be us ed for the t ype of wel ds consi dered in this study are max-
i mum pri nci pal stresses. For hi st ori es compr i sed of mul t i pl e ampl i t udes, a l i near damage
summat i on pr ocedur e is used.
The wel ds in ol d paddl es were cl assi fi ed as F2 wel ds and t hose in the new paddl es as
Cl ass E to account for r epor t edl y bet t er wel di ng pract i ce in the new paddl es. For each paddl e,
t wo sets of fat i gue curves represent i ng mean and mean mi nus t wo st andard devi at i ons (es-
sent i al l y a l ower bound curve) wer e used to pr edi ct l ower bound and average bl ocks to
failure, as shown in Table 3.
R e f e r e n c e s
[1] Society of Automotive Engineers, Fatigue Design Handbook, AE-10, Society of Automotive Engi-
neers, Warrendale, PA, 1988.
[2] Jhansale, H. R. and McCann, D. R. "Fatigue Analysis Techniques for Vintage Steam Turbine/
Generator Components," Advances in Fatigue Lifetime Predictive Techniques, ASTM STP 1122,
M. R. Mitchell and R. W. Landgraf, Eds., American Society for Testing and Materials, Philadelphia,
1992, pp. 474-489.
[3] Socie, D., "A Field Recording System with Applications to Fatigue Analysis," Technical Report
132, Electrogeneral Corporation, Minneapolis, MN, 1977.

J. Edgar Zapat a I and St ephen C. Anderson t
Fatigue Cracki ng of a Wel ded Roll Used in
a Paper-Mi l l Roll Press
REFERENCE: Zapata, J. E. and Anderson, S. C., "Fatigue Cracking of a Welded Roll Used
in a Paper-Mill Roll Press, " Case Studies for Fatigue Education, ASTM STP 1250, Ralph I.
Stephens, Ed., American Society for Testing and Materials, Philadelphia, 1994, pp. 77-85.
ABSTRACT: A 1-m fatigue crack was found after approximately four months of service in a
welded steel roll that had an intended life of 20 years. The roll was part of a roll press used
in a paper mill. The fatigue crack originated in a surface discontinuity and propagated parallel
to the girth weld. Metallographic examination revealed the discontinuity to have been produced
by entrapped welding slag. The discontinuity was 5.5-mm deep by 41-mm long and was small
with respect to the overall size of the roll (8.6-m long, 1.5-m in diameter, with a 110-mm wall
thickness). Furthermore, the nominal bending stresses in the part were found to be far below
the endurance limit of the material. A fatigue crack propagation study, based on Paris law
fatigue crack growth, was performed to determine if the stress level in the roll was sufficient
to propagate the fatigue crack from the initial flaw size in the transpired time. Further studies
were performed to investigate the effect of different flaw sizes and loading conditions on the
life of the roll.
KEYWORDS fatigue, weld, defect, slag, paper mill, roll
A paper mi l l rol l , whi ch was expect ed to f unct i on f or at l east 20 year s, devel oped a l ar ge
cr ack af t er onl y f our mont hs o f ser vi ce. Thi s f r act ur e caused s ome concer n as ot her r ol l s
wer e oper at i ng i n si mi l ar ci r cums t ances . Of si gni f i cant i mpor t ance was t hat t he oper at i onal
l oads pr oduced st resses that wer e subst ant i al l y be l ow t he endur ance l i mi t of t he mat er i al .
Background
Operation
The paper mi l l rol l in ques t i on was t he bot t om r ol l of a t wo- r ol l pr ess used i n t he pr o-
duct i on o f paper. The f unct i on of t he rol l pr ess was t o r educe t he wat er cont ent o f pul p as
an i ni t i al st ep i n t r ans f or mi ng pul p t o paper. The pul p was car r i ed on a conveyor be t we e n
t he t wo r ol l s o f t he r ol l press. Li near act uat ors wer e in pl ace above t he t op rol l t hat s er ved
t o pr oduce a di st r i but ed l oad on t he rolls. The t op r ol l pr oduced a di st r i but ed l oad on t he
pul p of 350 N/ mm. Thi s l oad was suppor t ed ent i r el y by t he bot t om rol l , as i l l ust r at ed in
Fi g. 1.
The bot t om rol l , in turn, was suppor t ed by t wo j our nal bear i ngs wi t h t he cent er - t o- cent er
di st ance bei ng 9 800 mm. It r ot at ed at 150 r pm and had been in cont i nuous oper at i on f or
about f our mont hs f r om t he t i me it was i nst al l ed unt i l , t he 1-m cr ack was det ect ed.
Senior mechanical engineer and company president, respectively, Anderson & Associates, Inc., 919
FM 1959, Houston, TX 77034-5425.
77

9800m
8600mm
350 N/ram
! ',
. 2867mm
Fract ure
Locat i on
i
Journal
-Beari ng
Both Sides
FIG. 1--Schematic of the bottom roll in the papermi l l roll press. The distributed l oad of 350 N/ mm
is shown to be acting over 8600 ram. The entire assembly is supported by j ournal bearings.
Fabr i c at i on
The rol l was fabri cat ed by t aki ng t hree mi l d steel plates, r ol l i ng t hem to round, and seam
wel di ng t hem to pr oduce t hree cyl i nder s of equal size. The t hree cyl i ndri cal sect i ons were
t hen j oi ned by t wo gi rt h wel ds to form the roll. Af t er wel di ng, the entire rol l was stress-
rel i eved by heat i ng to 580~ The rol l was t hen i nspect ed and the surface machi ned. A
r ubber coat i ng was appl i ed to the out si de di amet er surface to pr ovi de a l arger cont act area
wi t h the pul p. The final l engt h of the cyl i nder was 8600 mm, t he di amet er 1500 mm, and
t he wal l t hi ckness 110 mm. The rol l mass, not i ncl udi ng heads and j ournal s, was 32 200 kg.
Ex a mi n a t i o n
The rol l was t aken out of servi ce and the r ubber coat i ng r emoved for exami nat i on of the
fracture. The crack was ci rcumferent i al and was in the appr oxi mat e l ocat i on of one of the
gi rt h wel ds. The fracture was r emoved by t orch cut t i ng about 200 mm away from t he crack
surfaces. A uni form l ayer of rust was vi si bl e on the entire fracture surface, whi ch was
smoot h. The fracture ori gi n area was on t he out si de di amet er surface and was surrounded
by fai nt radi al marki ngs. The fracture ori gi n area and areas cl ose to the final st ages of the
fract ure were r emoved and cl eaned el ect r ol yt i cal l y in a sodi um- cyani de sol ut i on [1]. The
rust came of f easi l y and in a uni form fashi on. No excessi ve cor r osi on or mechani cal damage
was found in t he ori gi n area. The ori gi n site appear ed to have occur r ed at the surface where
a pecul i ar mar k was present, shown in Fi g. 2, Thi s mar k was appr oxi mat el y 41- r am l ong in
the ci rcumferent i al di rect i on, 5. 5-ram wi de in t he radi al di rect i on, and 1-mm deep in t he
l ongi t udi nal di rect i on. The mar k was rest ri ct ed to one of t he fracture surfaces; the mat i ng
fract ure surface was smooth.
The ori gi n area, crack-t ermi nat i on area, as wel l as i nt ermedi at e areas were exami ned in
the scanni ng el ect ron mi cr oscope (SEM). Al t hough macr oscopi cal l y the fracture faces were
free from mechani cal damage, at the mi cr oscopi c l evel no fract ure mode was di scerni bl e.
Evi dence of mechani cal damage was vi si bl e and corrosi on pi t s covered the surface. A di f-
ference, however, was di scer ni bl e bet ween the surface of the fract ure and the t opography at
the ori gi n site. As shown in Fi g. 3, the ori gi n site has a smoot her surface than the fracture.
A cl ear del i neat i on of the boundary bet ween t he t wo is al so vi si bl e.

ZAPATA AND ANDERSON ON FATIGUE CRACKING OF A WELDED ROLL 79
FIG. 2- - The pre-existing f l aw at the fracture origin is bounded by the arrows. Faint radial lines can
be seen surrounding the flaw. The scale is in inches (1 in. = 2.54 cm).
A cross-section was taken through both fracture faces at the origin site showing the entire
wall thickness. The cross section was surface ground and etched in 10% Nital. The fracture
could be seen to have originated at a discontinuity on the outside diameter surface. This
discontinuity was at the edge of the girth weld. The crack extended from this discontinuity
across the wall thickness perpendicular to the inside and outside diameter surfaces (Fig. 4).
A smaller cross-section containing the discontinuity was removed from the through-wall
cross section previously discussed. The sample was mounted, ground, polished, and etched
in 3% Nital. On one side of the discontinuity the girth weld acicular microstructure was
readily observed. On the other side of the weld the microstructure was consistent with a hot
rolled low-carbon steel. No coarse grain heat affected zone was evident on either side of the
discontinuity. Elsewhere, surrounding the weld, heat-affected zones were clearly discernible
(Figs. 5 and 6).
Hardness tests showed that the weld and the heat-affected zone were harder than the base
metal. The base metal hardness was approximately 66 HRB, the weld metal ranged from 85
to 90 HRB, and the heat-affected zone ranged from 83 to 87 HRB.
Student Questions
9 What is a possible explanation for the presence of the discontinuity at the fracture
origin?

FIG. 3--Fractograph showing the boundat 3' between the preexisting fla~t; top, and the.~ltigue crack,
bottom. The boultdat-v is indicated by the arrows.
FIG. 4--Photomacrograph of a cross-section taken through the fatigue fracture and the initial flaw.
The area to be mounted is marked in black. The scale is in inches (1 in. = 2.54 cm).

FIG. 5--Phot ornacrograph of the initial f l aw and the f at i gue crack. The scale is in inches (1 in. =
2,54 cm).
9 What is the fracture mechanism? How can this be substantiated?
9 What was the nominal stress at the origin site before the crack initiated?
9 How does the amplitude of the reverse bending stress in the roll compare to published
endurance limits for low-carbon steel?
9 How can a fatigue crack start and grow under such a low amplitude of cyclic stress?
How can this be substantiated?
9 What are possible remedies to the cracking situation? What are their limitations?
Analysis
Examination of Fig. 6, the photomacrograph taken through the fracture origin, gives some
insight into the presence of the discontinuity. The columnar weld microstructure is clearly
seen on one side of the discontinuity whereas the other side shows an unaffected base metal.
Furthermore, the base metal side tapers slightly with respect to the normal to the outer
diameter surface of the roll. This taper is probably the original bevel in the plate present to
facilitate welding. A foreign object must have been present to prevent the weld metal from
contacting the base metal in that area. Additionally, this object must have been an insulator
so as to shield the base metal from heat of welding, thereby preventing a heat-affected zone
adjacent to the discontinuity.
Macroscopically, the fracture had all the characteristic fatigue features. The crack was fiat
and perpendicular to the principal stress direction and radial features surrounded the initiation
area in a semicircular fashion. The smoothness and lack of excessive corrosion pointed to

FIG. 6--Photomicrograph of the initial flaw. The columnar microstructure of the weld is at the left.
The base metal, right, was not affected by the heat of welding, indicating that an insulator was present.
purely mechanical fatigue. The relative ease with which the rust was removed suggested that
the corrosion had not significantly attacked the surface. Furthermore, since the corrosion
product came off in a uniform fashion, the entire crack was probably exposed to the corrosive
environment for roughly the same amount of time. It was concluded that, at least for the
initial stages of the fatigue fracture, the environment did not contribute to the fracture because
it was isolated by the rubber coating.
Examination of the pre-existing flaw showed that it was long, slender, and parallel to the
welding direction. The flaw was not planar but exhibited some thickness. It was confined to
the weld metal and, as shown in the metallographic examination, it was an insulator. En-
trapped welding would clearly fit the description of such a defect.
At the microscopic level, no fracture features, like striations, were observable under SEM
examination. It is not uncommon for these to be erased in parts undergoing reverse bending.
Additionally, the crack faces were exposed to a slightly corrosive environment that would
serve to erase the microscopic fracture features.
The nominal stress in the roll was calculated by assuming the roll to be a simply supported
beam under a distributed load. The journal bearings were assumed to be incapable of sup-
porting a bending load, thus the justification for simple supports. The distributed load was
taken from the operational parameters as 350 N/ r am that acted over the roll length of 8600
mm. The section modulus was easily found from the dimensions to be 0.156 m 3. The bending
moment was primarily due to two factors, the distributed press load and the weight of the
bottom roll. The total weight of the roll, including the heads and journals was unknown, but
it was assumed that the same distributed load produced by the cylindrical section was con-
stant for the bearing-to-bearing distance. This was a reasonable assumption since the mass
of the roll was greatest close to the bearing, and the distributed press load only acted on the

cylindrical section. The bending moment at the fracture was therefore calculated to be 4.18
MN m. The static tension stress was in turn found to be 26.8 MPa on the outside diameter
surface at the fracture plane at the bottom of the roll.
A quick survey of the literature shows endurance limits for low-carbon steels to be much
higher than the fluctuating stress in the roll. Endurance limit values were found to range
from about 90 MPa up to 200 MPa for various conditions of welds and base metal [2-4].
The low nominal stress at the fracture compared to the reported endurance limits raised
questions about the effect of the initial flaw size on the life of the roll. A fatigue crack
propagation study was therefore undertaken to determine if such an initial flaw size could
account for a large through-wall fatigue crack in the time the roll was in service. The analysis
assumed the initial flaw was a sharp crack instead of a three dimensional discontinuity
detected. The Paris equation for fatigue crack propagation [5] was employed.
da/ dN = A ( mg) m
where
a = the crack length parameter, m;
N = the number of cycles;
K = the stress intensity range, MPa ~/-~;
A = a material constant; and
m = a material constant.
Rolfe and Barsom [6] report material constants of A = 3.6(10 -1~ and m = 3.0 for K in ksi
~]-~. and a in inches for steels with a ferrite/pearlite microstructure. The material constants
become A = 6.9(10 12) and m = 3.0 for K in Mpa X~mm and the crack length a in meters.
The equation for the stress intensity AK used was [71
A~aa
AK-
Q
where
= the tensile stress amplitude, MPa;
Q = crack shape parameter; and
a = the crack length parameter, m.
The tensile stress amplitude, A~r, was assumed to act only due to the loading and did not
consider any effects of residual stress.
The crack growth study using the operational loads found that the crack growth rate was
insufficient to account for the fatigue crack encountered. It was judged, however, that since
the failure occurred next to a weld, the effects of the residual stress could not be neglected.
Although a stress relief had been performed on the part, it is generally assumed that some
residual stresses remain. In the absence of actual data, it was assumed that the residual
stresses served to raise the alternating stresses so that the entire range of stress fluctuation
due to bending was tensile. The net effect was to double the amplitude of the stress from
26.8 to 53.6 MPa.
With the stress amplitude doubled by the presence of tensile residual stresses, the Paris
equation adequately predicted the crack propagation. Figure 7 shows fatigue crack propa-
gation curves considering both alternating stress situations and the approximate location of
the actual fracture. It should be noted that the constants A and m used in the Paris equation

8 4 CASE STUDIES FOR FATIGUE EDUCATION
~ o ~ ~ ~,
i 0 0 m ~
90 " ~ ~ _ " ~
7 0 ~ u '~"
6Q
f
'-" 50 ' ~
2 4o
3O
2O
10 ~
I I I
106 10 ] 108
Number o f Cycl es
FIG. 7 1 C r a c k depth versus the number o f cycles. The predi ct i ons were cal cul at ed wi t h and wi t hout
consi deri ng a tensile residual stress f r om welding. The initial f l aw depth was taken as 5. 5 ram.
90
80
7O
I = :
E 60
~. 5O
_~ 4 0
3O
2O
i 0
106 107 108
Number of Cycl es
4 -
o ~
....1
t ~
FIG. 8- - Cr ac k depth versus the number o f cycles. The predi ct i ons were cal cul at ed with and without
consi deri ng a tensile residual stress f r om welding. The initial f l aw dept h was taken as 0.25 ram. The
desi red life was 1.5(10 ~) cycles.

predict conservative estimates fatigue crack propagation and actual fatigue cracking should
lie to the right of the curves.
After ascertaining that the operational stresses were sufficient to cause the fatigue fracture
in the time expended, the next logical step was to determine what size of a flaw could be
tolerated for the roll to survive the desired life of 20 years. If a strict Paris crack growth
model was employed, the initial flaw size would have to be intolerably small. Figure 8 shows
two curves for an initial flaw size of 0.25 mm under the two stress situations as compared
to the desired life of 1.5 (109 ) cycles. It can be seen that this poses an unrealistic situation
as it is not possible to detect an initial flaw of that size. Using a strict Paris equation, the
desired life could not be designed for practically.
A viable alternative would be to consider the threshold stress intensity factor for fatigue
crack propagation. The flaw in this case, if assumed to be planar, would have given rise to
a stress intensity of 6.59 MPa ~/-m. A flaw of similar length but only 2.5-mm in depth would
have produced a stress intensity of 4.67 MPa ~ that is below the reported threshold stress-
intensity factor for fatigue crack propagation in a similar material [8]. Using threshold stress-
intensity factors for fatigue can allow for larger flaws than would otherwise be ruled out
when using a strict Paris equation.
References
[1] Dahlberg, E. R and Zipp, R. D., "Clean Breaks and SEM," Chemteck, Washingon, DC, Feb. 1985,
pp. 118-122.
[2] Boyer, H. E., Atlas of Fatigue Curves, American Society for Metals, Metals Park, OH.
[3] Bosch, R., Automotive Handbook, 2nd English Edition, Society of Automotive Engineers (SAE) Inc.,
Warrendale, PA, Delta Press Limited, 1986, p. 142.
[4] Campbell, J. E., Gerberich, W. W., and Underwood, J. H., Application of Fracture Mechanics for
Selection of Metallic Structural Material, American Society for Metals, Metals Park, OH.
[5] Paris, P. C. and Erdogan, E, "A Critical Analysis of Crack Propagation Laws," Journal of Basic
Engineering, Vol. 85, 1960, pp. 528-534.
[6] Rolfe, S. T. and Barsom, J. M., Fracture and Fatigue Control in Structures--Applications of Fracture
Mechanics, Prentice-Hall, Englewood Cliffs, NJ, 1977, p. 239.
[7] British Standards Institution, "Guidance on Some Methods for the Derivation of Acceptance Levels
for Defects in Fusion Welded Joints," PD6493:1980, British Standards Institution, March 1980.
[8] Taira, S., Tanaka, K., and Hoshina, M., "Grain-Size Effect on Crack Nucleation and Growth in
Long-Life Fatigue of Low Carbon Steel," Fatigue Mechanisms, ASTM STP 675, Jeffrey T. Fong,
Ed., American Society for Testing and Materials, Philadelphia, 1979, pp. 135-162.

Darrell Socie' and Ellen Segan 2
Welded Pipeline Fatigue Analysis
REFERENCE: Socie, D. and Segan, E., "Wel ded Pipeline Fatigue Anal ysi s, " Case Studies
f or Fatigue Education, ASTM STP 1250, Ralph I. Stephens, Ed., American Society for Testing
ABSTRACT: This case study shows the importance of considering welding flaws in pipelines
that may be subjected to cyclic pressures. The case study will demonstrate that the ASTM
specifications currently used to purchase welded pipe do not ensure adequate safety for cyclic
loading.
A piping system was installed using Specification for Seamless and Welded Austenitic Stain-
less Steel Pipe (ASTM A 312) welded pipe. Longitudinal welds are made without the addition
of filler metal by resistance welding from both sides of the pipe wall. This process frequently
results in lack of fusion (LOF) flaws in the pipe. Hydrostatic testing used to demonstrate
conformance with ASTM standards does not detect flaws that will grow in service.
An example of a failed pipeline system is presented. Pressure loading histories were obtained
during normal operation of the piping system. Crack growth materials data were available for
a fracture mechanics analysis of the system. A discussion of the ASTM standards as they relate
to this problem/~re included.
KEYWORDS: fatigue failues, design codes, lack of fusion (LOF) flaws
Background
Ther e have been sever al pr emat ur e f ai l ur es of st ai nl ess st eel f uel i ng syst ems des i gned in
accor dance wi t h t he Ame r i c a n Soci et y f or Mechani cal En g i n e e r s / Ame r i c a n Nat i onal St an-
dards Inst i t ut e ( AS ME / ANS I ) B31. 3 Pi pi ng Code f or Che mi c a l Pl ant and Pet r ol eum Ref i n-
er y Pi pi ng. Thes e f ai l ur es pose si gni f i cant safet y and envi r onment al risks. A f uel i ng s ys t em
is an under gr ound pi pi ng syst em that is des i gned to pump fuel f r om above gr ound st or age
t anks to f uel i ng t rucks that cont r ol t he fl ow of fuel to an ai rcraft . Fai l ur e consi st s of a r upt ur ed
pi pe that al l ows f uel to l eak i nt o t he sur r oundi ng soil. In addi t i on t o t he envi r onment al
damage, fuel in s ome cases has r eached t he sur f ace and coul d have r esul t ed in a seri ous fire.
Fai l ur es at t he pump house have not yet occur r ed but coul d be cat ast r ophi c as t he vol at i l e
fuel c ome s i nt o cont act wi t h el ect r i c mot or s and swi t ch gear. An anal ysi s of t he fuel s ys t em
f ai l ur es was made to i dent i f y t he cause or causes of t he f ai l ur e and to make r e c omme nda t i ons
f or new and exi s t i ng syst ems. The f ol l owi ng quest i ons we r e addr essed:
9 Why di d t he f uel l i nes fai l ?
9 Ar e mor e f ai l ur es l i kel y?
9 Di d t he f ai l ed pi pes meet speci f i cat i ons?
9 Is t he des i gn adequat e?
9 What is t he appr opr i at e cour se of act i on f or oper at i on of exi st i ng syst ems and des i gn
of fut ure s ys t ems ?
' University of Illinois at Urbana-Champaign, Mechanical Engineering Department, Urbana, IL 61801.
2 U.S. Army, Construction Engineering Research Laboratory, Champaign, IL 61826.
86

SOCIE AND SEGAN ON WELDED PIPELINE FATIGUE ANALYSIS 87
In any failure anal ysi s of a mechani cal syst em t here is al ways a quest i on as to whet her
t he fai l ure is due to a mat eri al or manufact uri ng flaw, over l oad in service, change in function
or i nt ended purpose, or a desi gn oversi ght . Al l of t hese fact ors must be consi der ed and t he
most l i kel y cause det ermi ned. Frequent l y, failures are a resul t of a combi nat i on of causes.
Inevi t abl y, r esponsi bi l i t y for causi ng and r epai r i ng the fai l ure wi l l be est abl i shed. In addi t i on
to the t echni cal quest i ons l i st ed above, t he answer to " Wh o caused the pr obl em?" is oft en
det er mi ned from the t echni cal eval uat i on t hrough l i t i gat i on. Wi t h this in mi nd, it is i mport ant
to det er mi ne the chai n of events that l ed to the failure.
Why Did the Fuel Lines Fail?
An 8-ft sect i on of 16-in. di amet er ASTM A 312 st ai nl ess steel schedul e 10 pi pe (from
ASTM Speci fi cat i on for Seaml ess and Wel ded Aust eni t i c St ai nl ess Steel Pi pe) that l eaked
in servi ce wa s r emoved for testing and anal ysi s. Before excavat i ng the fai l ed pi pe section,
t he entire fuel i ng syst em was drai ned and cl eaned of r esi dual fuel. A spark from the repai r
operat i on coul d have i gni t ed resi dual fuel and ext reme car e was taken whi l e worki ng around
t he fuel i ng syst ems. Fort unat el y, this part i cul ar fai l ure occurred on the side of a taxiway.
The pi pi ng syst em is bur i ed 6 ft into t he ground. Fuel l eaked into the ground, rose to the
surface, and was det ect ed in the grass. Had the pi pe rupt ure occurred underneat h the rei n-
f or ced concret e runway, it woul d have been difficult to accurat el y l ocat e the even more
expensi ve to repair. A schemat i c i l l ust rat i on of the pi pe is gi ven in Fi g 1. The pi pe was
fabri cat ed from a pl at e that was rol l ed into a cyl i nder and resi st ance wel ded al ong the seam.
Ther e was a vi si bl e l ongi t udi nal crack in t he wel d on t he out si de of the pi pe appr oxi mat el y
6-in. long. The owner of the syst em cl ai med that the fai l ure was a result of defect i ve pi pe
i nst al l ed by the cont ract or who i nst al l ed the system. Not surpri si ngl y, the cont ract or cl ai med
that the failure was t he resul t of an over l oad due to i mpr oper usage of the syst em by the
owner. Each si de can oft en find a consul t ant to support t hei r poi nt of view. In this case, an
i ndependent consul t ant was sel ect ed to det er mi ne the fai l ure origin. It was essent i al to con-
si der all possi bl e causes and not act as a referee bet ween opposi ng vi ewpoi nt s.
The cross sect i on of the pi pe was measur ed to det er mi ne i f the fai l ure was the result of
an overl oad. A t ransi ent pressure coul d have been pr oduced by the sudden cl osure of a val ve
and the resul t i ng pressure coul d have burst the pipe. Pressure in the syst em coul d also have
exceeded the desi gn pressure due to some mechani cal mal f unct i on in the pump house. Mea-
surement s of the pi pe shape di d not show any i ndi cat i on that the pi pe was pl ast i cal l y de-
f or med duri ng failure. The duct i l e stainless steel pi pe woul d have bul ged out war d i f excessi ve
pressure caused t he failure. Radi ogr aphy showed that a cr ack- l i ke flaw ext ended wel l beyond
the vi si bl e surface crack. These observat i ons l ead to t he pr el i mi nar y concl usi on that an
over l oad was not r esponsi bl e for the failure. The pi pe was cut into smal l sect i ons to exami ne
the cr ack- l i ke flaw.
A section of pi pe 32-in. l ong and 6-in. wi de cont ai ni ng the wel d was r emoved from the
pi pe for det ai l ed met al l ur gi cal exami nat i on. The pi pe was t ransversel y sect i oned at various
l ocat i ons al ong the length. Fi gur e 2 shows the fracture surface that was reveal ed duri ng
sect i oni ng. Three zones are shown in Fig. 2: l ack of fusion, fatigue crack growth, and final
fracture. Fat i gue cracks that have not yet grown to the out si de surface were al so observed
in many of the ot her t ransverse sections. They were et ched to show the wel ds wi t h a t ypi cal
result, shown in Fi g. 3. Ar r ows i ndi cat e the ori gi nal posi t i on of the seam in the pipe. In a
good wel d, the wel ds are l ocat ed on the cent erl i ne of t he seam. Bot h i nsi de and out si de
wel ds are l ocat ed off cent er in this section. The out si de wel d (top of Fi g. 3) shows evi dence
of a growi ng fat i gue crack. The i nsi de wel d compl et el y mi sses the wel d seam and the out si de
wel d does not penet rat e to the cent er of t he plate. Fi gur e 4 shows a si mi l ar exampl e. Here

FIG. 1--Pipe illustration.
the i nsi de wel d is off cent er but the out si de wel d penet rat es to the cent er of the plate. This
results in a smal l er initial flaw size than the one shown in Fig. 3. A fatigue crack has not
yet formed. These t ypes of flaws woul d not be found duri ng a visual i nspect i on of the out si de
of the pipe. Pi pes are purchased in 40-ft lengths so that it woul d also be difficult to find
t hem on the i nsi de as well. These flaws coul d have been found duri ng an ultrasonic or
radi ographi c i nspect i on of the pipe. Ironi cal l y, the field-butt wel ds where two pi pes are joi ned
were radi ographed. There were no requi rement s for radi ographi ng the factory wel ds that
failed in service. Internal LOFs (lack of fusion) that were not vi si bl e from ei t her surface
were also found and are shown in Fig. 5. This t ype of flaw is very difficult to find with
ei t her radi ography or ul t rasoni c i nspect i on techniques. Both t echni ques require an air gap
for good resolution. Al l of the transverse sect i ons near the failure crack show LOFs that
ext end appr oxi mat el y one- hal f to t wo-t hi rds of the wal l thickness.
Operat i ng pressures of the syst em were moni t ored for five weeks to det ermi ne the distri-
bution of operat i ng pressures duri ng normal operat i on of the fuel i ng system. These pressures

FIG. 2--Fracture surface showing: (a) fracture zone (0.011 in.; 0.028 cm), (b) fatigue zone (0.045
in.; 0.114 cm), (c) lack of fusion zone (0.110 in.; 0.280 cm), and (d) pipe thickness (0.165 in.; 0.420
cm).
were measured at the poi nt of failure after the fai l ed pi pe was repl aced. A hi st ogram of the
pressures is given in Table 1. Dat a have been normal i zed to occurrences per year. Rainflow
cycl e count i ng was used to det ermi ne the ranges. These measurement s were used for a fatigue
anal ysi s of the pi pi ng system. Pressures were measured wi t h a pressure t ransducer and re-
corded with appropri at e bat t er y- power ed l ong-t erm data acqui si t i on equipment. Fi gure 6
shows the pressure for a 2 h peri od when two ai rcraft were bei ng fueled. Mul t i pl e pressure
FIG. 3--Off'cent er welds.

FIG. 4- - Of f center il~side weld.
drops to near zero were observed as the valves were opened and pressure increases were
observed when the valves were closed. Several pressure cycles were observed for each air-
craft fueling. Test data showed that the maximum operating pressure never exceeded 250
psi (1724 kPa), which was within the design pressure of 275 psi (1896 kPa) for this system.
Similar measurements made on another fueling system confirmed that the measured pressures
were typical of normal service usage.
A simple limit load calculation was used to make an estimate of the final fracture zone,
shown in Fig. 2. Readers unfamiliar with these calculations and assumptions will find an
excellent review in Chapter 2 of Ref 1. Stainless steel has a high ductility and is plastically
deformed in the net section before failure. As a result, plane strain fracture toughness cal-
culations were not appropriate. Hoop stress, ~hoop, in the uncracked ligament could have been
estimated from the stress calculated with thin-walled tube assumptions.
PF
%oop- t - a (1)
Operating pressure in the pipe was denoted P (250 psi; 1724 kPa), pipe wall thickness t
(0.165 in.; 0.419 cm), pipe radius r (8 in.; 20.32 cm), and flaw size, a. The maximum flaw
size was estimated from a simple lower bound limit-load calculation by substituting the flow
stress %ow, for the hoop stress and treating the pipe as a plate.
O'yield + O'ultmlate
~rn~ = 2 (2)
Test reports from the pipe manufacturer showed that the stainless steel piping had tensile
strength varying from 81.2 to 94.2 ksi (559.85 to 649.49 MPa), yield strengths varying from

FIG. 5--Internal lack of fusion (LOF).
33.8 to 54.6 ksi (233.04 to 376.45 MPa), and elongation varying from 31.3 to 63%. The
resulting average flow stress was 66 ksi (455.05 MPa). An estimated final crack size of 0.135
in. (0.34 cm) and uncracked ligament of 0.030 in. (0.008 cm) was calculated. This compares
to the uncracked ligament of 0.011 in. (0.03 cm), shown in Fig. 2. As expected, a simple
lower bound solution overestimates the actual uncracked ligament size. Bending stiffness in
TABLE 1--Measured pressure ranges.
Pressure Range, psi " Occurrences per year
260 35
240 17
220 122
200 1502
180 716
160 506
140 506
120 646
100 1380
80 3807
60 5396
" 1 psi = 6.8948 kPa.

0 I I I
o_ 0 0.5 I.O 1.5
T i m e , h o u r s
FIG. 6 - - P r e s s u r e i n t he p i p e d u r i n g a n a i r c r a f t f u e l i n g o p e r a t i o n .
i
2 . 0
t he l ongi t udi nal di rect i on was i gnor ed in the si mpl e cal cul at i on. Thi s cal cul at i on cl earl y
shows that onl y a smal l fract i on of the pi pe wal l is needed to support the nor mal operat i ng
pressure in the fuel i ng syst em. It al so provi des further veri fi cat i on of t he concl usi on that the
pi pe was not over l oaded in service. An uncracked pi pe woul d requi re mor e than six t i mes
t he normal operat i ng pressure to cause a burst or over l oad failure.
Fract ure mechani cs was used to make an est i mat e of t he t i me requi red for the LOF to
gr ow to a critical size for failure.
~ da
Ny = . C A K " (3)
The cycl i c stress i nt ensi t y factor, AK, for l ongi t udi nal cr acks in a pi pe subj ect ed to internal
pressure is avai l abl e from the l i t erat ure [2]. Mat eri al s dat a for crack gr owt h (C and m) in
st ai nl ess steel are al so avai l abl e [3]. Ini t i al , ao, and final, a~, crack sizes can be sel ect ed from
t he precedi ng figures and cal cul at i ons. Numer i cal i nt egrat i on usi ng the t echni ques out l i ned
in Ref 4 were used to obt ai n est i mat es of the fatigue l i ves gi ven in Fig. 7 for various initial
flaw sizes. These est i mat es were based on mean mat eri al dat a and shoul d be i nt erpret ed as
50% pr obabi l i t y of fai l ure and not desi gn lives. They compar ed wel l with the observed fai l ure
t i me of one year.
Met al l ogr aphi c anal ysi s in Fi gs. 2 t hrough 5 showed that LOFs were r esponsi bl e for
nucl eat i ng fatigue cracks. Nor mal operat i ng pressures of t he syst em were sufficient to cause
t he crack to grow to a cri t i cal si ze that al l owed the pi pe to rupture and l eak during the l ast
few cycl es of l oadi ng. Est i mat ed l i fe from the fatigue anal ysi s for the i ni t i al crack of 0.11
in. (0.28 cm) t aken from Fig. 2 was wel l within the normal scatter in fat i gue life when
0 . 1 2
0 . 1 0 e -
U
I =
- - 0 . 0 8
r.~ 0.06
0
0.O4
0
002
c
0.00
1 0 1 0 0 lOCK)
Est i mat ed Life, Ye a r s
FIG. 7 - - E s t i m a t e d f a t i g u e l i f e a s a f i m c t i o n o f i n i t i a l f l a w si ze.

compared to the observed life of one year. This correlation provides additional verification
of the measurements, observations, and calculations and allows us to concl ude that the failure
was a result of flaws in the pipe and not abnormal operation.
Are More Failures Likely?
Experience with the systems provides further substantiation of the fatigue calculations.
Fatigue cracks have been discovered in other unfailed sections of pipe that have been ex-
cavated to l ook for LOF flaws. ASTM A 312 pipe has been extensively investigated for
applications in the nuclear industry. A summary of the relevant work from Cipolla et al.
[5,6] on lack of penetration defects in double seam welded A 312 stainless steel pipe is
given below by reproducing appropriate conclusions from their work.
9 Neither radiographic examination nor conventional ultrasonic examination techniques
are able to detect the defect reliably.
9 A log normal distribution of flaw size data indicated a mean flaw size of 12% of the
wall thickness and the size of flaw for 95% occurrence level with 95% confidence is
36% of the pipe wall thickness.
Results of these studies and experience with fueling systems demonstrates that ASTM A
312 pipe will contain LOFs, many of which cannot be identified during the manufacturing
process using conventional post weld inspection techniques. The piping system contains
additional LOFs that will nucleate fatigue cracks and grow to failure under normal operation.
There is no known way to determine the quality of the pipe (more than 10 000 feet) that is
already installed in a single fueling system. Therefore, one must concl ude that more failures
are indeed likely unless the operating conditions are changed to eliminate the cyclic loading.
Did the Failed Pipes Meet Specifications?
Sixteen inch schedule 10 pipe was purchased, according to ASTM A 312. The maj or
specifications relevant to this investigation are summarized below with the same numbering
system used in the specification:
3.0 General Requirements
(a) material conformi ng to ASTM A 530/ A 530M 3
5.0 Materials/Manufacture
(a) no additional filler metal
6.0 Chemical Composition
10.0 Mechanical Tests
(a) transverse or longitudinal tension tests
(b) flattening test
(c) hydrostatic test
12.0 Workmanship, Finish, and Appearance
(a) reasonably st rai ght / workmanl i ke finish
(b) imperfections can be removed with grinding providing wall thickness require-
ments are maintained
$5 Supplementary Requirements
3 Specification for General Requirements for Specialized Carbon and Alloy Steel Pipe.

( a ) radiographic examination
The major specifications included in specification ASTM A 530/ A 530M are summarized
as follows:
6.0 Hydrostatic Test
(a) not less than 50% of specified mi ni mum yield strength for austenitic steel
11.0 Variation in Wall Thickness
(a) mi ni mum wall thickness at any point shall not be more than 12.5% under the
nominal wall thickness specified.
Chemical composi t i on was measured and compared to the specification for 304L stainless
steel. Particular attention is placed on distinguishing between 304 and 304L. Austenitic
stainless steels with higher carbon contents are known to be sensitive to intergranular attack
known as weld decay. Al t hough there was no mi croscopi c evidence that weld decay played
any role in the failure, it was considered important to verify the chemical composition. I f
the pipe did not meet the composition requirements, further investigation to determine how
the chemical composition may have affected the failure would be needed. In this case the
pipe material was found to meet the nominal compositional requirements for 304L.
All of the mechanical specifications relate to performance under static loading. These tests
will not determine the performance under cyclic loading. A static pressure test was used to
test the integrity of the pipe to meet the A 312 specification. The tensile requirements for
ASTM A 312 304L stainless steel require mi ni mum tensile and yield strengths of 70 and
25 ksi (482.63 and 172.369 MPa), respectively. However, delivered pipe typically exceeds
these mi ni mum requirements as described above. Pipes containing the LOFs shown in Figs.
2 through 5 could easily pass the hydrostatic test specified in the Specification for General
Requirements for Specialized Carbon and Alloy Steel Pipe (ASTM A 530). It is interesting
to note that the design code ASME/ ANSI B31.3 allows operating stresses as large as 2/3 of
the mi ni mum yield strength while the ASTM standard only requires hydrostatic testing to
I/2 of the mi ni mum yield strength.
Lack of fusion in the pipe has clearly been demonstrated to be a contributing factor to
the failures, yet there is no specific requirement for weld quality. It is common practice in
the pressure vessel industry to have specific requirements for weld quality and to make the
weld inspection requirements part of the purchase specification. No weld inspection re-
quirements were specified when purchasing this pipe; therefore, it could incorrectly be as-
sumed by the supplier that weld quality was not critical. It is interesting that the owner did
not specify radiography for the factory longitudinal welds but did specify 100% radiography
for the field installed butt welds that j oi n sections of pipe together. No failures have yet been
observed in any of the field welds. Stresses in these welds are only hal f the magnitude of
the hoop stresses.
The specification clearly states that mi ni mum wall thickness at any point shall not be more
than 12.5% under the nominal wall thickness specified. Schedule 10 pipe calls for a nominal
wall thickness of 0.188 in. (0.477 cm). The mi ni mum allowable thickness is 0.164 in. (0.416
cm) compared to the measured value of 0.165 in. (0.420 cm) in the failed pipe. The question
now becomes: Does an LOF constitute a lack of wall thickness? Common sense would
dictate that the weld is part of the pipe wall and should be subjected to the wall thickness
requirement. The counter argument is that LOFs are a known flaw that results from the
welding process and are excluded from the wall thickness requirements, particularly i f no
radiography is specified. The pipe supplier may concl ude (rightly or wrongly) that there is

no part i cul ar concern about LOFs si nce no r adi ogr aphy was request ed. Does the presence
of an LOF come under the good wor kmanshi p pr ovi si on of the speci fi cat i on? The si t uat i on
is further compl i cat ed by the st at ement in the speci fi cat i on that i mperfect i ons can be r emoved
with gri ndi ng provi di ng wal l t hi ckness requi rement s are mai nt ai ned. There is no quest i on
that del i vered product s were abl e to meet the accept ance tests but t here is quest i on as to
whet her the pi pe meets the mi ni mum wall requi rement s in regi ons where there is l ack of
fusion.
Is the Design Adequate?
The pi pi ng syst em was desi gned with AS ME/ ANS I B31.3. As is the case for many
syst ems, there is no specific desi gn code for under gr ound fuel i ng syst ems. As a result, the
desi gner must use some j udgment in sel ect i ng the appropri at e sect i ons of the code. The code
is used to det er mi ne the size and wal l t hi ckness (schedul e) of the pi pe requi red for a gi ven
pressure and number of operat i ng cycl es.
ASME/ANSI B31.3 Summary
A short summar y of relevant sect i ons of AS ME/ ANS I B31.3 are reproduced here usi ng
the numberi ng syst em from the code. 4
The Code sets forth engi neeri ng requi rement s deemed necessary for safe desi gn and con-
struction of pressure piping. Whi l e safety is the basi c consi derat i on, this factor al one wi l l
not necessari l y govern the final speci fi cat i ons for any pi pi ng system. The desi gner hand-
book does not do away wi t h the need for the desi gner or for compet ent engi neeri ng
j udgment .
300.2 Definitions
Severe cyclic conditions--Those in whi ch SE, comput ed in accordance with para.
319.4.4 exceeds 0.8Sa (as defi ned in para. 302.3.5), and the equi val ent number of cycl es
(N in para. 302.3.5) exceeds 7000; or ot her condi t i ons that the desi gner det ermi nes wi l l
produce an equi val ent effect.
NoTE- - Bot h SE and SE are used in the code and have di fferent meani ngs.
301.2 Design Pressure
The desi gn pressure of each component in a pi pi ng syst em shall be not less than t he
pressure at the most severe condi t i on of coi nci dent internal or ext ernal pressure and
t emperat ure.
301.10 Cyclic Effects
Fat i gue due to pressure cycl i ng, thermal cycl i ng, and ot her cycl i c l oadi ngs shall be
consi dered in the desi gn of pi pi ng.
4 Permission to reprint ASME B31.3, Chemical Plant and Petroleum Refinery Piping, granted from
the American Society of Mechanical Engineers, New York, NY.

301. 5 Dynamic Effects
See Appendix F, para. F301.5.
301.5.1 I mpact - - I mpact forces caused by external or internal conditions (including
changes in flow rate, hydraulic shock, liquid or solid slugging, flashing, and geysering)
shall be taken into account in the design of piping systems.
302.2.4 Al l owances f or pressure and temperature variations, metallic pi pi ng--Occa-
sional variations of pressure or temperature, or both, above operating levels are char-
acteristic of certain services. The most severe conditions of coincident pressure and
temperature during the variation shall be u s e d . . .
302.3.5 Limits of calculated stresses due to sustained loads and displacement
st rai ns--(a) Internal Pressure Stresses. Stresses due to internal pressure shall be consid-
ered safe when the wall thickness of the piping component, and its means of stiffening,
meet the requirements of para. 304. (b) External Pressure Stresses. Stresses due to
external pressure shall be considered safe when the wall thickness of the piping com-
ponent, and its means of stiffening, meet the requirements of para. 304. (c) Longitudinal
Stresses S L. The sum of longitudinal stresses due to pressure, weight, and other sustained
loadings S~ shall not exceed SH in (d) below. The thickness of pipe used in calculating
S L shall be the nominal thickness T minus mechanical, corrosion, and erosion allowance
c. (d) Allowable displacement stress range S A. The allowable displacement stress range
(see para. 319.2.3) for the comput ed displacement stress range SE (see para. 319.4.4)
shall be
SA : f ( I . 2 S Sc + 0. 25 S.)
In the above equation
S,. = basic allowable stress at mi ni mum metal temperature expected during the dis-
placement cycle under analysis,
SH = basic allowable stress at mi ni mum metal temperature expected during the dis-
placement cycle under analysis, and
f = stress range reduction factor for displacement cycl e conditions for the total
number of cycles over the expected life (Table 1).
302.3.6 Limits o f calculated stresses due to occasional l oads--(a) Operation. The sum
of the longitudinal stresses due to pressure, weight, and other sustained loadings, SL and
of the stresses produced by occasional loads, such as wi nd or earthquakes, may be as
much as 1.33 times the basic allowable stress given in Appendi x A. For castings, the
basic allowable stress shall be multiplied by the casting quality factor E,. Where the
allowable stress value exceeds two-thirds of yield strength at temperature, the allowable
stress value must be reduced as specified in Note (3) in para. 302.3.2. Wind and earth-
quake forces need not be considered as acting concurrently. Table 2, which follows,
shows the stress range reduction fact ors?
304. Pressure Design of Metallic Components
304.1.2 Straight pi pe under internal pressure--(a) For t < D/ 6, the internal pressure
design thickness for straight pipe shall be not less than that calculated in accordance
with

SOCIE AND SEGAN ON WELDED PI PELI NE FATIGUE ANALYSIS
TABLE 2--Stress range reduction factors. 5
Cycles Factor f
< 7000 1.0
7 000 to 14 000 0.9
14 000 to 22 000 0.8
22 000 to 45 000 0.7
45 000 to 100 000 0.6
> 100 000 0.5
97
PD
t =
2SE
The following nomenclature is used in this equation:
t = design thickness,
P = internal design gage pressure,
D = outside diameter of pipe,
S = allowable stress value from Table A-l , and
E = quality factor from Table A- I A or A-l B.
(b) For t > D/ 6 or for P / S E > 0.385, calculation of pressure design thickness for
straight pipe requires special consideration of factors such as theory of failure, effects
of fatigue, and thermal stress.
305. 2. 3 Pi pe f o r severe cycl i c condi t i ons - - Onl y the following pipe may be used under
severe cyclic condition: (c) Stainless Steel Alloy Pipe (ASTM A 268, 6 seamless; ASTM
A 312, seamless; ASTM A 358, 6 E,. > 0.90; ASTM A 376; 6 ASTM A 430; 6 and ASTM
451,6 E,. > 0.90).
319. 2. 3 Di spl acement stress r a n g e - - As contrasted with stresses from sustained loads,
such as internal pressure or weight, displacement stresses may be permitted to attain
sufficient magnitude to cause limited overstrain in various portions of a piping system.
When the system is initially operated at the condition of greatest displacement (highest
or lowest temperature, or greatest imposed movement) from its installed condition, any
yielding or creep brings about a reduction or relaxation of stress. When the system is
later returned to its original condition (or a condition of greater displacement), a reversal
and redistribution of stresses occurs which is referred to as selfspringing. It is similar
to cold springing in its effects.
5 This table is reproduced here from the ASME Piping Code B31.3, with permission from ASME.
6 AS T M A 268--Specification for Seamless and Welded Ferritic Martensitic Stainless Steel Tubing
General Service.
ASTM A 358--Specification for Electric-Fusion-Welded Austenitic Chromium-Nickel Alloy Steel
Pipe for High Temperature Service.
ASTM A 376--Specification for Seamless Austenitic Steel Pipe for High-Temperature Central-Station
Service.
ASTM A 430--Specification for Austenitic Steel Forged and Bored Pipe for High-Temperature
Service.
ASTM A 451--Specification for Centrifugally Cast Austenitic Steel Pipe for High Temperature
Service.

319.4.5 Required Weld Quality Assurance--Any wel d at whi ch SE exceeds 0.8SA (as
defi ned in par agr aph 302.3.5 for any port i on of the pi pi ng syst em, and the number of
cycl es N exceeds 7000, shal l be ful l y exami ned in accor dance wi t h par agr aph 341.4.3.
341.4.3 Examination-severe cyclic conditions--Metallic pi pi ng to be used under se-
vere cycl i c condi t i ons shall be exami ned to the ext ent speci fi ed herei n or any great er
extent speci fi ed in t he engi neeri ng desi gn. Accept ance cri t eri a are as st at ed in paragraph
341.3.2 and in Tabl e 341. 2. 3A, appl i cabl e to severe cycl i c condi t i ons, unl ess ot herwi se
specified. (b) Al l ci rcumferent i al butt and mi t er gr oove wel ds and all fabri cat ed branch
connect i on we l d s . . , shall be exami ned by 100% radi ography.
Design Calculation Summary
Const ruct i on proj ect s such as fuel i ng syst ems usual l y have a f or mal desi gn analysis. Al l
cal cul at i ons support i ng the desi gn are document ed. The fol l owi ng cal cul at i ons were used to
select the si ze of the pi pe used in t he fuel i ng syst em [7]. Some edi t i ng has been done to
el i mi nat e cal cul at i ons not rel event to the pi pe failure.
NOTE- - Met r i c conversi on for t he dat a that fol l ows is: 1 ksi = 6. 8948 MPa.
Maximum pressure--l.5 t i mes maxi mum syst em pressure equals 225 psi; maxi mum
al l owabl e pump pressure equals 250 psi. Check compl i ance wi t h ANSI B31.3.
301.2 Design pressure--302.2.4, over 7000 cycl es per l i fet i me, must use most severe
condi t i ons.
301.5.1--Impact by surge is consi der ed si nce short durat i on stresses are al l owabl y
hi gher than static pressures.
302.2.3 Design temperature--Since t emperat ure is bel ow 100~ use fl ui d temperature.
302.3--Allowable stress (SE) is the maxi mum al l owabl e stress (S) mul t i pl i ed by the
qual i t y fact or (E). Qual i t y factor, defi ned in 302.3.4, equal s 0.85.
302.3.2 Basis for allowable stress--Minimum t ensi l e stress = (70 ksi ) ( 89 = 23.3; 2A
mi ni mum yi el d = (25 ksi)(2/3) -- 16.7; use 16.7.
302.3--Allowable stress SE = (16.7 ksi)(0. 85) = 14.2 ksi.
302.3.5--Allowable displacement stress range:
SA = f ( 1. 25 S c + 0.25 SH),
S c = 3", = 14.2,
f = 0.7 for up to 45 000 cycl es (assumed),
SA = 14.9, and
SA > SE; therefore, di spl acement stress is not a factor.
304.1 Internal Pressure
12.5% t ol erance on t hi ckness.
P = (0.188.0 0. 875) (14.2 ks i ) / 8 = 292 psi.
The desi gner then concl udes that al l pi pe sizes and schedul es are accept abl e per the code.
We now consi der the desi gn of t he system. Sel ect i on of the appr opr i at e wal l t hi ckness for
the st eady-st at e pressure in the pi pi ng sysem is st rai ght forward. Consi der at i on of the cycl i c
l oadi ng is somewhat probl emat i c. Longi t udi nal wel ded ASTM A 312 pi pe is not permi t t ed
i f the desi gner det er mi nes that severe cycl i c condi t i ons exi st (see Sect i on 305.2.3 of the
repri nt ed ANSI B31.3 Code). The cri t i cal issue is in det er mi ni ng the magni t ude of the stress

cycles and the number of stress cycles that constitute severe cyclic loading conditions. The
code contains several references to severe cyclic conditions. The only reference to allowable
stresses for severe cyclic loading are given in Section 302.3.5 of ANSI B31.3 for displace-
ment stresses. Paragraph 319.2.3 of B31.3 makes it quite clear that stresses due to internal
pressure are not displacement stresses. It appears that the designer only considered fatigue
by comput i ng the displacement stress range. A designer more familiar with fatigue would
have recognized a problem with Section 302.3.5 of the analysis. How can the comput ed
maxi mum stress for fatigue loading (14.9 ksi; 102.73 MPa) exceed the allowable static stress
(14.2 ksi; 97.90 MPa)?
The code requires that cyclic loading be considered but only provides guidance for the
case of displacement stresses. Could the requirements of the code be met by applying a
stress range reduction factor from Table 302.3.5 of ANSI B31.3 to the design allowable
stress, SE? There is nothing in the code that would prohibit this approach. The result would
have been a larger pipe wall thickness as the allowable stress would be reduced to (0.7)(14.2
ksi) = 9.9 ksi (97.90 and 68.26 MPa, respectively), resulting in schedule 20 pipe rather than
schedule 10 pipe. The designer would have considered cycl i c loading from internal pressure,
made an allowance for it with a stress range reduction factor, selected the appropriate wall
thickness and concl uded that severe cyclic loading conditions did not exist. Since the de-
signer concl uded severe cyclic loading conditions do not exist, welded A 312 pipe is per-
mitted and paragraph 341.4.3 of ANSI B31.3 is not applicable; therefore, radiography is not
mandatory.
Designers refer to codes such as this one because they are not experts in all aspects of
failure. Unfortunately, only an expert woul d have recogni zed the possibility of crack-like
LOF flaws and concl uded that severe cycl i c loading conditions existed for the fueling system.
What is the Appropriate Course of Action for Operation of Existing Systems?
Results of these studies and experience with fueling systems demonstrate that ASTM A
312 pipe contains LOFs, many of which cannot be identified with conventional post-weld
inspection techniques. The piping code recognizes this and does not permit ASTM A 312
welded pipe to be used under severe cyclic loading. This pipe should not be used in cyclic
applications unless the influence of LOFs is specifically taken into consideration in the
design. As a result, we believe that continued operation of the systems will result in more
failures. Reduci ng the operating pressures in the system will prolong the life of the system
but it is impossible to determine how l ong since the flaw size distribution is unknown.
Conclusions
The technical aspects of the failure are clearly established. LOF flaws initiated fatigue
cracks that grew to failure under normal operations of the fueling system. Limitations in the
design code and purchasing specifications of the pipe pl ayed a major role in the failure.
Designers, through the various design codes, need to consider the possibility of flawed welds
under cyclic loading. Design codes need to be i mproved in this area.
Discussion Questions
The following questions set forth by the author are listed here for further discussion.
9 What is the role and responsibility of the designer, pipe supplier, contractor, and system
operator in the failure?

9 How can t hese smal l LOFs be found in a buri ed pi pel i ne? What t echni ques are avai l abl e
to i nspect the pi pe bef or e i nst al l at i on?
9 Assumi ng the same size LOF, how woul d the burst pressure and fat i gue lives be al t ered
i f 304 was used i nst ead of 304L?
9 I f 100% r adi ogr aphy had been specified, what is the maxi mum LOF size that woul d be
per mi t t ed in the pi pe? How wi l l it be measur ed?
9 Do you t hi nk the pi pe meet s the speci fi cat i ons? How coul d the ASTM speci fi cat i on be
i mpr oved? Remember that a good speci fi cat i on has a means to measure, quantify, and
val i dat e compl i ance. Const ruct i on proj ect s are frequent l y awar ded on the basi s of the
l owest bidder. Shoul d the pi pe suppl i er pay for 12.5% more mat eri al than needed to
meet the mi ni mum speci fi cat i on?
9 Is consi der i ng di spl acement stress range as an i ndi cat i on of severe cycl i c condi t i ons a
cont ri but i ng fact or to the fai l ure? Coul d par agr aph 304.1.2 (b) of ANSI B31.3 be in-
t erpret ed by a non-expert that fatigue onl y needs to be consi dered i f t > D/ 6 ?
9 What t ypes of l oadi ng const i t ut e severe cycl i c l oadi ng? I f the cycl i c stresses were low
enough, t he LOFs woul d not grow si nce t hey woul d be bel ow the t hreshol d stress
intensity. Woul d this be severe cycl i c l oadi ng? How coul d the desi gn code be i mpr oved?
Woul d schedul e 20 pi pe be bet t er than schedul e l 0 pi pe? Based on the work of Ci pol l a
et al., schedul e 20 pi pe woul d be expect ed to have larger LOF flaws. Is it r easonabl e
to assume that desi gners are expert s on fracture mechani cs?
References
[1 ] Broek, D., The Practical Use of Fracture Mechanics, Kluwer Academic Publishers, The Netherlands,
1989.
[2] Raju, I. S. and Newman, J. C., Jr., "Stress Intensity Factors for Internal and External Surface Cracks
in Cylindrical Vessels" Journal of Pressure Vessel Technology, Vol. 104, 1982, pp. 293-298.
[3] James, L. A., "Fatigue Crack Propagation in Austenitic Stainless Steels," Atomic Energy Review,
Vol. 14, No. 1, 1976, pp. 37-85.
[4] Brussat, T. R., "An Approach to Predicting the Growth to Failure of Fatigue Cracks Subjected to
Arbitrary Uniaxial Cyclic Loading," Damage Tolerance in Aircraft Structures, ASTM STP 486, 1971,
pp. 122-143.
[5] McNaughton, W. E, Egan, G. R., and Cipolla, R. C., Significance of Centerline Lack-of-Penetration
in Double Welded Steel Pipe--Fracture Analysis, Aptech Engineering Services, Inc, Palo Alto, CA,
1979.
[6] Cipolla, R. C., Grover, J. L., Beaupre, G. S., and McNaughton, W. E, Lack-of-Penetration Defects
in Double-Seam-Welded SA-312 Stainless Steel Pipe, Aptech Engineering Services, Inc, Palo Alto,
CA, 1980.
[7] "Supplement to Final Design Analysis," Ellsworth Air Force Base, South Dakota, U.S. Army Corps
of Engineers, July 1984.

Shr i kant P Bhat ~
Thermal Fatigue Analysis: A Case Study
of Recuperators
REFERENCE: Bhat, S. R, "Thermal Fatigue Analysis: A Case Study of Recuperators,"
Case Studies for Fatigue Education, ASTM STP 1250, Ralph I, Stephens, Ed., American Society
ABSTRACT: Continually increasing energy cost is an impetus to increasing the operating
temperatures in industrial recuperators. However, higher service temperatures increase the risk
of thermal fatigue failures. This paper presents a case study involving thermal fatigue analysis
of a recuperator in an integrated steel plant.
The paper first describes the background and the rationale for a proposal to operate recu-
perators at the Inland 80" Hot Strip Mill at a higher than normal temperature for the purpose
of fuel savings. A quick engineering analysis identified thermal fatigue as a high risk item that
needed to be analyzed before deciding to operate the recuperators at a higher temperature. To
quantify the resulting thermal effects, temperature profiles were recorded on an experimental
basis for one furnace run at both the normal and higher operating temperatures. The paper
describes the steps and assumptions involved in analyzing the data and estimating the thermal
fatigue life. The calculations indicate that increasing the operating temperature from 1100 ~ to
1250~ (593 to 677~ may reduce the thermal fatigue life of the recuperator by as much as
63%. Based on these calculations and practical difficulties in achieving higher degree of skin
temperature regulation, the option of operating recuperators at a higher temperature was con-
sidered unattractive.
KEYWORDS: thermal fatigue, recuperators, elastic strain, plastic strain, cumulative damage
I n spite of maj or i mpr ovement s , ener gy cost s r epr esent 28 to 30% of steel manuf act ur i ng
cost i n the Uni t ed St at es [1]. Thus, t he economi c i mpet us for i mpr ovi ng t he effi ci ency of
recuperat ors, a c o mmo n l y used met hod for wast e heat recovery, is qui t e clear. Concept ual l y,
a recuperat or r ecycl es the heat f r om wast e exhaust gases f r om a f ur nace to pr eheat the
i nc omi ng combus t i on ai r or fuel. Thi s resul t s i n l ower wast e gas t emper at ur e and r educed
fuel and air i nput [1]. I n the earl y 80s, the Ener gy Technol ogy Team at I nl and Steel began
f i ndi ng ways to r ecover mor e ener gy f r om wast e gases as a means to r educe total ener gy
cost s at the 80" Hot St ri p Mi l l . One means of i ncr easi ng t he effi ci ency of recuperat ors is to
i ncr ease t he oper at i ng t emperat ures. The pr oposal i ncl uded r ai si ng t he r ecuper at or ski n t em-
per at ur e so that t he combus t i on ai r woul d be pr eheat ed to a hi gher t emperat ure. However,
i n r ai si ng the oper at i ng t emper at ur es, t rade off bet ween ener gy savi ngs and the pot ent i al
r educt i on i n component life must be consi der ed. Obvi ousl y, it is pr udent to carry out a ri sk-
benefi t anal ysi s bef or e changi ng the oper at i ng t emper at ur e. The Ener gy Technol ogy Team
i dent i fi ed t her mal expansi on, creep, hi gh- t emper at ur e gaseous corrosi on, and t hermal fat i gue
as the maj or ri sk fact ors for furt her anal ysi s. Engi neer i ng est i mat es made, based on the
physi cal di mens i ons and mat er i al s' data, pr ovi ded by t he manuf act ur er s of t he recuperat or,
Section manager, Inland Steel, Research and Development Department, 3001 East Columbus Drive,
East Chicago, IN 46312.
101

convi nced the t eam that t hermal expansi on, creep, and hi gh-t emperat ure gaseous corrosi on
woul d not det eri orat e the r ecuper at or l i fe si gni fi cant l y as a result of i ncreasi ng the operat i ng
t emperat ure. (See the next sect i on for the magni t ude of t emperat ure i ncrease bei ng consi d-
ered. ) To address t he remai ni ng cri t i cal i ssue of t hermal fatigue, t he i nvol vement of the R&D
Depart ment and the present aut hor was requested. The case st udy present ed here is based
on the t hermal fat i gue anal ysi s conduct ed as part of an engi neeri ng anal ysi s of the effects
of i ncreasi ng t he operat i ng t emper at ur e of the 80" Hot Strip Mi l l recuperator. The intent is
to demonst rat e how basi c pri nci pl es of fatigue anal ysi s may be appl i ed in det ermi ni ng
whet her or not rai si ng the oper at i ng t emperat ure wi l l have det ri ment al effects on the servi ce
l i fe of steel mi l l recuperat ors.
Topics f or Further Study
What are the di fferent mat eri al s used in the const ruct i on of an i ndust ri al recuperat or and
what factors go into deci di ng the opt i mum operat i ng t emperat ure?
Thermal History
In this section, attention is f ocused on i dent i fyi ng what dat a are pert i nent to represent the
ant i ci pat ed t hermal cycl i ng hi st ory of the recuperat or and how to i ncorporat e the t ypi cal
vari at i ons in mi l l operat i ng condi t i ons.
Based on experi ence and r ecommendat i ons of the equi pment supplier, the recuperat ors at
t he Inl and 80" Hot St ri p Mi l l were operat ed wi t h a ski n t emperat ure set poi nt of 1100~
(593~ The pr oposal consi st ed of i ncreasi ng the ski n t emperat ure to 1250~ (677~ by
cont rol l i ng wat er sprays in the wast e gas stream. Thi s i ncrease in ski n t emperat ure was
est i mat ed to i ncrease the preheat ai r t emperat ure by about 120F~ 2 resul t i ng in a final ai r
t emperat ure of 800~ (427~ Si nce the t hermal l oad on the recuperat ors depends on the
mi l l worki ng schedul e, any real i st i c t hermal fatigue anal ysi s has to be based on the actual
t hermal profile rat her than the set point. Therefore, to quant i fy the effects of i ncreasi ng the
recuperat or skin t emperat ure, one furnace was sel ect ed as the candi dat e to be run at the
hi gher t emperat ure on an exper i ment al basis. A mi ni mum of 48 h of t emperat ure profi l e
recordi ngs were suggest ed to capt ure t ypi cal vari at i ons in mi l l operat i ng condi t i ons.
Fi gur e 1 is a schemat i c of the recuperat or. 3 The skin t emperat ure was moni t or ed at a poi nt
where previ ous experi ence had shown to i ndi cat e the hi ghest operat i ng t emperat ure (Fig. 1).
Thi s appr oxi mat el y represent s t he regi on where the t emperat ure di fference bet ween the out -
goi ng wast e gas and the i ncomi ng air is maxi mum and is t ypi cal l y subj ect to maxi mum
fluctuations in t emperat ure. The ski n t emperat ure profi l es were r ecor ded on an expanded
scal e for the t wo recuperat ors in the furnace. For each recuperat or, t emperat ure profiles were
recorded at two l ocat i ons on the same el evat i on (i dent i fi ed as north and south t hermocoupl es).
In addi t i on, to veri fy the i mpact of i ncreasi ng t he ski n t emperat ure, a six channel Beckman
r ecor der was used to moni t or and r ecor d combust i on ai r and wast e gas t emperat ures.
Temperat ure profiles were r ecor ded as det ai l ed above wi t h skin t emperat ure set poi nt s at
1100~ (593~ for 64 h and at 1250~ (677~ for 53 h. (The mi l l operat i ons di ct at ed the
di fferent lengths of the observat i on peri od). In the fol l owi ng, dat a from the run with set
poi nt at 1100~ (593~ wi l l be used to est abl i sh the base line for the anal ysi s.
2 The nomenclature of 120F ~ to denote a range (A) of temperature, and 120~ to denote the actual
temperature, will be used throughout this paper. The author gratefully acknowledges one of the reviewers
for suggesting this unambiguous notation.
3 Bugyis, E. J., unpublished reports, Inland Steel Company, 1982-83.

BHAT ON THERMAL FATIGUE ANALYSIS 103
SPRING HANI O
SUPPORTS
OF FIN ~, !
NASTE GASES
OUT
COLD
AIR
IN
" 1 " BEAM OF
RECUPERATC
SUPPORT TO~
THERMOCOU
FOR SKIN TEl
i
HECUPERATOR SHELL
SLOTTED - ALLOWS
FIN TO EXPAND
VERTICALLY
EXPANSION JOINT
ALLOWS FIN TO
EXPAND VERTICALLY
PACKING PERMITS
/ INDEPENDENT
q MOVEMENT OF
INNER AND OUTER
SHELLS
HOT
AIR
OUT
WASTE GASES
IN
AIR JACKET GLAND
FIG. 1--Schematic dr awi ng o f t he cross sect i on o f a r ecuper at or at t he I nl and 80" Hot St ri p Mi l l
( not to scale).

Visual inspections of temperature recordings revealed very similar profiles on both recu-
perators at the two recording locations. Thus, only one location on one recuperator was
chosen for detailed analysis.
Thermal Fatigue Analysis
When a component undergoes temperature fluctuations and the natural expansi ons/ con-
tractions are either wholly or partially inhibited, thermal stresses are produced that can lead
to fatigue failures. This is called "t hermal fatigue." It is emphasized that it is not thermal
expansion that causes fatigue but rather the restraint of this expansion (or contraction) that
leads to fatigue. The details of the nature and problems of thermal fatigue can be found in
Ref 2.
For the purpose of the present analysis, the recuperator is considered essentially free-
hanging, and, as seen in Fig. 1, a packi ng gland at the bot t om of the recuperator allows the
inner and outer shells to expand independently of one another. Further, it is assumed that
over a period of time, the recuperator reaches the naturally expanded state corresponding to
the operating temperature. This is especially true since the recuperator is slowly heated.
(Typical firing time is approximately 5 h.) However, during it' s operation, due to variations
in heat input demands at the furnace, fluctuations occur in the recuperator skin temperature
lasting less than 30 rain per cycle. In the present analysis, it is assumed that this short interval
of time is not sufficient for "nat ural " (entire skin) expansi on/ cont ract i on of the recuperator.
Further, for ease in calculation, it is assumed that these local thermal expansions/contractions
are fully constrained and are the cause of thermal fatigue. In other words, the damage
accumulates due to temperature fluctuations rather than due to a specific operating temper-
ature. The direct effects of changing the operating temperature (within the range considered
here) are the minor changes in the material cyclic properties, which, as an engineering
approximation, are ignored in the present study. However, inspection of the temperature
recordings revealed that increasing the operating temperature results in a greater number of
temperature fluctuations of larger magnitude. Steps involved in calculating the relative effects
of these temperature fluctuations are explained below. Because the proposed change in set
temperature is comparatively small (120F ~ or 67C ~ and the primary interest is in the relative
effects of such changes in the operating temperature, a simple uniaxial analysis can be made,
although temperature fluctuations may lead to more compl ex multiaxial strain/stress
situation.
Step 1--Temperature Fluctuations
From the expanded temperature profile recordings, fluctuations were categorized into four
groups of magnitude: 99, 198, 297 and 396F ~ (55, 110, 165 and 220C~ respectively. In
general, the cycle time was found to be independent of both the magnitude of temperature
fluctuations and the operating set point. Typically, each fluctuation lasted about 15 to 20
min. The number of fluctuations in each group for the entire recording period were counted
without regard to their sequence. The results are summarized in Table 1 for both 1100~
(593~ and 1250~ (677~ aim skin temperatures. It is significant to note that at the higher
operating temperature, there were three temperature cycles of 396F ~ (220C ~ magnitude;
whereas, none was found of this magni t ude at the lower operating temperature. In addition,
the number of fluctuations increased as the amplitude of temperature cycles decreased for
bot h operating temperatures.

T
A
B
L
E

l
-
-
T
h
e
r
m
a
l

f
a
t
i
g
u
e

c
a
l
c
u
l
a
t
i
o
n
s

f
o
r

r
e
c
u
p
e
r
a
t
o
r
.

M
a
g
n
i
t
u
d
e

o
f

T
e
m
p
e
r
a
t
u
r
e

F
l
u
c
t
u
a
t
i
o
n
,

F
~

~

9
9
(
5
5
)

S
e
t

P
o
i
n
t

l
l
0
0
~

(
5
9
3
~

S
e
t

P
o
i
n
t

1
2
5
0
~

(
6
7
7
~

b

1
9
8
(
1
1
0
)

2
9
7
(
1
6
5
)

3
9
6
(
2
2
0
)

9
9
(
5
5
)

1
9
8
(
1
1
0
)

2
9
7
(
1
6
5
)

3
9
6
(
2
2
0
)

N
o
.

o
f

t
e
m
p
e
r
a
t
u
r
e

c
y
c
l
e
s

1
5

9

1

(
S
t
e
p

1
)

T
o
t
a
l

s
t
r
a
i
n

r
a
n
g
e

(
~
A
T
)

1
.
0
3

2
.
0
5
9

3
.
0
8
8

1
0
-
3

(
S
t
e
p

2
)

P
l
a
s
t
i
c

s
t
r
a
i
n

r
a
n
g
e

(
A
e
p
~
)

0
.
0
5
1

1
.
0
8

2
.
1
1

1
0

3

(
S
t
e
p

2
)

N
o
.

o
f

C
y
c
l
e
s

t
o

F
a
i
l
u
r
e

>

1
0
6

3
0

0
0
0

1
0

0
0
0

(
S
t
e
p

3
)

D
a
m
a
g
e

n
~
I
N
~

(
S
t
e
p

4
)

0

0
.
0
0
0
3

0
.
0
0
0
1

T
o
t
a
l

d
a
m
a
g
e

f
o
r

t
h
e

0
.
0
0
0
4

o
b
s
e
r
v
a
t
i
o
n

p
e
r
i
o
d
,

"
2

n
,
/
N
,

(
S
t
e
p

4
)

P
r
e
d
i
c
t
e
d

l
i
f
e

1
8
.
3

y
e
a
r
s

0

9

9
-
-

1
.
0
3

9
-
-

0
.
0
9
9

9
.
.

~
>
1
0
6

9

"
"

0

6

2

3

2
.
0
5
9

3
.
0
8
8

4
.
1
1
8

1
.
1
3

2
.
1
5

3
.
1
8
8

3
0

0
0
0

1
0

0
0
0

6
0
0
0

0
.
0
0
0
2

0
.
0
0
0
2

0
.
0
0
0
5

0
.
0
0
0
9

6
.
7

y
e
a
r
s

O
3

"
1
-

O

Z

"
1
"

m

"
O
b
s
e
r
v
a
t
i
o
n

p
e
r
i
o
d
,

6
4

h
.

h
O
b
s
e
r
v
a
t
i
o
n

p
e
r
i
o
d
,

5
3

h
.

r
-

c

w
l

z

>

7

o
n

C
o
p
y
r
i
g
h
t

b
y

A
S
T
M

I
n
t
'
l

(
a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
)
;

T
h
u

M
a
y

8

1
0
:
4
2
:
1
2

E
D
T

2
0
1
4
D
o
w
n
l
o
a
d
e
d
/
p
r
i
n
t
e
d

b
y
U
n
i
v
e
r
s
i
d
a
d

D
e

C
h
i
l
e

p
u
r
s
u
a
n
t

t
o

L
i
c
e
n
s
e

A
g
r
e
e
m
e
n
t
.

N
o

f
u
r
t
h
e
r

r
e
p
r
o
d
u
c
t
i
o
n
s

a
u
t
h
o
r
i
z
e
d
.

Step 2--Conversion of Thermal to Mechanical Strain
A t emperat ure fluctuation of AT results in a total strain fluctuation of
= ~AT (1)
where ~ is the coeffi ci ent of t hermal expansi on. Obvi ousl y, this represent s an upper-bound
for Ae,, cal cul at ed as i f the component was t ot al l y const rai ned from any t hermal expansi on.
Thus, these cal cul at i ons are conservat i ve and provi de some bui l t -i n engi neeri ng safety fac-
tors. Knowi ng the yi el d stress, ~r , and the modul us of elasticity, E, at the mean operat i ng
t emperat ure, the el ast i c port i on ( of this total strain and then the pl ast i c strain (
component can be cal cul at ed
2xe,q = ~ - at the oper at i ng t emperat ure (2)
A~: p/ = A~: r - A ~ e /
Not e that it is the pl ast i c strain that is most damagi ng for fatigue anal ysi s.
Comb! ni ng Eqs 1 and 2, we get
(3)
0" 3
2xe,, t = s AT - -~- (4)
The out er shel l of the recuperat or is made of the AI SI 304 st ai nl ess steel and the rel evant
mat eri al propert i es are listed in Tabl e 2 [3].
The cal cul at ed t ot al and pl ast i c strain ranges cor r espondi ng to the observed t emperat ure
fluctuations are l i st ed in Table 1.
Topics f or Further Study
Equat i ons 2 t hrough 4 make use of hi gh t emperat ure monot oni c propert i es (yi el d strength
and modul us of el ast i ci t y) to est i mat e the strain component s. How si gni fi cant are the changes
in t hese propert i es due to cycl i ng at hi gh t emper at ur es?
Step 3--Estimation of Fatigue Life
Specific t hermal fatigue dat a for 304 stainless steel that ful l y encompass the operat i ng
condi t i ons of a recuperat or (that is, numerous t i me dependent processes) were not avai l abl e
TABLE 2--High temperature properties of 304 stainless steel (extracted from Ref 3).
Coefficient of thermal expansion,
32 to 1200~ (0 to 650~
Modulus of elasticity
9 1100~ (593~
9 1250OF (677~
Yield strength
9 ll00~ (593~
9 1250OF (677~
10.4 l0 6 i n. / i n. / F ~ (18.72 10 6 cm/ cm/ C o)
23.5 x 106 psi (162 GPa)
21.5 106 psi (148 GPa)
23 000 psi (159 MPa)
10 000 psi (69 MPa)

BHAT ON THERMAL FATIGUE ANALYSIS 107
in the literature. In vi ew of proj ect t i me const rai nt s and t he need for maki ng an engi neeri ng
assessment rather than a ri gorous anal ysi s, appr oxi mat e anal ysi s using exi st i ng dat a was
consi der ed adequat e. The approach t aken consists of est i mat i ng t hermal fatigue life from
i sot hermal low cycl e fatigue dat a for the mat eri al [2,4,5].
Topics for Further Study
For the purpose of correl at i ng t hermal cycl i c life to i sot hermal fat i gue life, should the
i sot hermal tests be carri ed out at the average t emperat ure of t hermal excursi ons or at anot her
effect i ve t emper at ur e? Hi ghl i ght the r ol e of t emperat ure r egi me and possi bl e met al l urgi cal
changes in the mat eri al that can influence the choi ce of t est t emperat ure.
Fat i gue lives cor r espondi ng to the pl ast i c strain component s cal cul at ed in St ep 2 were
est i mat ed from t he publ i shed st rai n-l i fe dat a at high t emper at ur es [6]. Whi l e frequency, hol d
t i me, wave shape, and envi ronment al l are known to have significant influence on fatigue
l i fe at high t emperat ures, in this case study, life was est i mat ed based on t he results publ i shed
by Kuwabar a and Ni t t a [6] at 600~ 0.2 Hz frequency, and no hol d peri ods. In ot her words,
t he t hermal fat i gue l i fe over the t emperat ure range T~ and T 2, correspondi ng to the upper
and l ower l i mi t s of t emperat ure fluctuations, is est i mat ed from the low cycl e fatigue dat a at
600~ (1112~ for the pl ast i c strain range est i mat ed from t hese t emperat ure fluctuations.
These results at vari ous strain levels are l i st ed as the number of cycl es to failure in Table 1.
In vi ew of a number of appr oxi mat i ons i nvol ved, t hese val ues are onl y t rend i ndi cat ors and
caut i on shoul d be exerci sed in usi ng t hese numbers for any ot her purpose. It is cl ear that as
t he magni t ude of t emperat ure fluctuations i ncreases, the fat i gue life decreases rat her rapidly.
Thus, the recuperat or can wi t hst and t emper at ur e fluctuations of 99F ~ (55C ~ al most forever,
but for 396F ~ fluctuations life is onl y about 6000 cycl es.
Step 4--Cumulative Damage
It is cl ear from the above di scussi on that the r ecuper at or is subject to t hermal cycl es of
di fferent magni t udes and, therefore, one must consi der the cumul at i ve effects of these events
in est i mat i ng the final failure life. In the present cal cul at i ons, use is made of the l i near
damage equation due to Mi ner [7]. Thus, i f N~ is the number of cycl es to fai l ure correspond-
i ng to a t emperat ure fluctuation of magni t ude ATe, and i f n~ is the actual number of cycl es
spent, then the damage due to t hese cycl es is
F/1
D, = - - (5)
NI
The fai l ure is supposed to occur when the sum of all t he damagi ng events equal s 1.
Hi
E ~ = 1 (6)
lv,
Table l lists the cal cul at ed damage duri ng the obser vat i on per i od due to each group of
t emper at ur e fluctuations and the t ot al damage for t he peri od. The resul t s are present ed for
bot h the 1100~ (593~ and 1250~ (677~ operat i ons. Convert i ng t hese into t i me-scal e
results in an est i mat ed l i fe of appr oxi mat el y 18.3 years at 1100~ (593~ and 6.7 years at
1250~ (677~ Whi l e a number of factors cont ri but e to the overal l economi cs of operat i ng
t he recuperat ors, l ong lives, in excess of 10 years, are favored in commer ci al usage. In fact,

t ypi cal l i fe is about 15 years. In vi ew of a number of si mpl i f yi ng assumpt i ons made during
the cal cul at i ons, the pr edi ct ed life of 18 year s at 1100~ (593~ operat i on agrees reasonabl y
wel l wi t h what is obser ved in commer ci al pract i ce. Assumi ng that t hese assumpt i ons are
equal l y val i d at the hi gher operat i ng t emper at ur e (1250~ or 677~ t hese cal cul at i ons in-
di cat e t hat the life of the recuperat or may be r educed by as much as 63%. In ot her words,
t hermal fat i gue may short en the r ecuper at or l i fe subst ant i al l y i f operat ed at 1250~ (677~
Conclusions
The above cal cul at i ons i ndi cat e that i ncreasi ng the oper at i ng ski n t emperat ure from 1100
to 1250~ (593 to 677~ may reduce t he recuperat or life due to t hermal fatigue by as much
as 63%. Thi s drop is di r ect l y at t ri but abl e to the 396F ~ (220C ~ fluctuations seen onl y at the
hi gher operat i ng t emperat ure. Further, dat a in Tabl e 1 i ndi cat e t hat i f t hese l arge fluctuations
in t emperat ure can be avoi ded, then t he hi gher operat i ng t emper at ur e resul t s in compar abl e
l i fe and, therefore, can profi t abl y be used for fuel savings. As i ndi cat ed earlier, the recuper-
at or ski n t emperat ure is cont rol l ed by wat er sprays in the wast e gas stream. Thus, the mag-
ni t ude and frequency of ski n t emperat ure fluctuations are a funct i on of how wel l the wat er
sprays per f or m and can be cont rol l ed duri ng the mi l l operat i on. Therefore, it was r ecom-
mended to the Energy Technol ogy Team to l ook into possi bl e modi fi cat i ons to achieve a
hi gher degree of t emper at ur e regul at i on. However, pract i cal consi derat i ons such as the un-
pr edi ct abi l i t y of furnace cut -backs due to mi l l del ays, and t he transients i mposed during
l i ght -ups and shutdowns, the opt i on of oper at i ng recuperat ors at a hi gher t emperat ure for the
pur pose of fuel savings was consi dered unat t ract i ve gi ven the pot ent i al damage due to ther-
mal fatigue.
Acknowledgments
The per mi ssi on of the Inl and St eel Company to publ i sh this paper is appreci at ed.
The cooperat i on of the Ener gy Technol ogy Team during the course of this wor k and hel p-
ful comment s of B. S. Levy duri ng the preparat i on of t he manuscri pt are grat eful l y
acknowl edged.
References
[1] Bugyis, E. J. and Taylor, H. L., "Heat Exchange in Steel Industry," Industrial Heat Exchangers,
A. J. Hays, W. W. Liang, S. L. Richlen, and E. S. Tabb, Eds., American Society for Metals, Metals
Park, OH, 1986, pp. 3-12.
[2] Manson, S. S., Thermal Stress and Low Cycle Fatigue, McGraw-Hill Book Company, New York,
1966.
[3] American Society for Metals, Source Book on Industrial Alloys and Engineering Data, American
Society for Metals, Metals Park, OH, 1978.
[4] Bhat, S. P., High Temperature Cyclic Deformation in Nickel, TD-Nickel and A1-4% Cu Alloy Aged
to Contain O' and 0"," Ph.D. Thesis, University of Pennsylvania, Philadelphia, 1978.
[5] Taira, S., "Relationship Between Thermal Fatigue and Low Cycle Fatigue at Elevated Temperatures,"
Fatigue at Elevated Temperatures, ASTM STP 520, A. E. Carden, A. J. McEvilly, and C. H. Wells,
[6] Kuwabara, K. and Nitta, A., "Thermal-Mechanical Low-Cycle Fatigue Under Creep-Fatigue Inter-
action on Type 304 Stainless Steels," Fatigue of Engineering Materials and Structures, Vol. 2, 1979,
pp. 293-304.
[7] Fuchs, H. O. and Stephens, R. I., Metal Fatigue in Engineering, John Wiley & Sons, New York,
1980.

Richard C. Ri cd
Shell and Detail Fracture Formation in
Railroad Rails
REFERENCE: Rice, R. C., "Shel l and Detail Fracture Formati on in Railroad Rails," Case
Studies for Fatigue Education, ASTM STP 1250, Ralph I. Stephens, Ed, American Society for
ABSTRACT: Detail fractures have been observed in railroad rails for many years and some
serious railroad accidents have been attributed to these insidious rail defects [1]. In a study
completed in the early 1980s for the Department of Transportation [2], it was found that nearly
10% of over 10 000 1.6-km segments of revenue-service railroad rail contained one or more
detail fracture defects. Although in the aggregate, detail fractures are relatively uncommon.
(They comprised only about 3V2% of the 30 000 defects observed in the study cited in Ref 2.)
These defects are of great concern in terms of rail reliability and safety since detail fractures
typically initiate and grow within the head of the rail, and they may go unidentified until the
strength of the rail is so deteriorated that a sudden failure occurs, which can lead to a train
derailment.
Over the past decade, Battelle has worked closely with the Volpe National Transportation
Systems Center (VNTSC) and the Federal Railroad Administration (FRA) on the problem of
shell and detail fracture formation in railroad rails [3-7]. This problem remains a significant
one for the rail transportation and freight industry, in spite of the fact that a considerable body
of knowledge now exists regarding the cause of these defects.
This paper was prepared in the form of a case study for fatigue education to give students
interested in the discipline an opportunity to test their growing expertise on this real-world,
present-day engineering issue.
This paper is presented in three parts. The first part of the paper is informational. It provides
a brief description of shell and detail fractures in railroad rails, and offers a variety of examples
of both types of defects. Other potentially important information for the case study is also
provided, including:
(1) typical three-dimensional residual stress data on used rails,
(2) wheel/rail loading spectra,
(3) thermal stresses in rail,
(4) approximate "live" stresses induced in rails by passage of railroad wheels, and
(5) crack initiation and fatigue crack growth properties of rail steel.
The second part of the paper reviews some of the fundamental questions that the railroad
community has faced regarding shell and detail fractures in rails. The student is asked to take
the role of an independent researcher, perhaps working for the railroad community. The task
is to assess the available information and provide informed responses to each of these questions
(and perhaps others that arise in class discussions).
The third part of the paper provides a discussion on each of the questions posed in Part 2.
An overall interpretation of shell and detail fracture formation processes is also offered, along
with some comments on steps taken by the railroad industry to minimize their occurrence and
reduce the risk that those that do occur will lead to rail failures.
For use as a case study for fatigue education, it is recommended that the three parts of this
paper be given to students sequentially over a period of time sufficient to allow an independent
Manager, Batelle, Structural Integrity Projects Office, 505 King Avenue, Columbus, OH 43201.
109

evaluation of related literature. In this way each response in Part 3 can be used more as an
example of one plausible solution to the problem, rather than the only solution.
KEYWORDS: railroad rails, fatigue analysis, shelling, detail fractures, transverse defects,
residual stresses, case study, failure analysis
Thi s paper was prepared for present at i on at the ASTM Sympos i um on Case St udi es for
Fat i gue Educat i on, hel d in At l ant a, GA on 18 May 1993. There were several pri me goal s
of the symposi um. One goal was to st i mul at e interest in st udent s of fatigue and fracture
mechani cs t hrough exami nat i on of non-t ri vi al , real -worl d case studies. Anot her goal was to
faci l i t at e st udent ' s i nt egrat i on of the el ement s of engi neeri ng that commonl y ent er into a
fatigue desi gn or failure anal ysi s, i ncl udi ng mechani cs of mat eri al s, desi gn of experi ment s,
field dat a acqui si t i on, physi cal metallurgy, fract ography, mat eri al charact eri zat i on, damage
model i ng, and life predi ct i on. St i l l another goal was to dramat i ze the poi nt that most fatigue
desi gn pr obl ems do not have one unique solution, in fact the " bes t " sol ut i on is often di ct at ed
by financial constraints, t i me l i mi t at i ons, avai l abi l i t y of pert i nent mat eri al and processi ng
i nformat i on, l i abi l i t y concerns, and publ i c percept i on.
Part l - - Background
Shells attd Detail Fractures
For those unfami l i ar wi t h fat i gue cracks in rai l road rails, the t er mi nol ogy requires some
expl anat i on. A shell crack is a fatigue crack that forms under the runni ng surface of a rail
and paral l el to that surface. Shel l s in rai l road rai l s may be surface connect ed, but oft en are
not, somet i mes growi ng to a consi derabl e length i f undet ect ed by nondest ruct i ve means. I f
a shell does devel op near the surface of the rai l or gets sufficiently long, it may cause spal l i ng
of the rail, but these shells, or head checks as they are somet i mes cal l ed, do not nor mal l y
cause total rai l failure.
For over 40 years, cracks that devel op from a shell and pr opagat e appr oxi mat el y perpen-
di cul ar to the runni ng surface t hrough the head of a rai l have been known as detail fractures
[8]. More recently, the Sperry Rai l Defect Manual [9] has defi ned a det ai l fracture from
shel l i ng as " a progressi ve fracture starting from a l ongi t udi nal separat i on cl ose to the runni ng
surface of the rai l head, then t urni ng downwar d to form a t ransverse separat i on subst ant i al l y
at ri ght angl es to the runni ng surface. " I f a crack forms away from a shell and grows
per pendi cul ar to the runni ng surface it is known as a t ransverse defect. True t ransverse
defect s are less common t oday than detail fractures, because moder n rai l processi ng tech-
niques have l argel y el i mi nat ed the bl ow hol es that were common t ransverse crack starters.
A dramat i c exampl e of a det ai l fracture is shown in Fig. 1. Thi s det ai l fracture, l abel ed
AT-5, shows the t ypi cal l y obser ved oval shape that expands wi t h i ncreasi ng rail traffic across
the head of the rail, whi l e r emai ni ng cont ai ned wi t hi n the head of the rail until qui t e large.
Si nce detail fractures can grow through over 50% of the head area without bei ng vi si bl e to
the naked eye, they can l ead to cat ast rophi c rai l failures. In fairness, it should be acknowl -
edged that the vast maj ori t y of det ai l fractures are i dent i fi ed t hrough the peri odi c use of rol l -
by nondest ruct i ve i nspect i on devi ces. The det ai l fractures that escape det ect i on are the ones
that somet i mes cause seri ous rai l failures. (Rail AT-5 di d fai l duri ng testing at the Trans-
port at i on Test Center (TTC), 2 but restraint bars pl aced l oosel y on ei t her side of the rail
2 The Transportation Test Center is located in Pueblo, CO and is managed by the Association of
American Railroads (AAR), which is headquartered in Chicago, IL.

RICE ON SHELL AND DETAIL FRACTURE FORMATION 111
FIG. l - - Det ai l f ract ure in rail AT-5.
spanning the section containing the detail fracture prevented derailment of the test train,
while allowing the natural evolution of the detail fracture.)
The AT-5 detail fracture (Fig. 1) shows two distinct regions of crack growth. The nearly
featureless portion of the fracture developed while the rail was in revenue service. The
uneven portion of the flaw developed while the rail was installed in the test track at TTC
and was subjected to controlled freight traffic. The growth of the detail fracture was moni-
tored as a function of tonnage carried. Periodic changes in the direction of traffic on the test
track contributed to the slight changes in the plane of the detail fracture as it grew, visible
in Fig. 1 as alternating light and dark bands emanating from the original detail fracture.
Examples of other detail fractures are shown in Figs. 2 through 5. Although each detail
fracture has unique features, there are obvious similarities between them. All of the transverse
cracks emanated from a shell crack beneath the gage corner of the rail head, and they all
propagated downward and across the head of the rail, largely contained by the gage face
and the running surface of the rail. The cracking pattern displayed in Fig. 2 is similar to
that shown earlier for AT-5 (Fig. 1). That is because this rail experienced a similar service
history to AT-5; it also was removed from revenue (normal) service with a known detail
fracture present, and then placed in controlled service testing at TTC in Pueblo, CO. How-
ever, this rail did not fail during testing at TTC; it was removed before failure and broken
open at a later date for detailed investigation.
Most of the shell cracks display an elongated irregularity in their center, which appears
to be their origin. Detailed examinations of these sites have often revealed complex inclusions
or stringers. An indication of such a complex inclusion containing aluminum, sulfur, silicon,
and manganese is provided in Fig. 6, where an energy dispersive X-ray (EDAX) trace is
shown of the shell initiation site in Rail SP-2.

FIG. 2--Detail and shell fracture in rail NW-4.
The shapes of the shell cracks display less consistency than observed on the transverse
cracks. Some of the shells are relatively short and essentially circular. Others are 2 to 3
times as long as they are wide (lateral dimension); however, in none of these cases does a
shell crack exceed a width of approximately 25 ram, and most are less than 20-ram in width.
These trends are fairly typical of other shells observed in defective rails removed from service
[3-7]. In some extreme cases shells have been found that are over 0.30-m in length, and
some of these very long shells do not have transverse cracks associated with them.
Typical Rail Residual Stress Patterns
Because railroad wheels produce very high Hertzian contact stresses in a rail, the running
surface deforms plastically and typically forms a work-hardened, compressive residual stress
layer at the running surface. Similarly, periodic impact loads by the wheel flange on the
gage face of the rail can produce a compressive residual stress layer at the gage face. Since
it was believed that these residual stresses played some role in the formation of shells, detail
fractures, and other types of rail defects, the Federal Railroad Administration (FRA) and the
Volpe National Transportation Systems Center (VNTSC) supported research to identify them
accurately and to determine whether the magnitude of these residual stresses varied with
increasing tonnage (a measure of rail traffic) and increasing wheel loads. A detailed, exper-
imental stress analysis technique developed at Battelle and documented by Groom [10] in
1980 was shown to provide reasonably precise and accurate three-dimensional residual stress
data for railroad rails. The repeatability of results developed with this intricate rail cutting
and finite element analysis (FEA) procedure was found to be within approximately 20 MPa.
The accuracy of the technique was found to depend on the severity of the stress gradients,
with degradation in accuracy at regions where the gradients are largest, such as at the gage
face of a rail.
Fairly consistent patterns of residual stress do develop in railroad rails as a result of
repeated wheel loads and subsequent running surface and gage face plastic deformation.

FIG. 3--Det ai l fracture in rail AT-IO. (a) View of transverse crack. (b) View of shell crack.
They also appear to develop and stabilize fairly quickly in the life of a rail. For practical
purposes, "shakedown" of the residual stress field does seem to occur [10]. For example,
the longitudinal residual stress patterns shown in Fig. 7 are typical for tangent (straight)
track. Very large compressive residual stresses are normally seen at the gage corner, and
significant compressive residual stresses are usually observed on the gage face and running
surface. These compressive residual stresses produce significant tensile residual stresses in
the interior of the rail, and they typically display maximum values in the upper center of

FIG. 4--Detail attd shell fracture in rail NW-22.
the rail toward the gage corner. (The small pocket of compressive residual stress near the
lower center of the rail head is not usually seen. Residual stresses in this region of the head
can effectively be ignored.)
Even more significant in-plane residual stresses have also been observed in all used rails.
The pattern of maximum in-plane principal stresses found in a typical rail is shown in Fig.
8. As shown in this figure, the gage face and head are usually in complete compression,
while the interior of the rail shows very high maximum principal stresses. The direction of
the maximum in-plane principal stresses does vary significantly through the cross-section of
the rail head, but there is a tendency for the maximum principal stresses to be oriented on
a diagonal downward from the gage corner toward the center of the rail, where the principal
stress vectors tend to flatten out.
Wheel~Rail Loading Spectra
Although the average static wheel rail load experienced in revenue freight and passenger
service is approximately 65 kN, dynamic wheel / rai l loads due to rail joints, rail corrugations,
wheel flats, and other wheel / rai l interface anomalies, can occasionally exceed 300 kN. As
a result, the statistical characterization of wheel / rai l loads has been a matter of considerable
interest for over 15 years [11]. An illustration of the widely varying loads that are experienced
in revenue service is provided in Fig. 9 [12].
The actual severity of a particular type of service can depend more on the magnitude of
the occasional high loads than on the average wheel / rai l loads. For example, it was found
that the cars used for simulated service testing at the Facility for Accelerated Service Testing
(FAST) in Pueblo, CO carried average wheel loads about twice those experienced in Amtrak
and Conrail service, yet the 1 in 10000 wheel occurrence (0.01% exceedance) for FAST

FIG. 5--Det ai l fracture in rail SP-2. (a) View of transverse crack. (b) View of shell crack.
was onl y about half that for Amt rak and Conrai l (Fig. 9). 3 This result is directly attributable
to the very high standards of wheel and rail mai nt enance at FAST, where j oi nt mismatches,
wheel flats, and rail corrugations are essentially nonexistent, compared with field service
condi t i ons where wheel and rail abnormalities, or both develop and can remain for a period
of t i me before they are addressed.
3 These results may not apply to other rail lines operating at other times. These wheel/rail exceedance
curves are offered only as representative data for conditions prevalent in the early 1980s.

FIG. 6--Energy dispersive X-ray analysis of shell initiation in rail SP-2.
Occasional, very large dynamic wheel / rai l loads may accelerate the crack growth rate of
a detail fracture, although they are commonly more significant in limiting the maximum flaw
size that can be withstood by the rail before failure. The static strength of a rail decreases
(on average) with increasing flaw size of a detail fracture. This trend is shown (albeit with
considerable variability) in Table 1, from Orringer et al. [13], who developed residual
strength data from about 30 used and defective rails. These data were developed from a
Hori zont al Di mensi on, X, mm
-30 -20 -10 0 10
E
>" 30 o6
g -
E
>
10
-1,50 -1,25 -1.00 -0.75 -0.50 -0.25 0 0.25 0, 50
Hori zont al Di mensi on, X, in.
FIG. 7--Longitudinal residual stress pattern in a used railroad rail (1 ksi = 6.895 MPa).

RICE ON SHELL AND DETAIL FRACTURE FORMATION 1 1 7
Hor i z ont al Di mens i on, X, mm
-30 -20 -10 0 10
, 7 O l O . o e ,
c o r n e r / .
"_ ~ 2 5 >.-
~5 t5
"m 20 ,~
.O 0. 75 -5 K . / ~ - -
>
0. 50
i ; ~ 1 - 10
0.25
- 1, 50 -1. 25 - 1. 00 -0.75 -0. 50 -0.25 0 0. 25 0, 50
Hor i z ont al Di mens i on, X, in,
FI G. 8--hi-plane, maximum principal residual stress pattern in a used railroad rail (1 ksi ~. 6.895
MPa).
Load. kN
0 100 200 300 400
0 10 20 3o 40 So 60 70 80 90
V e r t i c a l Wh e e l L o a d , k i p
FIG. 9--Statistical characterization of wheel~rail loads on concrete ties.

TABLE 1- - Four - poi nt bending static strength results ,for
defective rails.
Apparent Flaw Size, Breaking Force,
Rail ID % of Head Area (kN)
132 RE RAIL
NW5 3 547
NW18 l 0 391
NW12 18 436
NW21 18 378
NW 15 20 480
NW9 22 471
NWI 6 22 658
NWI 0 25 383
NWI 4 25 338
NW 17 25 369
NW25 26 240
NW8 28 383
NW 1 34 400
NW19 35 356
NW13 36 365
NW7 37 431
NW4 46 133
NWI 1 54 338
NW20 58 205
NW22 89 102
136 RE RAIL
AT6 13 356
AT4 20 347
SP~8 24 365
SP2 28 325
SP31 44 285
SP34 44 258
SP1 46 231
SP6 53 311
SP30 57 133
SP15 59 316
SP27 61 231
SP12 63 231
SPI 9 69 236
SP7 78 276
SP28 87 187
f our - poi nt l oa di ng t es t f i xt ur e wi t h an out s i de s pan of 1, 52 m a nd an i nt e r i or s pan b e t we e n
l oa d a ppl i c a t i on poi nt s of 0. 2 m. Ea c h r ai l was l oa de d he a ddown. The e qui va l e nt r ai l b e n d i n g
mo me n t was f o u n d t o be a p p r o x i ma t e l y M(N.m) = - 1 . 4 7 P , whe r e P is t he appl i ed l oad.
The det ai l f r act ur e di me ns i ons f or di f f er ent r ai l s wi t h s i mi l a r si ze f l aws wer e s uf f i ci ent l y
cons i s t ent , so t ha t a ve r a ge f l aw s ha pe st at i st i cs c oul d b e t a bul a t e d as a f unc t i on of f l aw si ze
a nd e xpr e s s e d i n t e r ms of pe r c e nt a ge o f he a d ar ea, as s h o wn i n Tabl e 2. Th e c ons i s t e nc y i n
t he d i me n s i o n a l pa r a me t e r s of e a c h si ze r a nge o f det ai l f r act ur es i s appar ent , wi t h t ypi cal
coef f i ci ent s of va r i a t i on i n t he r e a di ngs ( bas ed on 5 or 6 r ai l s wi t hi n e a c h si ze r ange) of 5
t o 10%.

TABLE 2--Average dimensions and location of detail fractures in railroad rail [12].
Average Dimension, mm (Standard Deviation, mm)
Depth of Center
Average Upper, Lower, of Ellipse Distance of Center
Flaw Size, Minor Axis Minor Axis Major Ax i s Beneath Running of Ellipse
(% of Head Area) of Ellipse of Ellipse of Ellipse Surface from Gage Face
12.5 9.1 (1.0) 9.7 (1.3) 13.2 (1.3) 21.8 (1.8) 19.8 (1.0)
17.5 10.7 (1.5) 13.0 (0.8) 15.7 (1.3) 19.6 (1.8) 22.4 (0.8)
22.5 9.9 (0.8) 14.7 (1.0) 17.3 (1.5) 17.5 (3.0) 22.4 (0.5)
27.5 11.7 (1.0) 16.3 (0.8) 19.3 (0.8) 16.8 (1.5) 24.1 (0.5)
33.5 12.2 (1.0) 17.8 (2.0) 21.3 (0.8) 14.0 (2.0) 24.9 (1.3)
41.0 12.7 (1.5) 19.8 (1.0) 23.6 (1.8) 12.2 (2.5) 25.7 (1.3)
Thermal Stresses in Rail
Muc h of t oday' s rai l is cont i nuous l y wel ded rai l ( CWR) , made up of sect i ons of rai l
wel ded t oget her to f or m l ong, cont i nuous sect i ons of track. Thi s is done to avoi d the com-
monl y obser ved adver se effects of rai l j oi nt s, for exampl e, j oi nt fai l ures and wheel / r ai l
i mpact l oads f r om hei ght mi s mat ch at a j oi nt . However, t hese l ong cont i nuous sect i ons of
rai l can devel op subst ant i al t ensi l e or compr essi ve t her mal stresses. These t her mal stresses
devel op as a r esul t of a t emper at ur e di fferent i al bet ween t he so-cal l ed rai l neut r al t emper at ur e
(at whi ch t he l ongi t udi nal t hermal stress is zero) and t he cur r ent rail t emper at ur e. Ini t i al l y,
t he rai l neut ral t emper at ur e is appr oxi mat el y the s ame as t he l ayi ng t emper at ur e for t he rail,
but over t i me t he neut r al t emper at ur e can shift up or down rel at i ve to t he l ayi ng t emperat ure
because of shifts i n t he bal l ast beneat h t he ties and ot her di st ur bances to t he or i gi nal t rack
condi t i on. Temper at ur es above t he neut r al t emper at ur e l ead to l ongi t udi nal compr essi ve
st resses i n t he rai l whi l e t emper at ur es bel ow the neut r al t emper at ur e l ead to l ongi t udi nal
t ensi l e stresses i n t he rail. Thi s is an especi al l y seri ous concer n i n l ong sect i ons of st rai ght
( t angent ) track, al t hough cur ved CWR can also suffer f r om gage- wi deni ng or l at eral shi ft
due to t her mal stresses.
Typical Wheel-Induced Rail Stresses
The nat ur e and severi t y of wheel - i nduced rail st resses var y greatly, dependi ng on a vari et y
of rai l and servi ce paramet ers. The speci fi c case cons i der ed here is that of a 133 kN wheel
force appl i ed to a 67. 6 k g / m rail (Fig. 10) [14]. The l oad appl i cat i on poi nt was chosen to
be appr oxi mat el y 20 mm off-center, t owar d the gage side, to r epr esent a fai rl y t ypi cal l oadi ng
poi nt (based on c o mmo n l y obser ved head wear pat t erns). The predi ct ed l ongi t udi nal stresses
are shown (Fig. 11) for a rai l cr oss- sect i on i mmedi at el y under the l oad poi nt . Not e that t he
mi n i mu m stress magni t udes are shown, si nce t he head is al mos t compl et el y i n compr essi on.
Of course, t he great est compr essi on stresses are pr edi ct ed near the r unni ng surface. The
st resses decrease i n absol ut e magni t ude fart her down i n t he head of t he rai l and fart her
t owar d the field side of t he rail. The ma xi mum i n- pl ane compr essi ve- st r ess pat t ern is even
mor e l ocal i zed ar ound the poi nt of cont act (Fig. 12). The di rect i on of ma xi mum i n- pl ane
compr es s i on st resses is essent i al l y vert i cal di r ect l y under t he l oad poi nt and sl ant ed away
f r om the l oad poi nt t oward t he gage face and the cent er of t he rail. ( The static and act i ve
st resses that are pr oduced wi t hi n a rail are ful l y t hr ee- di mens i onal and very compl ex to
model . For pur poses of this di scussi on, onl y the axi al component s of stress are consi der ed. )

1 2 0 CASE STUDIES FOR FATIGUE EDUCATION
I
I
I
= i
T
I
Ii
I i
I
I
I
I
_1 NEUTRAL AXIS
~ ,
I
I
I
=1
+l
177
2, 6.
I
I +- - - I . 4 0 ~ - d I
I _
t l
- - ~ 8R
_ . t - r R . a 8 " R .
l -
~/~
I +
~ - 6 " - - " t
t
l
_1
~ - I 9
t O
n
I
I
I
A Ye~
S q I n Per cent
H e a d . . . . . . . . . . . . . . . . . . 4 . 8 6 36. 4
W e b . . . . . . . . . . . . . . . . . . . 3. 62 27. 1
B a s e . . . . . . . . . . . . . . . . . . 4. 87 36. 5
T o t a l . . . . . . . . . . . . 13. 35 100. 0
Mo me n t o f i n e r t i a . . . . . . . . . . . . . . . 9 4 . 9
S e c t i o n mo d u l u s , h e a d . . . . . . . . . . . 23. 9
S e c t i o n mo d u l u s , b a s e . . . . . . . . . . . . 28. 3
R a t i o mJ . t o a r e a . . . . . . . . . . . . . . . 7. 11
R a t i o s . m. h e a d t o a r e a . . . . . . . . . . 1. 79
R a t i o h e i g h t t o b a s e . . . . . . . . . . . . . 1. 22
C a l c u l a t e d we i g h t , l b p e r y d . . . . . 136. 2
F i g . 6 - - 1 3 6 R E r a i l s e c t i o n .
References, Vo| . 63, 1962, PP. 498, ?68.
( 1 9 7 1 )
F I G. l O- - S t a n d a r d cross- sect i onal di mens i ons f o r a 136 RE rai l (1 in. = 25. 4 ram).

Ho r i z o n t a l Di me n s i o n , X, mm
-30 -20 -10 0 10
1.75
; a o e / ~ \ \ -aorol f / I I f . I
-20 t o -25 ksi
"-~ 1.25 > :
>" 30 ~"
~1 10 to 15 ksi
> -5 re - 10 ksi
0. 50
10
0. 25
*1. 50 - 1. 25 -1 .OO - 0. 75 - 0. 50 -0.25 0 0. 25 0. 50
Ho r i z o n t a l Di me n s i o n , X, i n.
F I G . l l --Predi ct ed longitudinal stress pattern due to a 133 kN off-center wheel load (1 ksi = 6.895
MPa).
Ho r i z o n t a l Di me n s i o n , X, mm
- 3 0 - 2 0 - 1 0 0 1 0
I I I I I
. , - 40 ksi
- 20 ks~ \v, / -30 ksi
~ageer 0 t o 5 ksi
4O
1, 50
E
>: ~.25 >:
0 . 7 5 2 0 . ~
>
0. 50
10
0. 25
- 1. 50 - 1. 25 - 1. 00 -O.75 - 0. 50 -O.25 0 0. 25 0. 50
Ho r i z o n t a l Di me n s i o n , X, i n.
FIG. 12--Predicted i n-pl ane, max i mum pr i nci pal st ress pat t ern due to a 133- kN of f - cent er wheel l oad
(1 ksi = 6. 895 MPa).

122 C A S E S T U D I E S FOR FATI GUE E D U C A T I O N
Rail Crack Growth and Fatigue Properties
In a maj or st udy of the fatigue crack pr opagat i on propert i es of st andard carbon-st eel rai l
(St andard Speci fi cat i on for Car bon- St eel Rai l s, ASTM A1- 68a) [15] undert aken in the mi d-
1970s, dat a were devel oped for a range of stress rat i os and t est t emperat ures, as wel l as for
each ort hogonal cracki ng pl ane in the rai l head [15]. Thr eshol d experi ment s were also done
at three stress rat i os (R = - 1 . 0 , 0.0, and 0.50) to devel op est i mat es of t hreshol d stress
i nt ensi t y levels, bel ow whi ch crack-growt h rates woul d asympt ot i cal l y appr oach zero [16].
Wi t h this col l ect i on of dat a it was possi bl e to devel op a rel at i vel y general crack growt h
model for average crack growth trends rangi ng from near t hreshol d t hrough fracture insta-
bi l i t y for a range of di fferent stress ratios [17]. The equat i on is stated as fol l ows
Kax
d N d -a -a = C ( 1 - R ) 2 (K2rnax - - K~th) K c 7 ~ r n a x ( 1 )
where Km,x is the maxi mum crack t i p stress i nt ensi t y at any poi nt in t i me, and K c and K,h
are est i mat es of the pl ane stress fract ure t oughness and t hreshol d stress i nt ensi t y factor as
shown in Fi g. 13. I f t he stress cycl e i nvol ves a stress rat i o l ess than zero, the value of R in
Eq 1 was t aken to be zero (that is, the compr essi ve part of the cycl e was not consi dered to
influence the crack growt h rate). The average dat a trends compar ed to the crack growt h
predi ct i ons are shown in Fig. 13 for L / T ori ent at i on cracks (l ongi t udi nal l oadi ng on a trans-
verse crack wi t hi n the rai l head).
The fatigue crack i ni t i at i on propert i es of rai l steels were al so charact eri zed in a later st udy
[18]. These results are present ed (Fi g. 14) in terms of equi val ent strain versus cycl es to crack
i ni t i at i on, where the equi val ent strain par amet er was defi ned as
Ee q ~_ (AE)m (O.max), m ( 2 )
O
- - ~ Trend
10- 2 ~ O O / k P r e d i c t i o n
K c = 6 0 . 4 M P a ~ / ~ [3 ~ .
" 8 - - [ 3
1 0 - ` = _ [ 3
~o I G s
1 - 7 . . . . . Y
1 0 - s ~ I
J ~ I I I I I I I l l
8 9 I 0 2 0 so 4 0 5 0 6 0 eo 8 0 9 Ol 0 0 2 0 0
Stress I n t e n s i f y , A K , M P o v f m "
FIG. I 3 - - L/ T orientation fatigue-crack growth properties of rail steel, predicted versus actual.

-230
-23~
- 2 4
- 2 4
J
- 2 5 0
- 2 5 5
- 2 , 6 0
- 2 6 5
S t r e s s R a h o S p e c i m e n L o c x l t l o ~
0 - I 0 C e n t e r ( d a t a p o i n t o p e n ) 0
~ 1 0 0 I 0 S u r f a c e (k.=ft i ~ l l f o f g a t e p o i n t s o l i d ) qD
~ 0 5 0 I n t e r m e d i a t e ( r i g h t h a l f o f d a t a p o i n t s o l i d )
N o t e E a c h p o i n t r e p r e s e n t s a n a v e r a g e o f ( ~ } m ( s / E I - m
f r o m 2 t o 7 c r a c k i n i t i a t i o n v a l u e s r = n ~ x )
1 m - O O
5 x I 0 6 c y c l e s
- - < ~ ~ / C o n s t o n t a m p l i t u d e
2 ~ 0 [ ~ (cPers~enntt ORr~P' ~ tuutd e )
. . . . . . . . . . . . . . . . . \ i I " " ~ I
\ \ I I " I R=O~O
I I
i 0 4 r o 9 10 ~ i o e
C y c l e s t o I n i t i a t i o n
FI G. 14- - Crack initiation f at i gue properties af rail steel as a f unct i on o f equivalent strain.
with an optimum m value of approximately 0.6. An approximate relationship between cy-
clically stable stresses and strains was also developed as follows
2S, for At less than Axe(l)
Axe = - - E - ' (3)
( s . ) ' " ' ,
Ae = 2 \ K) / , for > Ae(1), -< Ae(2)
{ s , , ) ) , , , :
AX~ = 2 \ K~ _ ] , AXe > A~( 2 )
( 4)
( 5 )
where K~ and K2 are strength coefficients, n, and n2 are strain hardening exponents, and
Ae(1) and Ae(2) are specific values of Ae that denote break points in the tri-log-linear stress-
strain curve approximation. The approximate values for the constants in Eqs 3 through 5 are
as follows for the tested rail material
E=20 104MPa,
K( = 2.36 9 1 0 4 MPa,
Ae(1) = 2.7 9 103,
n~ = 0.677,
S,,(I) = 270 MPa,
K, = 5 . 5 5 X 103 MPa,
Ae(2) = 4.0 x 1 0 - 3 ,
n, = 0.443, and
S,,(2) = 354 MPa.
A small number of periodic overstrain, crack initiation fatigue experiments were also

TABLE 3--Periodic overstrain results on new rail material.
Strain Range, % Strain Range
Components, %
Small Lar ge Frequency of Stable Stress Total Cycles
Cycles Cycles Overstrain Elastic Plastic Range, MPa to Failure
0.375 0.800 10 every 1000 0.338 0.037 643 209 240
0.342 0.033 629 217 600
0.375 (I.800 10 every 10 000 0.345 0.030 676 270 200
0.346 0.029 645 289 400
0.300 0.800 10 every 1000 0.289 0.011 546 422 820
0.296 0.004 545 401 270
0.300 0.800 10 every 10 000 0.268 0.032 525 900 700
0.271 0.029 516 930 785
pertbrmed to evaluate the potential influence of occasional high-stress excursions on the
long-life fatigue resistance of rail steel. Those results are presented in Table 3 and shown in
Fig. 15. Compared with the expected cycles to crack initiation based on previously generated
constant amplitude data, the periodic overstrain tests produced fatigue lives 3 to 5 times
shorter than would have been expected if the 0.375% strain range cycles were non-damaging
and produced fatigue lives 10 to 50 times shorter than would have been expected if the
0.300% strain range cycles were non-damaging. Significantly, the predictions of fatigue life
for simulated service, variable amplitude loading conditions were much more accurately
predicted when using the periodic overstrain fatigue curve for the damage summations com-
pared with the constant amplitude fatigue curve [16].
In addition to the information presented here, there are numerous other sources of data
on the fatigue and crack growth properties of rail steels, for example, the work of Fowler
[191 and Park and Fletcher [20].
-2.30
k Frequency of
Overstrain
- 2. 35 ~ O 10 ever y 1000
[ ] 10 ever y 10, 000
-2.40 Overstrain level = 0 . 8 0 %
-2.45 w ~ ~ Const ant amplitude
j ~ e l i n e curve
"J - 2. 50
Const ant a mp l i t u d e l , / ~ \
~ 2 ~ 6 0
i
strain level J\0300% \
2 0 s i I I I \ I \ b J I t i I
10 r 105 106 107
Cycles t o Failure, N i
FIG. 15--Effect of periodic overstrains on long-life fatigue properties of rail steel.

Part 2 - - S t u d e n t Que s t i ons
The fol l owi ng quest i ons are present ed to st i mul at e a careful review of the background
i nformat i on pr ovi ded in Part 1 and to pr omot e di scussi on wi t hi n a cl ass or study group. To
hei ght en the si gni fi cance of t hese quest i ons it may be useful to consi der the posi t i on of
rai l road mai nt enance engi neers who must make deci si ons concer ni ng the i nt egri t y of the rail
l i nes they inspect and mai nt ai n. Or perhaps an even more meani ngful , hypot het i cal situation
to consi der woul d be that of an FRA engi neer within the Office of Safet y or the Office of
Research and Devel opment . Your t ask coul d be to assess the research that has been done
and to eval uat e whet her i mport ant basi c quest i ons are under st ood wel l enough to propose
regul at i ons regardi ng rai l mai nt enance, i nspect i on, and r epl acement that wi l l ensure the safe
operat i on of t oday' s rail t ransport at i on and freight syst ems. In addi t i on, i f represent at i ves
from the Associ at i on of Amer i can Rai l roads ( AAR) quest i on the meri t of t hese proposed
regul at i ons, how wi l l you defend your r ecommendat i ons? Does addi t i onal research need to
be performed, and i f so, speci fi cal l y what i ssues need to be addr essed and how can experi -
ment s be devi sed to bet t er understand the unresol ved i ssues?
What Causes Subsurfiwe Shells?
In order to prevent (or at l east mi ni mi ze) subsurface shells it is necessary to devel op an
underst andi ng of the condi t i ons that l ead to subsurface shells in a rai l road rail. To what
ext ent do factors such as the magni t ude of wheel loads, t ensi l e strength of the mat eri al , and
the r ai l / wheel i nt erface condi t i on influence the devel opment of shel l s? To what ext ent mi ght
t rack condi t i ons such as the t ype of bal l ast and the condi t i on and t ype of ties be expect ed
to influence the severi t y and frequency of shel l format i on?
Whr Do Only Certain Shells Become Detail Fractures?
What are the probabl e key factors that det er mi ne when a shell wi l l turn downward and
become a det ai l fract ure? And why do some shel l s grow to subst ant i al lengths whi l e never
turning downwar d to form det ai l fract ures?
Why Do Most Detail Fractures Grow Subsurface?
Det ai l fractures t ypi cal l y r emai n cont ai ned wi t hi n the head of the rail until t hey are quite
large. As a result, when left undet ect ed, the~, occasi onal l y l ead to cat ast rophi c fai l ure of a
rail. Based on the i nformat i on present ed in Part 1 of this paper and ot her relevant i nformat i on
that can be found, what factors woul d most l i kel y cont ri but e to this phenomenon? What
condi t i ons must exi st before a det ai l fracture does br eak t hrough to a free surface?
How Should Detail Fractures Be Modeled?
How coul d a model be formul at ed to represent the non- sel f - si mi l ar crack growt h of a
det ai l fract ure? The model shoul d accommodat e the growt h of a shell in a pl ane nearl y
paral l el to the runni ng surface and the subsequent t urni ng of the defect downwar d onto a
pl ane appr oxi mat el y per pendi cul ar to the ori gi nal shell. Pr edi ct ed growth rates shoul d be
nearl y const ant for det ai l fractures rangi ng from about 10 to 35% of the head area. Bot h
t hermal and resi dual stresses must be accommodat ed, in addi t i on to the dynami c, wheel-
i nduced rai l stresses,

What Factors Most Influence Rail Head Crack Initiation ?
There are numerous vari abl es that can si gni fi cant l y influence the rate of format i on of
fat i gue cracks, whi ch can subsequent l y l ead to shel l s and det ai l fractures. What factors woul d
be expect ed to most st rongl y i nfl uence rail head fat i gue crack i ni t i at i on rates? Some obvi ous
possi bi l i t i es are:
(1) rail cl eanl i ness,
(2) average axl e load,
(3) traffic severity,
(4) t rack vert i cal modul us,
(5) combi ned vert i cal and l at eral loads,
(6) t hermal stresses,
(7) resi dual stresses,
(8) head hardness, and
(9) degree of offset in wheel l oadi ng.
The first t ask in i dent i fyi ng the fact ors that most influence rai l head fat i gue crack i ni t i at i on
rat es is an assessment of the requi red backgr ound i nformat i on. What must be known to
pr oper l y address t he influence of one or all of t hese fact ors on rai l fat i gue crack i ni t i at i on
resi st ance? Wi t h avai l abl e i nformat i on present ed in Part 1 of this paper and elsewhere, whi ch
i t ems shoul d be chosen as most i mport ant in t erms of t hei r influence on rai l head fat i gue
cr ack i ni t i at i on l i ves? What is the j ust i fi cat i on for t hese sel ect i ons? Assumi ng these sel ect i ons
are reasonabl e, what measures coul d be t aken to mi ni mi ze t hei r effect and t hereby maxi mi ze
rai l head fat i gue crack i ni t i at i on resi st ance?
Part 3--Responses to Student Questions
Formation of Subsurface Shells
Subsurface shells are t ypi cal l y associ at ed wi t h compl ex i ncl usi ons (stringers) ori ent ed
l ongi t udi nal l y in the rai l head at a dept h of 6 to 10 mm beneath the runni ng surface. Re-
gardl ess of the dept h beneat h the runni ng surface, the shell i ni t i at i on process is ai ded by the
presence of compl ex inclusions. In most cases that have been exami ned [6,7] the i ncl usi ons
cont ai ned al umi num and silicon, and coul d gener al l y be cl assi fi ed as " oxi de- t ype" inclu-
sions. Al umi num is added for deoxi dat i on, whi l e cal ci um may arise from interactions wi t h
the slag to form non-met al l i c i ncl usi ons. Because nonmet al l i c (oxi de) part i cl es are hard and
bri t t l e, the l arger i ncl usi ons, especi al l y t hose ori ent ed paral l el to the runni ng surface, serve
as candi dat e shell crack i ni t i at i on sites.
The l oadi ng on t he rai l and its strength are al so factors in the l i kel i hood of shell cr ack
format i on. The frequency of shell cr ack format i on has been found to be di rect l y rel at ed to
the magni t ude of t he stresses generat ed duri ng r a i l / whe e l interaction. Everyt hi ng else bei ng
equal, i ncreases in the t ensi l e strength and fatigue resi st ance of the rai l steel have been found
to decrease the rate of shell cracki ng [21].
However, the bi ggest factor in decreasi ng the frequency of shell crack format i on appears
to be rail cl eanl i ness (whi ch r emoves the maj ori t y of the l i kel y shell cr ack initiation sites).
An i nt erest i ng st udy compl et ed i n Japan [22] on over 100 rails wi t h shell defects showed
that 70% of t he defect i ve rai l s were rai l s t aken from t he t op of a semi - ki l l ed i ngot (whi ch
gener al l y cont ai ns the largest concent rat i on of i ncl usi ons). Other studies in Germany [23]
and the f or mer Sovi et Uni on [24] have al so shown that shell format i on in t he heads of rai l s

RICE ON SHELL AND DETAIL FRACTURE FORMATION 1 2 7
can be r educed by decreasi ng the number and size of nonmet al l i c i ncl usi ons present in the
steel.
Mi cr ohar dness measurement s have al so been conduct ed on several rai l heads cont ai ni ng
shel l defect s to det ermi ne whet her t here was any rel at i onshi p bet ween the gradi ent s in the
hardness of t he rai l head surface mat eri al [3] and the l ocat i on of the shell in that same rail.
These st udi es showed that the shells were t ypi cal l y l ocat ed at t he i nt erface bet ween the
subst ant i al l y col d- wor ked surface l ayer and t he softer mat eri al in the bul k of t he rai l head.
Thi s dept h wi t hi n the head of the rai l al so cor r esponded appr oxi mat el y wi t h the regi on that
t ypi cal l y cont ai ns l arge i n-pl ane, t ensi l e resi dual stresses [6].
Transition of Shells into Transverse Defects
The dept h of a shell may be the most i mpor t ant fact or in det er mi ni ng whet her i t wi l l turn
downwar d and form a det ai l fracture. Shel l s not associ at ed wi t h det ai l fractures are t ypi cal l y
nearer t he runni ng surface. The total area and l engt h of a shell are poor i ndi cat ors of whet her
a shel l wi l l become a det ai l fracture [6].
Shel l s that initiate about 10 mm bel ow t he runni ng surface t end to be most cri t i cal in
t erms of subsequent f or mat i on of det ai l fractures. It is i nt erest i ng to note that this dept h
beneat h t he runni ng surface corresponds appr oxi mat el y with the dept h at whi ch the l ongi -
t udi nal resi dual stresses become tensile, as shown earl i er in Fig. 9. Even t hough the " l i ve"
l ongi t udi nal st resses that resul t from wheel passage at this dept h in the rai l head are com-
pressi ve, the t ensi l e resi dual stresses creat e a condi t i on from whi ch a t ensi l e stress excursi on
at the crack tip can result. In ot her words, when a shell crack branches downward at a dept h
wi t hi n the head of the rai l where it becomes exposed to si gni fi cant t ensi l e resi dual stresses,
t hose t ensi l e resi dual stresses may "l i f t " the ot herwi se compr essi ve stress excursi on into
tension. The t ensi l e stresses can l ead to a si gni fi cant Mode I stress intensity at the shell
crack tip, pr ompt it to turn, and cause it to pr ogr ess downward, essent i al l y per pendi cul ar to
the r ai l ' s runni ng surface. Thi s concept is di scussed in some det ai l in a l at er sect i on of this
paper ent i t l ed, Modelling of Detail Fractures.
A key assumpt i on in this i nt erpret at i on of t he crack growth process is that Mode I cr ack
growt h wi l l t end to pr edomi nat e in the det ai l fracture process. Thi s assumpt i on is based on
evi dence t hat cracks in rai l steel t end to " pr ef er " Mode I crack growth. Earl i er experi ment s
at Bat t el l e desi gned to force cracks in rai l st eel to grow in Mode II or III were unsuccessful
(St andard Speci fi cat i on for Carbon-St eel Rai l s, ASTM A1-68a) showi ng the strong t endency
for fat i gue cracks to propagat e per pendi cul ar to the maxi mum pr i nci pal stress di rect i on. This
appears to be true even in the earl y stages of shel l crack format i on. Evi dence suggest s that
it is onl y because the Mode I stress i nt ensi t y al ong the axis of the rail event ual l y exceeds
the i n- pl ane Mode I stress i nt ensi t y that the cr ack turns al most per pendi cul ar to its ori gi nal
di rect i on and t ransi t i ons from a shell to a t ransverse defect. Si nce the l ongi t udi nal tensile
resi dual stresses t end to be great est near t he cent er of the rai l head, the detail fract ure tends
to pr opagat e most r api dl y in that di rect i on.
It is cert ai nl y possi bl e that the size and ori ent at i on of the i ncl usi ons al so influence whet her
a shel l wi l l turn into a det ai l fracture. Unfort unat el y, no evi dence to support this argument
has been found si nce ext ensi ve mechani cal damage is general l y apparent on the shel l fracture
surfaces (caused by the force of the rai l road wheel s rubbi ng the t wo crack surfaces together).
Nondest ruct i ve t echni ques are needed for the exami nat i on of t hese defect s once a shel l crack
is det ect ed in a rail. Such t echni ques coul d be empl oyed to define the size, shape, and
ori ent at i on of t hese defect s and to t rack the growt h of a shell crack in service, or in si mul at ed
ser vi ce condi t i ons, such as at the Faci l i t y for Accel er at ed Servi ce Testing (FAST) at TTC.

Growth of Detail Fractures in the Head of a Rail
The physical evidence f r om retired, defective rails suggests that a detail fracture typically
remains contained within the head of a rail until it is quite large because of longitudinal
compressive residual stresses that tend to build up along the gage face and the running
surface. Until there is a maj or redistribution of these residual stresses due to propagation of
the detail fracture across the head of the rail, there is a minimal Mode I crack driving force
along the front of the crack near the gage face and running surface. Since the compressive
residual stresses near the running surface tend to be reinforced by continued wheel passages,
equilibrium demands that they be offset by an expandi ng tensile residual stress envelope
surrounding the detail fracture. This argument is supported by the common observation that
detail fractures tend to grow in a non-self-similar manner across the rail head and downward
through it, while there is very little crack growth upward or toward the gage face. Figures
1 and 2 provide examples of this phenomenon.
The angle of the transverse crack, once established, is usually not perfectly vertical, but
is tilted f r om vertical by 0.15 to 0.30 rad. It is believed that this angle is dictated by the
direction of traffic, and that drag forces created by the passage of undriven wheels cause a
shift in the three-dimensional variable stress field sufficient t o tilt the maxi mum axial prin-
cipal stress direction from horizontal. The detail fracture pattern in the AT-5 rail (Fig. 1) is
a dramatic example of this effect.
Modeling of Detail Fractures
There is no generally accepted approach for model i ng the non-self-similar crack growth
of a detail fracture. The issue has been examined in considerable detail by various research-
ers, for example, Orringer et al. [13] and Besuner [25] in the Uni t ed States, and Mair and
Groenhout [26] and Chipperfield [27] in Australia. An alternate approach for modeling the
growth of a detail fracture was also developed by the author in Ref 6 and explained in more
detail in Refs 28 and 29. A bri ef summary of the approach is included here.
The combi nat i on of residual and wheel-induced stresses apparently provide the maj or
driving force for detail fracture growth. I f these stresses are considered together, it is possible
to construct a reasonable pattern of the stresses within the rail that are effective in crack
initiation and growth. Figure 16 shows the pattern of stress ratios (the ratio of mi ni mum to
maxi mum stress in a wheel-induced stress cycle) that results f r om the combination of lon-
gitudinal residual stresses with predicted wheel-induced stresses. This figure shows that the
gage face and running surface withstand stress cycl es completely in compression, while the
interior of the head experiences stress cycles in tension, and even some cycles with tensile
mean stresses. This prediction of cyclic tensile stresses in the longitudinal direction within
the head of the rail is significant because it is generally conceded that fatigue cracks are
most prone to propagate in a cycl i c tensile stress field. Even i f a crack does initiate in a
largely compressive cyclic stress field [30], it is not likely t o propagate. Therefore, it appears
unlikely that transverse cracks woul d initiate near the gage face or running surface and even
i f t hey did they would not propagate, at least not perpendicular to the running surface.
Transverse cracks would have the greatest t endency for initiation and growth in the interior
of the rail where the majority of the stress cycl e is in tension, that is, stress ratios above
- 1. 0. This predicted trend is supported by field data on flawed rails containing transverse
defects and detail fractures.
The model that was developed to describe these cracking patterns is based on an embedded
elliptical flaw that can change in aspect ratio and tilt. Flaw growt h is considered at points

- 3 0 -20 -10 O 10
1.75
> >
10
~ 1 ~ t r 1 1 I .
- , . ~o - , . 5 5 - , . o o - ~ - 0 . 5 ~ - o . 2 ~ o o . ~ 5 o . 5 o
Hori zont al Di mensi on, X, in.
FIG. 16- - Est i mat ed pattern of stress ratios resulting f rom the combination of longitudinal residual
stresses and active stresses induced by a 133-kN wheel load.
on the perimeter of an initial small ellipse. The stresses effective in crack growth are assumed
to result from combi ned residual and wbeel-induced stresses. The crack growth rate is es-
timated based on the local Mode I stress intensity factor. After small increments of growth
have occurred around the ellipse, a new ellipse is established to approximate the expanded
flaw. The process is repeated incrementally and the size and shape of the flaw is monitored
as a function of wheel passages.
As mentioned earlier, the stresses effective for crack growth are assumed to arise from the
combi nat i on of residual and wheel-induced stresses. (Thermal stresses could also be influ-
ential; i f they are believed to be significant thay can be introduced additively throughout the
rail cross-section,) The stress field that is believed to drive transverse crack growth is shown
schematically in Fig. 17, Three component s of longitudinal stress are shown. The first, shown
as a broken line, represents the typical distribution of residual stresses rangi ng from the
running surface to the base of the head that develops approximately 20 mm inward from the
gage face. The second, shown as a dashed line, represents the wheel-induced stresses, which
are entirely compressive from the running surface down t hrough the rail head. The sum-
mation of these two longitudinal stress component s is represented by the third line, whi ch
is dotted. At any given depth, the difference between the first and third stress components
represents the predicted stress excursion. Near the running surface this stress excursion is
entirely compressive. Further beneath the running surface this stress excursion develops a
significant positive component, at least in a used rail that has developed appreciable longi-
tudinal tensile residual stresses. This predicted stress excursion is shown as a continuous
sine wave (Fig. 17), whi ch develops 3.14 rad out-of-phase with the applied wheel load.
To begin the crack growth estimation process, a particular point in the rail cross-section
was chosen and that point defined the centroid of the hypothetical shell and transverse
defects. The assumed shape and tilt of the shell and transverse crack were also specified.
The crack growth rates at any point along the periphery of either ellipse was comput ed based
on the local estimate of the crack tip stress intensity. The stress intensity formulation for a
completely embedded elliptical flaw was used [3l ] with no edge or near surface correction

R a i l
Head<
W h e e l
,,~ Z d i r e c t i o n ~ j R u n n i n g s u r f a c e
. . . ~. ~. . . . . . - . ~. . . - . . . ~ --...-2...-S...-..22-- .............. ~ ..........
Typieol s h e l l . . . . . .
initiation z o n ~ ~ ~'~-
~ . ~ ~-~--.~-~.~---~- Live stress
~....,m--~.l ;~~;;;;';;;;;;;;~';.; Tolol stress at wheel passage
~. ~ C o m p r e s s i o n
FIG. 17--Predicted longitudinal stresses shown on a vertical plane through the rail head.
fact ors i ncl uded. The stress range that was used in each comput at i on of stress i nt ensi t y was
based on t he combi ned resi dual and act i ve st ress excursi ons (in t ensi on) that were pr edi ct ed
per pendi cul ar to the pl ane of t he crack at 12 si t es uni forml y spaced around each el l i pse.
The resi dual stress field was assumed to r emai n constant, even t hough the cr acked area
obvi ousl y coul d not sustain any t ensi l e resi dual stress. This choi ce was made because it was
bel i eved to be conservat i ve, si nce static equi l i br i um requi red that t he r emai ni ng uncracked
area pr evi ousl y in t ensi on had to move to st i l l hi gher t ensi on i f t he areas of the head that
were exper i enci ng compr essi ve resi dual st resses r emai ned in compressi on.
The pat t ern of evol ut i on of each el l i pt i cal flaw was not const rai ned. The flaws coul d
el ongat e, t ransl at e, or rotate, as l ong as t hey r emai ned in the same pl ane. Af t er i ncrement al
ext ensi ons of each el l i pse, the new el l i pse was defi ned t hrough a regressi on pr ocedur e that
best descr i bed t he now- di st or t ed new crack front.
Usi ng this basi c cr ack growt h al gori t hm, several di fferent cases were run so that compar-
i sons wi t h physi cal l y obser ved cracki ng pat t erns coul d be made. The predi ct ed resul t s for a
det ai l fract ure growi ng in t he t ransverse pl ane are shown in Fi g. 18. It is evi dent from this
figure that t he i ni t i al t ransverse defect, whi ch was assumed to be a somewhat el ongat ed
el l i pse t i l t ed appr oxi mat el y 0.35 rad f r om vert i cal , grew pr edomi nat el y downward and i nwar d
t owar d the cent er of t he rail. Thi s pat t ern was si mi l ar to that obser ved in many act ual det ai l
fractures (for exampl e, see Fi g. 1).
An exampl e of t he shel l gr owt h pat t erns t hat wer e obt ai ned is shown in Fi g. 19. The initial
flaw was assumed to be el l i pt i cal . Thi s i ni t i al flaw i mmedi at el y began to el ongat e al ong the
l ongi t udi nal axi s of the rai l and to grow onl y moder at el y in t he t ransverse di rect i on. Thi s
pat t ern of shel l growt h has been observed in vari ous defect i ve rails.
Anot her i mpor t ant feat ure of the shell gr owt h anal ysi s was t hat the rate of shell growt h
was found t o be hi ghest for shel l s l ocat ed appr oxi mat el y 10 mm bel ow the rai l surface. Thi s
dept h cor r esponds f avor abl y wi t h the l ocat i on of many nonsurface-connect ed shells. At
dept hs bel ow 10 ram, the t endency for shel l cr ack growth di rrdni shed substantially. Thi s lent
further cr edence to the i dea t hat t here is a cri t i cal shel l depth, bel ow whi ch shells are less

R I C E O N S H E L L A N D D E T A I L F R A C T U R E F O R M A T I O N 131
H o r i z o n t a l D i me n s i o n , X, mm
- 30 - 2 0 - 10 10
1 7 5 , r r e - - o e J ' ' ' I
. - - 4O
i / i / / C _
9 - - 20
Roi l
~ - Gag e face Cerr~rline
I
0 . 5 0 Assumed center of initial I
elliptical flaw is I
0. 59 inches below rumingeurface J - - 10
0. 45 inches in from g~ge face j
0 . 2 5 J I I I I I
- 1. 5 - 1. 25 -1 .OO - 0. 75 - O. 50 - 0. 25 O 0. 25 0 . 5 0
H o r i z o n t a l D i me n s i o n , X, i n.
FIG. 18--Predicted crack growth pattern f or a transverse defect in a rail head.
1. 50
._~
>." 1. 25
1. 00
0 , 7 5
>
E
E
>_-
3 0 ~"
o
E
>
l i kel y to cont i nue as shells and are mor e l i kel y to turn downwar d and cont i nue in the trans-
verse pl ane as a det ai l fracture.
Fact ors Inf l uenci ng Rai l He ad Crack Ini t i at i on
Many factors i nfl uence the l i kel i hood of fatigue cr ack i ni t i at i on in rai l road rails. Li ke
many ot her cri t i cal structural el ement s, fat i gue cracks in rai l road rai l s t end to initiate at
poi nt s of stress concent rat i on, ei t her geomet r i c or mi crost ruct ural . In the head of a rai l road
rail, t he most l i kel y site for crack i ni t i at i on is a nonmet al l i c inclusion. The rat e at whi ch
cracks wi l l form is di rect l y rel at ed to the severi t y of the l oadi ng on the rai l and the degree
of const rai nt i mposed on the rai l by t he t i es and bal l ast . Good bal l ast and ties, whi ch t oday
are oft en made of concret e, pr ovi de hi gher values of rai l vert i cal modul us, and this t ends to
r educe rai l bendi ng stresses. However, the rai l vert i cal modul us has little effect on the severi t y
of wheel / r ai l cont act stresses, whi ch are pr i mar i l y dependent on the qual i t y of the wheel /
rai l interface.
A number of t echni ques are used to mi ni mi ze the severi t y of wheel / r ai l cont act stresses
and to mi ni mi ze t hei r effect. Head gri ndi ng or profi l i ng is somet i mes used to maintain an
opt i mum wheel / r ai l cont act pattern. Rai l l ubri cat i on is used regul arl y on some rai l lines [32].
The use of head- har dened and heat -t reat ed rai l s (to achi eve hi gher overal l rai l hardness and
strength) is becomi ng mor e common [33,34].
There have been a vari et y of studies that have focused on predi ct i on of the effect of
oper at i onal and mat eri al vari abl es on fat i gue crack i ni t i at i on rat es in rai l steel, for exampl e,
Refs 35 and 36. Excerpt s from a t hi rd st udy [37] are i ncl uded here.

132 CASE STUDI ES FOR FATI GUE EDUCATI ON
25
20
15
0. 6-
E
E
>:
=-
O
E
c5
L)
>
10
j_ 5 Hor i z ont al
Di mens i on,
I X , i n.
- 0. 8 - 0. 6 I | - 0 . 4 ~ 0. 2
0. 8
-5
-10
-15
-20 --
-25 --
-30 p
-20
-~._o ! 0
U e')
'-2 ~, 9
~E>
I I I c 5 . 1 . 2 ~ I I I
-15 -10 -5 0 5 10 15 20
FIG. 19--Predicted crack growth pattern for a shell defect in a rail head.
In this study it was found that variations in rail loads, track support conditions, and the
quality and configuration of the rails combi ne to form a large number of different, potential
operating conditions. Variations in some of these conditions greatly affect the stress state
and the rate of fatigue damage accumulation while variations in some other operating con-
ditions (within realistic ranges) do not produce significant differences in rail stresses and
predicted fatigue lives.
The anticipated influence of these various factors in combination on rail fatigue lives was
addressed through a 45-element parametric study. The variety of conditions examined in this
parametric study is illustrated in Table 4. (Obviously, onl y selected combinations of these

RICE ON SHELL AND DETAIL FRACTURE FORMATION
TABLE 4--Matrix of conditions chosen f or the parametric study of rail fatigue.
133
Variable Units Levels Examined
59.2
Rail size kg/m 65.6
67.6
89
Wheel/rail verdcai force kN 133
222
0
Wheel/Rail lateral force kN 44
-13.3
mm from centerline of rail, 0.0
Wheel/rail loading point positive toward gage face
20.1
13.8
Track vertical modulus MPa 34.5
55.2
0.0
Rail head wear mm 7.6
variables were chosen for analysis.) To complete the analysis for each parametric case, a
total of 26 locations within the head of the rail were examined. The location of these points
coincided with nodal points within the finite element mesh used to generate values of stress
and strain.
The most critical locations in the rail head were typically in the upper third of the head
toward the gage face. Interestingly, these sites generally corresponded to the location of
maximum tensile residual stress. All analyses were based on an eight-level train spectrum
developed from a combination of several different train spectra, as shown in Table 5. Periodic
overstrain fatigue data trends were used for the cumulative damage analysis.
Using a uniaxial fatigue, equivalent strain model, computed axial rail stresses due to wheel
loading were combined with estimated residual stresses to make crack initiation life predic-
tions. Unworn and worn rails were treated as two major categories; type of traffic, lateral
load, and vertical modulus were treated as subcases within these categories. The position of
the contact patch was found to be a minor factor on predicted lives, except for the worn rail
cases, where the contact patch, when shifted from the rail center line toward the gage face,
produced significantly higher stresses and shorter fatigue life estimates.
Crack initiation life estimates varied from about 3 to 20 TN of traffic [300 to 2 500 million
gross tons (MGT)] for the various cases examined. A 45 kN lateral load typically reduced
TABLE 5--Equivalent train spectrum used f or crack initiation fatigue analyses.
Severity Level, Cumulative Frequency
% of Maximum Cycles per Block of Occurrence, % Wheel Load Equivalent
100.9 1 0.3 222 kN, unit train
83.2 5 2.0 ...
76.6 6 4.0 ...
68.4 18 10.0
58.4 30 20.0 133 kN, mixed freight
48.4 60 40.0 133 kN, unit train
35.0 60 60.0 ...
23.3 120 100.0 ...

cr ack i ni t i at i on l i fe est i mat es by about a factor of 2, whi l e a l ow vert i cal modul us (of about
14 MPa) r educed l i fe est i mat es by al most a fact or of 3 compar ed wi t h near - opt i mum vert i cal
modul us condi t i ons (of about 55 MPa). Heavy traffic, i nvol vi ng wheel l oads roughl y 20%
about nor mal mi xed traffic, wer e pr edi ct ed to r educe crack i ni t i at i on l i ves by roughl y 35%.
A subst ant i al l y worn rail, wi t h about 18% of t he head area removed, showed subst ant i al l y
hi gher axi al stresses due to wheel l oadi ng, and t herefore, crack i ni t i at i on l i ves were pr edi ct ed
to be about a fact or of 2 l ower t han for unworn rail. (In act ual servi ce t he wear is nor mal l y
gradual , whi ch woul d t end to make the worn r ai l l i fe est i mat es somewhat conservat i ve,
because t he cri t i cal damage sites woul d shift sl i ght l y over t i me. )
Beyond t he factors consi der ed in the ori gi nal par amet r i c anal ysi s, t he influence of t hree
ot her fact ors on sub-surface cr ack i ni t i at i on wer e al so exami ned: (1) t hermal stresses, (2)
nonmet al l i c i ncl usi ons, and (3) cl ass of track. A seri es of hypot het i cal cases were const ruct ed
based on t hree di fferent l oad severi t y curves, as shown in Fi g. 20, represent i ng: (1) a nomi nal
l ocat i on in a wel l mai nt ai ned track, (2) a hi gh l oad l ocat i on in a wel l - mai nt ai ned track, and
(3) a hi gh l oad l ocat i on in a l ower cl ass track. Bot h 0 and 70 MPa t hermal stress condi t i ons
were exami ned. The 70 MPa t her mal stress was assumed based on a - AT of about 40~
f r om the neut ral t emperat ure and a coeffi ci ent of t hermal expansi on of 8.6 10 6 m/ m/ C.
The influence of pot ent i al i ncl usi ons at cri t i cal sites in the rai l cr oss- sect i on were si mpl y
represent ed by stress concent rat i on factors rangi ng f r om 1.0 (no i ncl usi on) to 3.0 (rel at i vel y
flat, hard i ncl usi on in t he r ai l cross-sect i on).
The anal yses t hat were per f or med showed t hat i ncl usi ons woul d be expect ed to be qui t e
damagi ng i f t hey happened to l i e i n a t ensi l e resi dual stress zone in the r ai l head. Reduct i ons
99. 99 ! ~ I I I I i
99. 9
99
90
50
i0
1[
0. 1
0.Ol
0
3-SIGMA LOCATION. ,
~ POOPeR-CLASS TRACK
I
3-S I(~WA LOCATION.
WELL-MAINTAINED TRACK*
" - . . \
\ \ .
(~'~ "AVERAGE" F I E ~
OF RAIL)
\
NOt a t anomaly, s uch
as a r a l l j o l t \
\
FLAT ~ IMPACT EFFECTS , , . , , . . , . ~. ~
\
10 20 30 40 50 60 70
FIG. 20--Example of vertical wheel~rail load spectra due to spatial effects along a track and due to
track geometry quality.

in fat i gue l i fe of a fact or of 10 were pr edi ct ed for a defect i ve r ai l wi t h a K, = 3 i ncl usi on
compar ed wi t h a rai l wi t hout such a defect. I f a 70 MPa t her mal stress was al so superi m-
posed, t hen the crack i ni t i at i on fatigue l i fe was r educed by anot her fact or of 3. In compar i son
wi t h a nomi nal l oad l ocat i on in a wel l mai nt ai ned track, fat i gue l i ves were pr edi ct ed to be
4 t i mes l ower at hi gh l oad l ocat i ons in a l ower cl ass t rack (wi t h a l ow vert i cal modul us of
14 MPa). Dependi ng on t he exact combi nat i on of del et eri ous fact ors, servi ce lives as low as
0.2 TN (20 MGT) were predi ct ed, compar ed wi t h the " g o o d " rai l l i fe predi ct i on of about
20 TN (2500 MGT) . It shoul d be not ed that t hese anal yses di d not i ncl ude wear as a rai l
r emoval criteria; many rai l s are r emoved from servi ce because of l oss of head area and head
gri ndi ng or both, bef or e t hey accumul at ed about 4 TN (500 MGT) of traffic.
Hypothesis Concerning Detail Fracture Formation
As a resul t of the vari ous research efforts undert aken over t he past decade at Bat t el l e [ 3-
7,28,29] a hypot hesi s concer ni ng the process of det ai l fracture f or mat i on has been devel oped.
The hypot hesi s is based on t wo key el ement s:
Large cycl i c compr essi ve stresses occur in the rai l head. These stresses cause fatigue
damage and occasi onal cracks, especi al l y at sites where nonmet al l i c (oxi de) i ncl usi ons
are present. However, most of t hese cracks are of little consequence and wi l l nor mal l y
show erratic growth, i f any. A few cracks whi ch devel op in the area of l arge cycl i c
t ensi l e stresses ( i mmedi at el y under the pl ast i cal l y def or med l ayer of mat eri al ) wi l l grow
as a result of the cycl i c i n-pl ane t ensi l e stresses. Such a cr ack devel ops into a shell.
Al l shel l cracks wander and devel op smal l pert urbat i ons f r om their, "st r ai ght " path.
I f a shel l wanders upwar d it enters a r egi on with r api dl y decreasi ng cycl i c i n-pl ane
tension. Therefore, the pert urbat i on wi l l not persi st , and anot her pert urbat i on wi l l keep
the shel l crack appr oxi mat el y in its pl ane. A pert urbat i on downwar d al so causes the
crack to ent er a r egi on of rapi dl y decr easi ng cycl i c i n- pl ane tension. However, a per-
t urbat i on downwar d al so turns the pl ane of the crack i nt o a di rect i on of hi gh cycl i c
axi al t ensi on (al ong the axi s of the rail). As a result, such a pert urbat i on can persi st ,
turning the crack deci si vel y downward wher e it encount ers i ncreasi ng cycl i c axi al ten-
sion, and a det ai l fract ure is formed. Dependi ng upon the rel at i ve posi t i on of the shell
in the head of the rail, it may or may not form a det ai l fracture. The most cri t i cal dept h
wi l l vary from rai l -t o-rai l and depend on the severity of the wheel / r ai l l oadi ng, which,
in turn wi l l influence the resul t ant rai l deformat i ons and resi dual stress fields that
devel op.
The hypot hesi s suggest s t hat shells can pr opagat e in t ensi on and Mode I (crack openi ng)
i f t hey happen to i ni t i at e in the zone j ust bel ow the wor k har dened surface l ayer of the rai l
head. ( Al t hough this hypot hesi s does not speci fi cal l y address cr ack initiation, evi dence ci t ed
earl i er suggest s that most shells do i ni t i at e at oxi de i ncl usi ons or stringers that happen to be
l ocat ed in this regi on of hi gh mul t i axi al stress. ) Dependi ng on t he exact nature of t he resi dual
stress fi el d and the al i gnment of the shell, t hi s hori zont al flaw wi l l cont i nue to gr ow under
the i nfl uence of subst ant i al l ongi t udi nal t ensi l e stress excursi ons. I f stress condi t i ons are right,
a downwar d pert urbat i on in t he di rect i on of t he shell wi l l cause i t to turn gr adual l y into a
t ransverse pl ane and once agai n grow in Mode I, but as a det ai l fracture rather t han a shell.
The det ai l fracture wi l l be const rai ned to devel op as an embedded flaw in the ear l y stages
of growt h because the cr ack wi l l be i mpeded at the gage si de and surface of the rai l head
by compr essi ve l ongi t udi nal stresses. Af t er t he det ai l flaw grows t hrough a si gni fi cant port i on
of the rai l head (usual l y about 35%), and t he i nt ernal t ensi l e resi dual stresses are redi st ri b-

uted, the flaw will break out on the gage side of the rail. At this point the growth rate of
the flaw will accelerate significantly and failure of the rail will be imminent.
Conc l us i ons
Detail fractures from shells represent a serious concern to the rail transportation and freight
industry in terms of rail reliability and safety since detail fractures typically initiate and grow
within the head of a rail and may go unidentified until the strength of the rail is so deteri-
orated that a sudden failure may occur, which sometimes leads to a train derailment.
Much has been learned regarding the causes of shell defects in railroad rails and effective
means for reducing the frequency of their occurrence have been identified, in particular
through improvements in the cleanliness of the rail steel to reduce the number and size of
nonmetallic inclusions. Plausible concepts have been developed for model i ng the transition
of shells into detail fractures and for predicting the pattern and rate (at least qualitatively)
of crack growth as a function of various material and operating variables.
Increases in tonnage and axle loads on many U.S. railroad lines have placed greater stress
on rails and increased the likelihood of many different types of rail defects, including detail
fractures. At the same time, profitability of many rail lines has declined, which has forced
many railroads to minimize their budgets for rail inspection and replacement. In response to
these opposi ng forces, the railroad industry has responded with comprehensive inspection
programs to find and remove defective rails f r om track. These inspection systems include
detailed track records, most of which are comput eri zed [38]. The FRA has also conduct ed
extensive research to develop i mproved guidelines for rail inspection [39]. Accordi ng to these
new guidelines, it is now possible for railroads to adjust the frequency of rail inspection
based on the rate of defect occurrences and their growth characteristics [40]. The next maj or
hurdle will be to extend this knowl edge base to the safe use and maintenance of high speed
rail lines that will undoubtedly be developed in the United States in the late 1990s and early
21 st Century.
Acknowl edgment s
The research work that forms the basis for this case study was completed by numerous
individuals over a period of several decades. Many of those individuals have been cited in
the list of references. At Battelle, Mr. Donal d Ahlbeck, Mr. Jeffrey Hadden, Dr. James
Kennedy, Dr. Frank Dean, and Mr. Robert Prause have been among the most significant
contributors. Mr. Ravi Rungta, currently at General Motors, Harrison Radiator Division, was
also a key contributor on the rail fractography work. Dr. David Broek, currently a private
consultant, was a maj or contributor in the crack growt h prediction work. Mr. Harold Harri-
son, who is now the President of Salient Systems, Inc., led much of the early work on
wheel/rail load measurements.
Within the Department of Transportation, Dr. Oscar Orringer and Mr. James Morris (now
deceased) at VNTSC, Mr. Arne Bang at the FRA, and Dr. Roger Steele at the Association
of Ameri can Railroads have been major contributors and supporters of this research.
Re f e r e nc e s
[1] National Transportation Safety Board, "Broken Rails: A Major Cause of Train Accidents," Report
NTSB RSS 74-1, National Transportation Safety Board, Washington, D.C., 2 Jan. 1974.

[2] Mack, G. A. and Hopper, A. T., et al., "Statistical Analysis of Burlington Northern and Atchison,
Topeka and Santa Fe Rail Failures," Task 5, final report to the Department of Transportation from
Battelle Columbus Laboratories, Cambridge, MA, Contract DOT-TSC-1708, 8 June 1982.
[3] Rungta, R., Rice, R. C., and Buchheit, R. D., "Post-Service Rail Defect Analysis," first interim
report to the Transportation Systems Center, Department of Transportation, Cambridge, MA, July
1982.
[4] Rice, R. C., Rungta, R., and Buchheit, R. D., "Post-Service Rail Defect Analysis," second interim
report to the Transportation Systems Center, Department of Transportation, Cambridge, MA, Jan.
1983.
[5] Rice, R. C., Rungta, R., and Broek, D., "Post-Service Rail Defect Analysis," third interim report
to the Transportation Systems Center, Department of Transportation, Cambridge, MA, July 1983.
[6] Rice, R. C., Rungta, R., and Scott, P. M., "Post-Service Rail Defect Analysis," fourth technical
report to the Transportation Systems Center, Department of Transportation, Cambridge, MA, June
1984.
[7] Rungta, R. and Rice, R. C., "Shelling Behavior in Rails," final report to the Transportation Systems
Center, Department of Transportation, Cambridge, MA, Feb. 1987.
[8] Campbell, J. E., McIntire, H. O., and Manning, G. K., "Summary Report on the Examination of
Rails Which Contain Detail Fractures," Rail, Appendix 10-b, American Iron & Steel Institute,
Birmingham, AL, 1949, pp. 608-620.
[9] Rail Defect Manual, compiled by Sperry Rail Service, Sperry Division of Automation Industries,
Inc., Danbury, CT, 1964, 4th printing, 4/ 68, p. 38.
[I0] Groom, J. J., "Residual Stresses in Railroad Rails," Battelle Columbus Laboratories, Columbus,
OH; paper presented at the 1980 Fourth SESA International Congress on Experimental Mechanics,
25-30 May, see also Federal Railroad Administration Report FRA/ ORD83/ 05, May 1983.
[11] Ahlbeck, D. R., Harrison, H. D., Tuten, J. M., and Johnson, M. R., "Wheel/Rail Load Dat a- - A
Statistical Characterization," Interim Report to the U.S. Department of Transportation, Federal Rail-
road Administration, Jan. 1979.
[12] Harrison, H. D. and Tuten, J. M., "Correlation of Concrete Tie Track Performance in Revenue
Service and at the Facility for Accelerated Service Testing," Report DOT/ FRA/ ORD-84/ 02. 1, Vol.
I, A Detailed Summary, Final Report, Department of Transportation, Washington, D.C., Aug. 1984.
[13] Orringer, O., Tang, Y. H., Gordon, J. E., Jeong, D. Y., Morris, J. M., and Perlman, A. B., "Crack
Propagation Life of Detail Fractures in Rails," Report DOT/FRA/ORD-88-13, final report, U.S.
Department of Transportation, Federal Railroad Administration, Washington, D.C., Oct. 1988.
[14] Manual for Railway Engineering, American Railway Engineering Association, Chicago, IL, 1971,
pp. 4-1-6.
[I5] Feddersen, C. E., Buchheit, R. D., and Brock, D., "Fatigue Crack Propagation in Rail Steels,"
Report FRA/ ORD-77/ 14, interim report by Battelle to the U.S. Department of Transportation,
Federal Railroad Administration, Washington, D.C., June 1977.
[16] Broek, D. and Rice, R. C., "Fatigue Crack Growth Properties of Rail Steels," final report, Part I,
submitted by Battelle to the U.S. Department of Transportation, Transportation Systems Center,
Washington, D.C., July 1977.
[17] Broek, D., Rice, R. C., and Steele, R. K., "Prediction of Fatigue Crack Growth in Railroad Rails,"
presentation to the Transportation Research Board, 5 Aug. 1977.
[18] Rice, R. C. and Broek, D., "Fatigue Crack Initiation Properties of Rail Steels," Report DOT/
FRA/ ORD-82/ 05, final report to the U.S. Department of Transportation, Federal Railroad Admin-
istration, Washington, D.C., March 1982.
[19] Fowler, G. J., "Fatigue Crack Initiation and Propagation in Pearlitic Rail Steels," Ph.D. Thesis,
University of California, Los Angeles, 1976.
[20] Park, Y. J. and Fletcher, E B., "Fatigue and Fracture Toughness of Standard Carbon and High
Strength Rail Steels," Paper 82-HH-21, American Society of Mechanical Engineers, New York,
1982, pp. 1-8.
[21] Marich, S., "Research on Rail Metallurgy," Paper 78-663-8, Bulletin 663, American Railway En-
gineering Association, Chicago, IL, 1988.
[22] Ito, A. and Kurihara, "Shelling of Rails Experienced in Japanese Railways," Permanent Way Society
of Japan, Vol. 8(2), No. 27, 1965, pp. 17-32.
[23] Heller, W., "The Development and Further Development of the Quality and Service Properties of
Modem Rail Steels," ETR-Eisenbahntechnische Rundschau, Jan./Feb. 1971, pp. 71-78.
[24] Zolotarsky, A. E and Rauzin, Y. R., "Increasing the Strength of Rails and their Reliability in Service
on the Railways of USSR," Rail International, Dec. 1971, pp. 908-915.

[25] Besuner, E M., "Fracture Mechanics Analysis of Rails with Shell-Initiated Transverse Cracks,"
Rail Steels--Developments, Processing, and Use, ASTM STP 644, D. H. Stone and G. G. Knupp,
Eds., American Society for Testing and Materials, Philadelphia, 1978, pp. 303-329.
[26] Mair, R. I. and Groenhout, R., "The Growth of Transverse Fatigue Defects in the Head of Railway
Rails," Rail Research Papers, VoL 1, Melbourne Research Laboratories, Melbourne, Australia, 1988,
pp. 170-186.
[27] Chipperfield, C. G., "Modelling Rail Head Fatigue Using Fracture Mechanics," Rail Research
Papers, Vol. l, Melbourne Research Laboratories, Melbourne, Australia, 1988, pp. 213-217.
[28] Rice, R. C. and Rungta, R., "Modeling of Embedded Fatigue Cracks in Railroad Rails," Fatigue
and Fracture of Engineering Materials and Structures, Vol. 11, No. 3, 1988, pp. 189-203.
[29] Rungta, R., Rice, R. C., Buchheit, R. D., and Brock, D., "An Investigation of Shell and Detail
Cracking in Railroad Rails," Microstructural Science, Volume 12, Corrosion, Microstructure, and
Metallography, Proceedings of the Sixteenth Annual Technical Meeting of the International Me-
tallographic Society, 1985, pp. 383-406.
[30] Suresh, S., "Crack Initiation in Cyclic Compression and Its Applications," Engineering Fracture
Mechanics, Vol. 21, No. 3, 1985, pp. 453-463.
[31] Tada, H., Paris, P. C., and Irwin, G. R., The Stress Analysis of Cracks Handbook, Del Research
Corporation, St. Louis, MO, June 1973, p. 26-31.
[32] Steele, R. K., "The Consequences of Truly Effective Lubrication Upon Rail Performance," ASME
Paper 84-WA/RT-17, American Society of Mechanical Engineers, New York, 1984.
[33] Mutton, P. J. and Epp, C. J., "Factors Influencing Rail and Wheel Wear," Rail Research Papers,
Vol. 1, Melbourne Research Laboratories, Melbourne, Australia, 1988, pp. 277-282.
[34] Marich, S., "Development of Improved Rail and Wheel Materials," Rail Research Papers, Vol. 1,
Melbourne Research Laboratories, Melbourne, Australia, 1988, pp. 147-164.
[35] Leis, B. N., "Microcrack Initiation and Growth in a Pearlitic Steel--Experiments and Analysis,"
Proceedings, 15th National Fracture Symposium, University of Maryland, July 1982.
[36] Jeong, D. Y., Orringer, O., Perlman, A. B., and Rice, R. C., "Beam Theory Predictions of Shell
Nucleation Life," Proceedings, Second International Symposium on Contact Mechanics and Wear
of Rail/Wheel Systems, University of Rhode Island, Kingston, RI, 8-11 July 1986.
[37] Rice, R. C., Ahlbeck, D. R., Kennedy, J. C., Leis, B. N., Meacham, H. C., and Sampath, S. G.,
"Analysis of Service Stresses in Rails, Report DOT-TSC-1663, final report submitted to the U.S.
Department of Transportation, Federal Railroad Administration, Washington, D.C., March 1983.
[38] Orringer, O. and Bush, M. W., "Applying Modem Fracture Mechanics to Improve the Control of
Rail Fatigue Defects in Track," Bulletin 689, American Railway Engineering Association, Chicago,
IL, 1983.
[39] Orringer, O., "Control of Rail Integrity by Self-Adaptive Scheduling of Rail Tests," Report DOT/
FRA/ORD-90-05, Volpe National Transportation Systems Center for the U.S. Department of Trans-
portation, Federal Railroad Administration, Washington, D.C., June 1990.
[40] "Improving Transportation Through Railroad Research: 1988-1991," Report DOT/FRA/ORD-92-
14, summary report, U.S. Department of Transportation, Federal Railroad Administration, Office
of Research and Development, Washington, D.C., July 1992.

Ralph L Stephen& 1 Timothy B. Adams ~ and Stephen L. Carlson I
Equating Damped Vibration to Constant
Amplitude Fatigue Loading for a Thick-
Walled Pressure Vessel
REFERENCE: Stephens, R. I., Adams, T. B., and Carlson, S. L., "Equat i ng Damped
Vibration to Constant Amplitude Fatigue Loading for a Thick-Walled Pressure Vessel,"
Case Studies f o r Fatigue Education, ASTM STP 1250, Ralph I. Stephens, Ed., American Society
ABSTRACT: A single impulse firing of an A723 steel thick-wall cannon tube has been con-
sidered by the U.S. Army to be equivalent in fatigue to a single 0.2 Hz laboratory simulated
constant load amplitude test. However, a high frequency (15 kHz) resonant damped vibration
has been found to accompany the firing in some situations and hence raise concern about the
single loading equivalent simulation fatigue test philosophy. This case describes the analytical
and test programs to determine a new fatigue equivalence that incorporates both the normal
impulse, the resonant damped vibration, and the constant amplitude simulation tests. Analytical
and experimental fatigue lives were based upon strain life (e-N), stress-life (S-N), and linear
elastic fracture mechanics LEFM ( da/ dN - AK) models. Quasi-equivalence testing was con-
ducted between 0.2 and 30 Hz using both constant amplitude and a damped vibration spectrum
with axial smooth specimens and SE(T) specimens. The analytical and experimental results,
modified for frequency effects, provide a new fatigue equivalence with significant differences.
KEYWORDS: fatigue (materials), resonant vibration, cumulative damage, life predictions,
frequency effects, pressure vessel, case study
The U. S. Ar my cr i t er i a for r el eas i ng a new t hi ck- wal l ed c a nnon t ube desi gn for field
servi ce requi re a dur abi l i t y or al l owabl e fat i gue l i fe veri fi cat i on. Thi s fat i gue life ver i f i cat i on
r equi r es r epeat ed exper i ment al fi ri ngs and si mul at ed l abor at or y pr essur i zat i ons. The fi ri ng
pr essur e- t i me cur ve and hence t he Lai n6 hoop st ress-t i me cur ve for t hese t hi ck- wal l ed pres-
sure vessel s is s hown schemat i cal l y i n Fi g. 1. The t i me per i od for a fi ri ng is i n mi l l i seconds.
The ma x i mu m hoop stress val ue is used i n r epeat ed l abor at or y si nusoi dal pr essur i zat i on t ube
t est s at 0. 2 Hz, as s hown i n Fi g. 2. The stress ratio, R = trmi,/tr~a x is zero, as shown. Based
upon equi val ent f at i gue damage or f at i gue life, each 0. 2 Hz cons t ant ampl i t ude, R = 0,
si mul at ed pr essur i zat i on has been s hown to correl at e essent i al l y 1 to 1 wi t h t he act ual firing.
Thus, for ma ny years, t he 1 to 1 equi val ence has be e n an i mpor t ant fact or i n t est i ng, and
hence, set t i ng safe-l i fe fi ri ngs for t hi ck- wal l ed c a nnon t ube.
It has been f ound t hat a r es onant var i abl e ampl i t ude vi br at i on at appr oxi mat el y 15 kHz
coul d be s uper i mpos ed on t he c a nnon t ube dur i ng fi ri ng, r esul t i ng i n a hoop st ress-t i me
hi st ory, as s hown i n Fi g. 3. Thi s r esul t ant vi br at or y st r ess- t i me hi st ory shows an i ncr easi ng
ampl i t ude phase wi t h zero me a n stress, a peak stress, %~ak, f ol l owed by a decr easi ng am-
pl i t ude phase wi t h a vi br at or y me a n stress, %m, si mi l ar to t he nor mal Lam6 pr essur i zat i on
i Professor and research assistants, respectively, The University of Iowa, Mechanical Engineering De-
partment, Iowa City, IA.
139

( ~ max
t i m e i n m i l l i s e c o n d s =
FI G. 1 - - S c h e m a t i c o f o r i g i n a l f i r i n g s t r e s s hi s t or y.
stress fol l owed by a dr op to zero stress. Thi s new 15 kHz vi brat ory stress hi st ory raises
si gni fi cant concern about the 1 to i equi val ence stated earlier. The goal of this proj ect is to
det er mi ne the equi val ence of the usual const ant ampl i t ude l abor at or y t est i ng at 0.2 Hz wi t h
R = 0 to that of t he new 15 kHz vari abl e, ampl i t ude vi br at or y stress history.
St ude nt Que s t i ons a nd I ns t r uc t i on
At this poi nt the cl ass shoul d prepare a list of i t ems that need to be consi der ed in order
to det er mi ne a new equi val ence.
1. What anal yt i cal , comput at i onal , or exper i ment al aspect s and model i ng should be
undert aken?
2. What l i t erat ure revi ews wi l l be needed to address some of t hese aspect s?
3. What aspect s are t he most i mpor t ant and what aspect s are secondar y or of smal l
influence?
4. What addi t i onal i nformat i on wi l l be needed f r om U. S. Ar my personnel ?
Pr o c e dur e s
Di scussi ons wi t h U. S. Ar my personnel pr ovi ded addi t i onal i nformat i on concerni ng t he
probl em. At first, it was desi rabl e to formul at e a compl et e range of equi val ence for many
frequency = 0 . 2 H z /
i
N = ? c y c l e s
]
5 1 0 1 5 2 0
t i m e ( s e c o n d s )
FI G. 2 - - L o w f r e q u e n c y c o n s t a n t a mp l i t u d e , R = O, s t r e s s hi s t or y .

STEPHENS ET AL. ON A THICK-WALLED PRESSURE VESSEL 141
( ] r v m
. . . .
increasing phase d ~ r ~ n g
0
t t r n e - m i l l i s e c o n d s O. 1 3 3
FIG. 3--Schematic representation of the resonant vibratory stress history.
values of %, ~ and ~r~,,. This was soon restricted to more realistic values with %o~, = 1034
MPa (150 ksi) and Crvm = 517 MPa (75 ksi). This provided an overload ratio, ~rr~,/crvm, of
2. The number of cycles in each vibratory load history to represent the actual input firing
was determined by the Ar my to be 20, as shown in Fig. 3. With this information available,
the key aspects in obtaining the desired equivalence for this ASTM A723, Grade 1, Class
4, Ni -Cr-Mo tempered martensitic alloy pressure vessel steel was then formulated as follows:
1. Possible elimination of the 15 kHz resonant vibration.
2. Equivalent fatigue life prediction models based upon:
(a) Local strain approach using low cycle fatigue strain-life (S-N) concepts.
(b) Nomi nal stress approach using stress-life (S-N) concepts.
(c) Fatigue crack growth using linear elastic fracture mechani cs (LEFM) concepts.
3. Simulated experimental test program based upon smooth specimens and precracked
specimens, or both.
4. Wave form quantification.
5. Cumulative damage models.
6. Mean stress and residual stress effects.
7. Cycl e counting methods.
8. Not ch or stress concentration effects.
9. Frequency effects for 15 kHz and 0.2 Hz.
10. Environmental effects of corrosion and temperature.
11. Biaxial effects.
12. Material properties needed.
13. Applicable standards.
Elimination o f the kHz Resonant Vibration
Many different cannon tube sizes and designs exist in the Army and design changes to
alter resonant vibrations are always possible. However, design changes may have restrictions
and can take substantial time. U.S. Ar my personnel preferred to directly attack the problem
assuming the resonant condition persisted and thus an equivalence was still sought.
Temperature, Corrosion, Biaxial and Residual Stress, and Notch Considerations
Cannon tubes are expected to be used under all climatic temperature conditions and could
involve climatic temperatures from - 4 5 ~ to 50~ Low climatic temperature fatigue resistance

involving bot h fatigue crack format i on and fatigue crack growth should be equal to or better
than that at r oom temperature for this quality Q&T alloy steel [1]. This generally has also
been assumed in previous Ar my cannon tube equivalence criteria. However, fracture tough-
ness, and hence critical crack size, can be reduced at the lower climatic temperatures. This
latter effect has been principally addressed and solved through stringent microstructure, com-
position, and Charpy V-notch i mpact requirements. Thus, it was assumed that room temper-
ature modeling and experiments should be conservative relative to low-temperature effects.
The actual elevated temperature of the cannon tube can be expected t o exceed that of the
highest climatic temperature due to heating effects during firing. In the regions of concern
for this resonant vibration loading, temperature should be less than 200~ as shown by
previous U.S. Ar my findings. This temperature is not high enough to form a significant
degradation of fatigue resistance for this alloy steel. Thus, it was assumed that room tem-
perature modeling and experiments woul d be satisfactory for all temperatures involved.
Corrosion from firing and f r om climatic conditions has been incorporated into the previous
successful equivalence criteria by firing each laboratory test cannon tube several times prior
to laboratory testing. It was assumed that the 15 kHz resonant vibration would not cause
increased corrosion problems, and hence, modeling and experiments excluding additional
corrosion considerations should be appropriate.
During firing a large tensile circumferential (hoop) stress and a smaller compressive radial
stress exists. For both the analytical and experimental portions of this case study, a uniaxial
state of stress was assumed.
Residual stresses and notch (stress concentration) effects significantly alter the fatigue
resistance of cannon tubes. These are both taken into consideration with the 1 to 1 equiva-
lence by laboratory testing of actual cannon tube breech sections. By considering the muzzle-
end region as the principal 15 kHz resonant vibratory concern, both the residual stress and
notch considerations are significantly reduced in importance. Thus, U.S. Army personnel
recommended this equivalence study assume residual stresses are small and that the theo-
retical stress consideration factor, K,, was equal to one. Other locations that contain signif-
icant residual stresses and K, values could be evaluated at a later date.
Student Questions
1. What information can you find to justify the above assumptions on temperature and
corrosion effects?
2. What climatic corrosive and firing corrosive environments actually exist with this can-
non tube?
3. Many fatigue life prediction software programs and many laboratory simulation com-
ponent tests omit temperature and corrosion environments. How are these omissions
incorporated into safe durability?
4. What relative changes in fatigue strength or fatigue life can occur in real components
under biaxial state of stress and by inducing surface residual stresses or reducing stress
concentrations?
Mathematical Representation of the Vibratory Loading Wave Form
An equivalence determination will involve fatigue life calculations and simulated fatigue
experiments. Bot h of these procedures require a quantification of the 15 kHz vibratory wave
form of Fig. 3. The needed information is the peak and valley of each cycl e or reversal in
Fig. 3. This waveform can be t hought of as a block loading of 20 sinusoidal cycles lasting

STEPHENS ET AL. ON A THICK-WALLED PRESSURE VESSEL 1 4 3
0.133 milliseconds. The stress amplitude of each cycle, however, is variable. For the first
ten cycles the waveform is centered around zero and thus can be t hought of as fully reversed
with the stress amplitude of each cycle increasing exponentially. This first hal f of the loading
is the increasing phase. At the first hal f of the eleventh cycle, a peak value is reached, Crp~,,
and then the stress amplitude begins to decrease exponentially. In contrast to the first ten
cycles, the last ten cycles (the decreasing phase) are centered about a positive mean stress
value, %,,. After the 20th cycle, the stress amplitude is zero. Then the waveform immediately
drops from the positive mean stress to zero stress. From this point, the next repetition of the
bl ock loading begins.
Given the exponential characteristics of the waveform, a quantitative representation was
produced by empl oyi ng the mathematical model of the free vibration of a viscously damped,
single degree-of-freedom, second-order syst em expressed by the following ordinary differ-
ential equations
m ~ ( t ) + cJc(t) + k x ( t ) = 0 (1)
o r
)~(t) + 2~to~(t) + to 2 x(t) = 0 (2)
The solution of these equations for the underdamped case (0 < { < 1) can be represented
in general form by
x ( t ) = A e - ~ " m sin COd(t) ( 3 )
The frequency of the waveform is known to be 15 kHz or 94 krad/ s. This is the same as
the damped frequency, to a, that has the fol l owi ng relationship to the undamped natural fre-
quency of the system and the viscous dampi ng factor,
cod = t%~v/1 - ~2 (4)
The viscous damping factor can then be found by its relationship with the logarithmic dec-
rement, 8, given as
r - (5)
-~v/(2,rr)2 + ~2
The logarithmic decrement is a quantity determined by the rate of decay of the vibration
n - 1 (6)
I f the amplitudes of t wo peaks are known, for example x 1 and x,, the logarithmic decre-
ment can be calculated. This in turn yields the viscous dampi ng factor and finally the un-
damped natural frequency can be calculated. The last variable of Eq 3 to be found is the
amplitude coefficient, A. The amplitude, x ( t ) , at the first peak, xl, is known. This will occur
at the first quarter cycl e or t = 0.016 67 milliseconds. Now A can be solved for by substi-
tuting the known values into Eq 3.

To appl y this model to t he quant i t at i ve formul at i on of the bl ock vi brat ory l oadi ng of
interest, bot h the i ncreasi ng and decreasi ng phases had to be cal cul at ed separat el y and su-
per i mposed at the peak stress of 1 034 MPa. The i ncreasi ng phase of the wavef or m was
si mpl y assumed to be a mi r r or i mage of t he second- or der syst em j ust formul at ed. The i ni t i al
peak ampl i t ude, x~, of this vi brat i on is known to be 1034 MPa and then essent i al l y damps
out to zero after ten and a hal f cycl es. Not e that because of t he overl ap at the peak stress,
an addi t i onal hal f-cycl e had to be added to t he i ncreasi ng phase. Next a val ue had to be
chosen for " z e r o" si nce the t heoret i cal wavef or m ampl i t ude cont i nues to osci l l at e at very
smal l ampl i t udes. "Essent i al l y zer o" was chosen r eal i st i cal l y to be one percent of t he initial
peak ampl i t ude, or xl t = 10.3 MPa. Wi t h t hese val ues and the frequency, the wavef or m was
generat ed and then reversed to yi el d the i ncr easi ng phase of the wavef or m shown in Fi g. 3.
Next, the decreasi ng phase had to be gener at ed and combi ned wi t h the i ncreasi ng phase.
The decreasi ng phase of the wavef or m decays exponent i al l y f r om t he peak value, but there
is a posi t i ve mean stress, cry, ., i mposed on t hese cycl es. In or der to make this por t i on of the
wavef or m consi st ent wi t h the assumpt i ons for bot h exponent i al decay and t he val ue of zero,
a separat e l ogari t hmi c decr ement had to be cal cul at ed that depended on t he mean stress.
Not e that because of t hese assumpt i ons, i mposi ng a mean stress on the decreasi ng phase
al so changes the stress ampl i t ude of t hese cycl es. Fi nal l y, t he i ncreasi ng and decreasi ng
phases were combi ned at the peak stress of 1 034 MPa to form the compl et e waveform. The
dampi ng charact eri st i cs of the compl et ed wavef or m are gi ven in Table 1 for bot h the in-
creasi ng and decreasi ng ampl i t ude phases. Fr om t hese values, the needed magni t udes of all
peaks and val l eys were obt ai ned.
Equi val ence Fat i gue Li f e Predi ct i on Model Sel ect i on
Bot h the l ocal strain model usi ng low cycl e fatigue st rai n-l i fe (e-N) concept s and the
nomi nal stress model usi ng st ress-l i fe (S-N) concept s were chosen to eval uat e the fat i gue l i fe
equi val ence. These model s assume the fat i gue l i fe of a component is that to form a crack
of several mi l l i met er s in size and is based upon the usual cr ack si ze of fracture t hat occurs
in reference smal l speci men const ant ampl i t ude tests. An LEFM model was al so used to
det ermi ne reasonabl e fat i gue crack growt h l i fe equivalence. These model s are cl ari fi ed in
Refs 2- 5.
The general e-N model used is
_ o r )
Ae 2~e e + _ _ = __ (2N) b + el ( 2N)C
2 2 2 E
(7)
al ong wi t h t he const i t ut i ve equat i on
TABLE 1--Dampi ng characteristics f or the variable amplitude
block history, crp~,k = 1 034 MPa, trvm = 517 MPa.
Increasing Decreasing
Characteristic Amplitude Amplitude
n 11 10
X1/ X, 100 67.7
0.461 0.467
0.0731 0.0741
w, rad/ s 94 500 94 510
A 1 160 1 161

The general S - N model is
~ l l n '
a_~=~+ % (8)
2 E \ K ' /
~-= % = ~) ( 2 N) b (9)
whi ch is t he el ast i c component of Eq 7. Bot h t he Mor r ow mean stress model (Eq 10), and
the Smith, Wat son, and Topper mean stress model (Eq 11), were used in the l ocal strain
based appr oaches
and
a ~ = (,~5 - < . ) ( 2 N ) "
+ ,E) (2.'V) c (10)
2 E
A~
O-maxE -~-- = ((I))2(2N) b + (r}~)E(2N) b+~ ( 1 1 )
O ' m a x = Or m - {- O" a (12)
The modi f i ed Goodman mean stress model (Eq 13), was used for the nomi nal stress appr oach
O" a Or m
+ T . = 1 ( 13)
where Sy is t he fat i gue st rengt h at a specific fat i gue life. Addi t i onal symbol s are defi ned in
Tabl e 2. Act ual tensile, cycl i c stress-strain, and l ow cycl e fat i gue A723 mat eri al propert i es,
TABLE 2--Monotonic, cyclic, and low cycle fatigue properties
of ASTM A723 steel [6].
Young's modulus, E, GPa
0.2% offset yield strength, Sy,, MPa
Ultimate tensile strength, S,, MPa
% elongation
% reduction in area
Strain hardening coefficient, K, MPa
Strain hardening exponent, n
Rockwell hardness, HR c
Cyclic strength coefficient, K' , MPa
Cyclic strength exponent, n'
0.2% offset cyclic yield strength, Sy,, MPa
Fatigue strength coefficient, ~r}, MPa
Fatigue strength exponent, b
Fatigue ductility coefficient, e)
Fatigue ductility exponent, c
Young's modulus at half-life, E' , GPa
200
1 170
1 262
13
5O
1 483
0.037
40
1 581
0.071
1 014
2 123
-0. 110
0.489
-0. 783
193

obtained by Koh and Stephens [6], and used in the e-N and S-N models are also given in
Table 2.
Fatigue crack growth models used were the Paris model (Eq 14) and the Foreman model
(Eq 15)
da
~-~ = C(AK)" (14)
da C(AK)"
7 - U = ( 1 - R ) K c - a X (15)
Partial constant amplitude fatigue crack growth data for A723 steel were available from
the Army; however, a wider range of da/ dN was desirable and additional fatigue crack
growth testing was performed to obtain the needed fatigue crack growth parameters.
The above three models must incorporate cycle counting procedures. The rainflow count-
ing model is by far the most common procedure because it brings into effect closed hysteresis
loops and mean stress. Rainflow counting was used to cycle count the vibratory load history
of Fig. 3 and to provide both stress amplitude and mean stress.
The above three models must also incorporate a cumulative damage procedure. For the ~-
N and S-N approaches, the Palmgren-Miner linear damage rule has been the most common
rule for the past several decades. It was decided, however, to evaluate other cumulative
damage models to determine if better models could be used in the equivalence calculations.
In order to review this topic comprehensively, three computer data bases were employed as
a means of locating references: COMPENDEX PLUS, which corresponds to the Engineering
Index, the National Technical Information Service (NTIS), and Applied Science and Tech-
nology. From these references and subsequent additional references, existing cumulative
damage models were evaluated as to their applicability to this problem. Numerous papers
and overviews of these methods were found in the literature search. References 7 through
11 include significant overviews of the different cumulative damage models. Many models
of differing complexity were found in the literature review. Non-linear damage and inter-
action effects are very difficult to overcome. The more complex models often require em-
pirical parameters that require extensive testing. It was decided, as has been done by most
others, that the simplicity and oftentimes reasonable accuracy makes the Palmgren-Miner
linear damage rule very attractive. It will be used in the ~-N and S-N fatigue life equivalence
calculations.
Frequency Effects
The previously mentioned three computer data bases were again used to obtain references
evaluating frequency fatigue effects in metals. Temperature, corrosion, and strain rate sen-
sitive stress-strain constitutive relations were shown to be important parameters affecting
frequency fatigue effects. Excluding the temperature and corrosion effects, frequency metal
fatigue effects between 0 and 200 Hz were usually small. Generally within this frequency
range the fatigue resistance based upon e-N, S-N, and da/ dN-AK could increase slightly with
increased frequency. This small influence was important in deciding later what test frequency
to choose for representing the constant amplitude 0.2 Hz simulated pressurizing.
Most recent kHz fatigue testing has been run at about 20 kHz since this frequency is
above the available range of the human ear. Four recent comprehensive publications on high

frequency fatigue were evaluated [12-15] along with other papers. Pneumatic, piezoelectric,
and magnetostrictive test systems have been used to test at frequencies in the kHz range.
Smooth, notched, and crack growth specimens have been tested. Various cooling media such
as air, water, water modified with corrosion inhibitors, and oil have been used to cool spec-
imens because of self-heating caused by internal friction during high frequency testing.
Figure 4 shows representative S-N curves for three carbon steels and one CrMo steel tested
at lower frequency (27 to 70 Hz) and at 20 kHz [16]. For all four steels, the S-N curve shifts
upward and to the fight for the high frequency loading indicating increased fatigue resistance
at the kHz frequency range. Figure 5 shows representative da/ dN-AK curves for several
steels tested at 70 Hz and at 22 kHz [17]. The band of results indicate significant increase
in fatigue crack growth resistance for the kHz frequency range.
The above S-N and da/ dN-AK frequency results are representative of much of the data.
However, quantitative values of increased fatigue resistance at kHz frequencies range from
almost zero to values higher than depicted in Figs. 4 and 5. This depends upon test speci-
mens, corrosion, heating, and microstrncture. Less increase in fatigue resistance was noted
by Kuzmenko [15] for steels with substantial Ni content. Puskar [18] has recommended
several quantitative correction factors for S-N and da/ dN-AK curves for steels. These along
with suggestions by Kuzmenko [15] form the following shifts in kHz fatigue resistance
relative to 0 to 25 Hz for this A723 steel that will be used in forthcoming fatigue life
calculations.
S-N Curve
1. Shift the fatigue limit from 107 cycles to 108 cycles.
2. Increase the fatigue limit by a factor of 1.1.
3. Retain the same slope for <108 cycles.
or
.=~
E
Curves 1: 0.07% C
Curves 2: 0.37% C
360 - Curves 3: 0.56% C
Curves 4: CrMo
\
N
3 2 0 - ~ ~ \ m_ _ 20kHz
N . ~ \ "x ~ 25 - 70 Hz
280
- . " " - " . . " - - - - - 3
240
200- ~ NN- - 2 _ _ _ 1
- - 1
160 I I t 1
105 106 107 108 109
Cycles - N
FIG. 4--Frequency effect on S- N curves f or four steels tested at 25 to 70 Hz and 20 kHz [16].

1 4 8 CASE STUDIES FOR FATIGUE EDUCATION
10 4
10 3
E
1 "02
x
I01
9 8
/ / I
/ / /
i / /
r
///
iii
7 0 H z - - - , 1 / , ' 1 "
4 ' / / / / , 3 2
/'/ 7 '
;? 1.
1 . . .I . . I . ,
a 5 6 7 8 10 20 30 t,0 50
K o ~ MPo m I/2
F I G . 5 - - F r e q u e n c y e f f e c t o n d a l d N - A K f o r s e v e r a l s t e e l s tested at 7 0 H z and 22 k H z [ 17] .
The above shi fts wi l l resul t in a f act or o f t we l v e i ncreas e i n l i f e for a gi ve n alternating st ress
l e v e l and a 31% i ncreas e i n alternating stress for a gi ve n l i f e.
e - N
1. Shi f t the el as t i c curve the s a me as the gi ve n preceedi ng S-N curve.
2. Shi f t the pl ast i c curve up 10%.
3. Ret ai n the s a me el as t i c and pl ast i c s l ope s .
The s e shi fts wi l l change the val ue o f ~r} t o 2788 MPa and e} t o 0 . 5 3 8 for 15 kHz . Thi s
resul t s in an i ncreas e i n f at i gue l i f e be t we e n a f act or o f 2 at hi gher strain ampl i t udes and a
factor o f 12 at l o we r strain ampl i t udes. The s e shi f t s al s o corres pond t o a 12% i ncrease i n
alternating strain for short l i ve s and up t o a 31% i ncreas e in alternating strain at the l onger
l i ves .
d a / d N- AK
1. I ncreas e AK, h by a factor o f 1.1
2. De c r e as e the regi on II Paris coef f i ci ent C by a factor o f 5.
3. Retai n the s a me Pads e xpone nt n.

STEPHENS ET AL ON A THICK-WALLED PRESSURE VESSEL 149
These shifts will have the effect of increasing the fatigue crack growt h life by over a factor
of 5, depending on the closeness to AK, h.
Additional information concerni ng the formulation of the above decisions can be found
in Ref 19.
Student Questions
1. At this point, the class should prepare constructive criticism of the fatigue life calcu-
lation models and the kHz frequency correction factor decision.
2. What t est i ng/ speci mens/ st andards should be considered t o aid in obtaining the proper
loading equivalence?
e- N and S- N Fat i gue Life Equivalence Calculations and Experiments
Cal cul at i ons
Fatigue life calculations for the constant amplitude 0.2 Hz loading of Fig. 2 with the Lam6
hoop stress O'ma x ( ~ v m ) : 517 MPa were obtained based on the S- N approach (Eq 9) with
the Goodman mean stress model (Eq 13) and the e-N approach using the Morrow model
(Eq 10), and the SWT model (Eq 11), along with A723 material properties, given in Table
2. These calculations indicated greater than 107 cycles woul d exist for the S- N model and
the Morrow model. The SWT model gave 4.4 x 106 cycles. These values are shown in Table
3 for eventual comparisons with experimental values.
The number of blocks to failure with the resonant vibratory stress history of Fig. 3 for
dampi ng characteristics given in Table 1 and the Lam6 hoop stresses %~ak = 1034 MPa and
(Tyro -~- 517 MPa were calculated using the Lifest | software program (SoMat Corporation,
Champaign, I L 1988). This program incorporates rainflow counting, linear cumulative dam-
age, and both S- N and E-N approaches using Goodman, Morrow, or SWT mean stress models.
Calculations were made with and without frequency correction factors and the number of
calculated blocks are given in Table 4.
The equivalent number of calculated constant amplitude, R = 0 cycles were obtained by
dividing values of Table 3 by values of Table 4 for each model. Equivalent cycles with and
without frequency correction factors are given in Table 5. These equivalent calculated values
are very high, ranging f r om >2 x 10 3 to >2 x 10 4 cycles.
Experi ment s
The objective of the test program was to determine an experimental equivalence to com-
pare with the calculated equivalence. The tests involved smoot h un-notched axial 6- mm
TABLE 3--Const ant amplitude (R = O) life calculations based
on S-N and e-N analysis plus experimental results.
Experimental
Calculated Lives-Cycles Lives-Cycles
S-N, Goodman e-N, Morrow e-N, SWT
>10 7 >10 7 4.4 x 10 6 >10 7

TABLE 4--Variable amplitude life calculations based upon S-N and ~-N analysis plus experimental
results.
Lives-Blocks
Average
S-N, Goodman ~-N, Morrow ~-N, SWT Experimental
Without frequency effects 407 458 390 4076
With 15 kHz frequency 4840 1240 1700 14 000
effect
diameter cylindrical specimens machined from stress-relieved thick-walled A723 pressure
vessels. The longitudinal axis was in the direction of the pressure vessel hoop stress. After
machining, specimens were polished with 600 grit emery paper. Specimens satisfied Practice
for Strain-Controlled Fatigue Testing (ASTM E 606). All tests were conducted in load control
with R = 0 using an 89 kN closed loop electrohydraulic test system with hydraulic grips.
No high frequency (kHz) testing was done because of experimental limitations. Thus, kHz
frequency correction factors previously described were also incorporated into the experi-
mental equivalence.
Twelve constant amplitude R = 0 tests depicted in Fig. 2 were performed with O'ma x ranging
from 517 to 1 034 MPa. Tests were run between 0.2 and 25 Hz with no noticeable frequency
effect. The test results are plotted as an S- N diagram in Fig. 6. The key stress level for
equivalence calculations was O'ma x ( ( T y r o ) = 517 MPa where a run-out at 107 occurred.
Two variable amplitude tests with (rpe~k = 1 034 MPa and [~vm = 517 MPa as depicted in
the block history of Fig. 3 were run at 20 Hz. A brief rest period at zero load was added
between each variable amplitude block. The number of blocks to fracture for these duplicate
tests were 2 685 and 5 466, with an average of 4 076. Since no kHz frequency testing was
possible, the 20 Hz variable amplitude blocks were adjusted by an empirical correction factor
based upon averaging the Morrow and SWT e-N frequency results in Table 4, that is,
(1 240 + 1 700) + (458 + 390) = 3.36. This was used to form a pseudo-experimental
average block kHz life of (3.36) 9 (4 076) -~ 14 000.
Both constant and variable amplitude experimental lives or blocks have been included in
Tables 3 and 4 for better comparison to calculated values. The experimental equivalence has
been included in Table 5 again for better comparison to calculated equivalencies. It is seen
in Table 5 that the experimental equivalence values are significantly less than the calculated
values.
TABLE 5--Equivalent number of constant amplitude (R = O) cycles with and without frequency
effect.
Average
S-N, Goodman e-N, Morrow e-N, SWT Experimental
Without frequency effects >24 600 >21 800 >11 300 >2500
With 15 kHz frequency >2 100 >8 100 2 600 >700
effect

3 x 1 0 3.
Q .
.=
1 0 3 9
10 2
10
I I I I
3 10 4 10 5 10 6 10 7 10 8
c y c l e s
FIG. 6- - S- N diagram f or constant amplitude, R = O, smooth axial specimens, A723 steel.
Student Questions
1. What test replication i mprovement s are needed?
2. Why do calculated and experimental equivalence differ? Is this a significant difference?
3. Final frequency corrected equivalence ranged f r om >700 to >8100 cycles. Why such
large equivalence when the previous loading was considered 1 to 1?
LEFM Fatigue Crack Growth Life Equivalence
Experi ment al
Fatigue crack growth experimental equivalence was obtained by simulating the thick-
walled pressure vessel with an SE( T) specimen having a thickness, t, of 5 mm and a width,
W, of 25.4 ram. Specimens were machi ned from stress-relieved thick-walled A723 pressure
vessels. The longitudinal axis was in the direction of the pressure vessel hoop stress. The
initial crack size and nominal test stress ~r . . . . (%,,), was chosen to have an initial stress
intensity factor, K, of 27.5 MPaW' m. This value was chosen to be consistent with initial
values of K for crack sizes and hoop stresses found in many similar thick-walled cannon
tubes [20]. Thus, the initial stress intensity factors for <rp~k and Cram from Fig. 3 were 55 and
27.5 MPa~/-m, respectively. An initial crack length of 5.5 mm was chosen for both variable
and constant amplitude tests. Speci mens were machi ned with three grip holes at each end
to allow for a monobal l gripping fixture that woul d allow for bot h tension and compression
loads.
Regions I and I I R ~ 0 fatigue crack growth data were obtained using load shedding and
constant amplitude loading, respectively, as per Standard Test Met hod f or Measurement of
Fatigue Crack Growt h Rates ( ASTM E 647). Test frequency varied from 5 to 30 Hz, de-
pending upon da/ dN. Cracks were monitored with a traveling telescope using stroboscopic
lighting. The fatigue crack growth data were reduced using a secant met hod to d a / d N- AK
Copyr i ght by ASTM I nt ' l ( al l r i ght s r es er ved) ; Thu May 8 10: 42: 12 EDT 2014
Downl oaded/ pr i nt ed by
Uni ver s i dad De Chi l e pur s uant t o Li cens e Agr eement . No f ur t her r epr oduct i ons aut hor i zed.

and t he resul t s are shown in Fi g. 7. Very l i t t l e scat t er is seen. The regi on II Pads equat i on
obt ai ned by l og- l og l i near r egr essi on was
da
= C(AK)" = 4.93 x 10 -H (mg) 2"36 (16)
Thi s val ue was modi f i ed to
da C
d N 5
- - - - - (AK)" = 9.86 X 10 12 (AK)2.36
(17)
in order to i ncor por at e kHz frequency effect descr i bed earlier. The t hreshol d stress i nt ensi t y
factor, AK, h, was obt ai ned at 10 -1~ m/ c y c l e as per ASTM E 647 to be 4.3 MPa~' - ~. Thi s
val ue was i ncr eased by 10% to 4.7 MP a V~ for kHz frequency effect. The stress i nt ensi t y
at fracture, Kic, was t aken as 163 MPaV' -m, as seen i n Fi g. 7. The above fat i gue cr ack growt h
propert i es wer e used l at er for bot h const ant and vari abl e ampl i t ude fat i gue cr ack growt h l i fe
cal cul at i ons.
Three SE( T) speci mens were t est ed usi ng the Fi g. 3 vari abl e ampl i t ude l oadi ng wi t h t he
pr evi ousl y descr i bed i ni t i al cr ack l engt h of 5.5 mm and Kpe,k i ni t i al and Kv,, i ni t i al equal to
55 and 27.5 MPaV' -m, respect i vel y. The frequency was 20 Hz and t he l oadi ng hi st ory was
repeat ed unt i l fracture. The cr ack l engt h versus appl i ed bl ocks dat a are shown in Fi g. 8
where i t is seen that very l i t t l e scatter exi st ed in t hese t r i pl i cat ed tests. Cr ack l engt hs at
fracture wer e bet ween 11 and 12 mm and the bl ocks to fai l ure were 1 681, 1 754, and
1 767, whi ch yi el ds an average bl ocks to fract ure of 1 734. Thi s number obt ai ned at 20 Hz
wi l l be adj ust ed to a pseudo kHz number of bl ocks t o be used in equi val ence det ermi nat i on.
0
0
E
1 0 " 5
1 0 - 6
1 0 - 7
Z
10 -8
"0
1 0 " 9
t
t
1 0 - 1 0 ~ , . . . . . . , . . . . . . . . , . . . . . . . .
1 0 0 1 0 1 1 0 2 1 0 3
AK [MPa~/m]
FIG. 7- - Fat i gue crack growth rate behavi or f or ASTM A723 steel, R = 0.05.

13
11
7
!
~, Test 1
r A
9 Test 2 / j ~
/ r
5 n I ,
0 1000
B l o c k s
FIG. 8- - Crack length, a, versus applied variable amplitude blocks.
2000
Cal cul at i ons
Fat i gue cr ack gr owt h l i ves were cal cul at ed for bot h const ant and vari abl e ampl i t ude tests
in order to obt ai n t he needed si mul at ed crack gr owt h l i fe equi val ence. The commer ci al
Li f est | and the NASA- devel oped FLAGRO | ( NASA, Johnson Space Center, Houston, TX,
1989) soft ware pr ogr ams were used for bot h const ant and vari abl e ampl i t ude cal cul at i ons.
Nei t her pr ogr am i ncorporat es i nt eract i on effects on fat i gue crack growt h life, but si mpl y
sums i ncrement al cr ack ext ensi on cycl e by cycl e usi ng numeri cal i nt egrat i on. The Li fest |
pr ogr am uses the Pari s (Eqs 14 and 16) and the FLAGRO | program) uses a modi fi cat i on of
the For eman (Eq 15). Bot h pr ogr ams i ncorporat e AK,~ and Kjc. Li fest requires rai nfl ow
count i ng whi l e FLAGRO uses successi ve peaks and val l eys. In each case the cal cul at i ons
were made for t he SE( T) speci men wi t h a i = 5.5 mm, and initial stress i nt ensi t y factors of
27.5 and 55 MPa, respect i vel y, for cr~,~ and %e~. The numeri cal i nt egrat i on was carri ed out
unt i l fract ure at Kic = 163 MPa Vm.
The vari abl e ampl i t ude cal cul at i ons were obt ai ned wi t h and wi t hout kHz frequency cor-
rect i ons. The number of cal cul at ed vari abl e ampl i t ude bl ocks for bot h Li f est and FLAGRO
are gi ven in Table 6 al ong wi t h t he average exper i ment al results. The pseudo hi gh-frequency
exper i ment al results were obt ai ned by usi ng t he rat i o of the cal cul at ed bl ocks wi t h and
TABLE 6--Variable amplitude fatigue crack growth block
calculations plus experimental results f or simulated SE(T)
specimen.
Lives-Blocks
Average
LifEst FLAGRO Experimental
Without frequency effects 1 550 1 189 1 734
With frequency effects 7 770 5 435 8 360

TABLE 7--Constant amplitude (R = O) fatigue crack growth
life calculations f or simulated SE(T) specimen.
LifEst Cycles FLAGRO Cycles
18 700 17 728
wi t hout frequency effects as: (7 770 + 5 435) + (1 550 + 1 189) = 4.82. This rat i o was
mul t i pl i ed by the experi ment al average bl ocks, 1 734, to obt ai n t he pseudo exper i ment al
val ue of 8 360 bl ocks in Table 6. Cal cul at i ons and exper i ment al l i ves wer e very similar,
bei ng less t han a fact or of 2. The Li f est bl ock cal cul at i ons were sl i ght l y l arger than the
FLAGRO values.
Cal cul at ed const ant ampl i t ude, R = 0, fatigue cr ack growt h cycl es to fai l ure are gi ven in
Tabl e 7. Agai n, the Li fest val ue was sl i ght l y l arger t han t he FLAGRO value. The fatigue
cr ack growt h equi val ence was det er mi ned by the rat i o of the const ant ampl i t ude cycl es to
fai l ure to the vari abl e ampl i t ude bl ocks to failure. Equi val ence wi t h and wi t hout kHz fre-
quency effects are gi ven in Table 8. Her e i t i s seen that very little di fference in cal cul at ed
or average exper i ment al equi val ence occurred. The fat i gue crack growt h equi val ence for t he
frequency cor r ect ed SE(T) model was bet ween 2.3 and 3.3. These equi val ence values are
si gni fi cant l y less t han t hose obt ai ned usi ng the r or S- N model s.
Student Questions
1. Does it seem reasonabl e t hat t wo fatigue cr ack growt h soft ware pr ogr ams shoul d gi ve
di fferent resul t s? Is the percent age di fference si gni fi cant ?
2. Comment on the use of t he SE( T) speci men as used in the exper i ment al and cal cul at ed
equi val ence as regards to a t hi ck- wal l ed pressure vessel. What bet t er si mul at i on
model (s) do you propose and why?
3. Why does the equi val ence usi ng ~-N and S- N approaches di ffer so much from t hat of
the LEFM approach?
4. What final equi val ence is most r easonabl e for this pr obl em?
5. Is addi t i onal eval uat i on needed to form r easonabl e equi val ence? What do you rec-
ommend?
Discussion of Equivalence Results
The equi val ence values usi ng S- N or r model s and experi ment s were bounded by >700
and >8 100 or an order of magni t ude di fference. The equi val ence val ues for LEFM based
fat i gue cr ack growt h model s and experi ment s were bet ween 2.3 and 2.4, or a negl i gi bl e
TABLE 8--Equivalent number of constant amplitude (R = O)
fatigue crack growth cycles with and without frequency effect.
Average
LifEst FLAGRO Experimental
Without frequency effects 12.1 14.9 11
With frequency effects 2.4 3.3 2.3

difference. A major concern, however, was how to resolve/explain the two to three orders
of magnitude difference between the S-N or ~-N results and the LEFM results. The answer
is best obtained by realizing that the S-N or e-N results were based upon no stress concen-
trations, that is, K, = 1, at the muzzle location of the resonant frequency. Normally, fatigue
failures in these thick-walled pressure vessels occur in the rifling where significant stress
concentrations exist. Thus, these equivalence values can be altered by incorporating a stress
concentration factor, K,, or fatigue strength reduction factor, K~ This will drastically reduce
the S-N or ~-N equivalence values. The addition of a stress concentration will have only a
small influence on the LEFM fatigue crack growth equivalence.
References
[1] Stephens, R. i., Ed., Fatigue at Low Temperatures, ASTM STP 857, American Society for Testing
and Materials, Philadelphia, 1985.
[2] Fuchs, H. O. and Stephens, R. I., Metal Fatigue in Engineering, Wiley Interscience, New York,
1980.
[3] Rice, R. C., Ed., Fatigue Design Handbook, 2nd Edition SAE Publication AE-10, Society of Au-
tomotive Engineers, Warrendale, PA, 1988.
[4] Bannantine, J. A., Comer, J. J., and Handrock, J. L., Fundamentals of Metal Fatigue Analysis,
Prentice Hall, Englewood Cliffs, NJ, 1990.
[5] Dowling, N. E., Mechanical Behavior of Materials, Prentice Hall, Englewood Cliffs, NJ, 1993.
[6] Koh, S. K. and Stephens, R. I., "Fatigue Life Predication of an Autofrettaged Thick-Walled Pressure
Vessel with an External Groove," Journal of Pressure Vessel Technology, Vol. 13, Aug. 1991, pp.
368-374.
[7] Fackler, W. C., Equivalence Techniques for Vibration Testing, Naval Research Laboratory, Wash-
ington, DC, 1972.
[8] Collins, J. A., Failure of Materials in Mechanical Design-Analysis Predication Prevention, John
Wiley and Sons, New York, 1981.
[9] Osgood, C. C., Fatigue Design, John Wiley and Sons, New York, 1982.
[10] Kutt, T. V. and Bieniek, M. E, "Cumulative Damage and Fatigue Life Prediction," AIAA Journal,
Vol. 26, No. 2, 1988, pp. 213-219.
[11] Buch, A., Fatigue Strength Calculations, Trans. Tech. Publications, Aedermannsdorf, Switzerland,
1988.
[12] Wells, J. M., Buck, O., Roth, L. D., and Tien, J. K., Eds., Ultrasonic Fatigue, The Metallurgical
Society of the American Institute of Mining, Metallurgical and Petroleum Engineers (AIME), War-
rendale, PA, 1982.
[13] Willertz, L. E., "Ultrasonic Fatigue," International Metal Review, Vol. 25, No. 2, 1980, pp. 65-78.
[14] Puskar, A., The Use of High Intensity Ultrasonics, Elsevier, New York, 1982.
[15] Kuzmenko, V. A., "Fatigue of Structural Materials at High-frequency Cyclic Loading," Advances
in Fracture Research (Fracture '84), Proceedings of the 6th International Conference on Fracture,
ICFG, S. R. Valluri, D. M. R. Taplan, E Rama Rao, and J. K. Knott, Eds., Pergamon Press, 1984,
Vol. 3, pp. 1791-1798.
[16] Puskar, A. and Golovin, S. A., Fatigue in Materials: Cumulative Damage Processes, Elsevier, New
York, 1985.
[17] Puskar, A., "Fatigue Properties of Steels at Low and High Loading Frequencies," Ultrasonic Fa-
tigue, Wells, J. M., Ed., The Metallurgical Society of the American Institute of Mining, Metallur-
gical and Petroleum Engineers (AIME), Warrendale, PA, 1982, pp. 223-227.
[18] Puskar, A., "Fatigue Process Modifications at High Frequency Loading," Basic Mechanisms in
Fatigue of Metals, Elsevier, New York, 1988, pp. 331-338.
[19] Adams, T. B., "Cumulative Fatigue Damage of a Repeated High Frequency Vibratory Loading,"
M.S. Thesis, The University of Iowa, Iowa City, IA, May 1991.
[20] Underwood, J. H., "Fracture Damage and Failure of Cannon Components by Service Loading,"
Technical Report ARLCB-TR-83004, Feb. 1983.

T. Dani el L i ebster 1 and Gr zegor z Gl i nka I
Development of a Numerical Model
Predicting Fatigue Lives of Tubular
Threaded Connections
for
RE F E RE NCE : Liebster, T. D. and Glinka, G., " De v e l o p me n t o f a Nu me r i c a l Mo d e l f or
Pr e di c t i ng Fa t i g u e Li v e s o f Tubul ar Th r e a d e d Co nne c t i o ns , " Case Studies for Fatigue Ed-
ucation, ASTM STP 1250, Ralph I. Stephens, Ed., American Society for Testing and Materials,
Philadelphia, 1994, pp. 156-173.
ABS TRACT: Tubular threaded connections are used in the off-shore oil industry for appli-
cations such as tension leg platforms and drill string components. The design of these com-
ponents requires full-scale fatigue testing to ensure their reliability. This is an extremely ex-
pensive undertaking (actual connectors may be as large as 1.2 m in diameter). Therefore, an
analytical method is needed to provide preliminary estimates of how various parameters will
affect the fatigue lives of the connections. This model must have the ability to estimate fatigue
life under variable amplitude cyclic loading conditions. The key to such a model is relating
the applied variable amplitude loading history to the local elasto-plastic stress-strain response
at the roots of the threads. The load dependent nature of the stress concentration factor in pre-
loaded threaded connections requires the use of a numerical solution to determine the local
stresses. The development of this model is described starting with a discussion of the type of
loading and fatigue failure that must be modeled, the assumptions that are made in developing
the model, and the synthesis of the numerical solution and its application to predicting the
fatigue life of a real connector geometry. The fatigue crack initiation life is calculated using
the strain-life approach. The model is then used to demonstrate the effects on fatigue life that
result from changing various aspects of the connector design, such as geometry and load
configuration.
KE Y WOR D S : threads, fatigue, stress concentration factor, load distribution, finite element
analysis (FEA)
Thr eaded connect i ons refer to al l scr ew- nut syst ems, st uds i n a bl ock, or t hr eaded pi pe
j oi nt s. The charact eri st i c par amet er s of t hreads, such as di amet er , pi t ch, and t hread form, are
st andar di zed for mos t appl i cat i ons. However , t hr eaded j oi nt s become compl i cat ed st ruct ures
t hat pose di ffi cul t des i gn and anal ysi s pr obl ems whe n t hey are l arge i n size and subj ect to
maj or cycl i c stresses. Appl i cat i ons of t hi s nat ur e i ncl ude pr essur e vessel end cl osures,
t hr eaded j oi nt s for off-shore t ens i on l eg pl at f or ms (TLPs), dri l l st ri ngs, and ot her large di -
amet er connect i ons used i n t he oil i ndust ry.
Under s t andi ng how l arge t ubul ar t hr eaded connect i ons wi l l per f or m i n t er ms of fat i gue
can onl y be quant i f i ed by conduct i ng fat i gue exper i ment s on act ual component s . The t est
connect i ons mus t be subj ect ed to l oad hi st ori es t hat are compar abl e to t he act ual servi ce
l oads. For ver y l arge connect i ons , t hi s t ype of t est i ng is ext r emel y expensi ve and t i me con=
sumi ng. Therefore, i t i s desi r abl e to have anal yt i cal me a ns to est abl i sh t he expect ed i mpact
1 Graduate student and professor, respectively, University of Waterloo, Department of Mechanical
Engineering, Waterloo, Ontario, Canada, N2L 3G1.
156

LIEBSTER AND GLINKA ON DEVELOPMENT OF A NUMERICAL MODEL 157
that parameters such as connect or geometry, loading, number of engaged thread pitches, and
so forth woul d have on the fatigue life.
An analytical model would be useful for deciding what aspects should be explored in full-
scale fatigue tests. The information from such an analysis would also provide useful infor-
mat i on pertaining to the design of the experimental program.
Many studies of threads have addressed the role of specific thread geometries on the
magnitude of the stress concentrations [1-4], information that is important for a fatigue
analysis. However, most of the information pertaining to whole threaded assemblies are of
an empirical nature, determined under static loading conditions, rather than a systematic
analytical or numerical analysis. For fatigue analysis of threads, it is necessary to find the
answers to such questions as:
1. When the threaded assembly is subject to cyclic traction, how are the reactions between
the different threads distributed?
2. Where are the regions with the maxi mum local stress amplitudes?
3. What are the stress concentrations?
Direct calculation of stresses by numerical model i ng of a threaded assembly does not in
itself pose any special technical problems; however, it is extremely cumbersome and expen-
sive for large assemblies, even for monot oni c loading. In the case of cyclic, variable ampli-
tude loading that induces local nonlinearities, such calculations are practically not feasible.
These difficulties are magnified for multiple repetitions of calculations, such as is the case
with parametric studies.
St a t e me nt o f t he Pr o bl e m
This case study examines the development of a fatigue life prediction model that was
required by a large petroleum company for the design of large tubular threaded connections
in off-shore structures. The probl em was to develop an analytical model that could be used
for screening connect or designs and establishing whi ch should be built for evaluation by
full-scale testing. The model needed to be efficient, yet encompass enough of the parameters
associated with fatigue so that the calculated results were reasonable within the scope of
existing fatigue life prediction methods. The relative effects of various geometric and load
parameters on fatigue crack initiation could then be evaluated to determine which aspects of
the design should be modified to gain the greatest benefit in terms of i mproved fatigue life.
A model that serves the purpose stated above should:
9 predict the loads on the threads, since ultimately it is the loads that cause the stresses;
9 predict the stresses at the location where the fatigue failure is expected to occur; and
9 have the ability to handle a variety of geometries, materials, and loading histories.
Ge ne r al Appr o a c h t o t he Pr o bl e m
The solution to this problem must consider the general characteristics of threaded assem-
blies. The threads, which may be geometrically and mechanically complex, are a small
component of a mechanical system that connects t wo large diameter pipes. Each thread pitch
is reproduced in the assembly many times. Therefore, the solution uses onl y a small, detailed
portion of the assembly (that is, a few thread pitches) t o characterize the load and stress
parameters that are affected onl y by the thread geometry. The local model is multiplied, as

necessary, to solve the large-scale problem of the full assembly, where the physical pa-
rameters of the complete assembly are taken f r om a numerical solution of the local model.
In general, the calculation process determines the macro properties of the threaded assem-
bly from local data obtained independently of the loading on the complete assembly. The
macro model is easily solved for different loads, making it possible to return to the local
level at any critical point in the threaded assembly, where the resulting stresses and strains
in the thread root can be obtained.
Background Information
I n developing any engineering simulation of a mechanical component , it is necessary to
ask questions whose answers will provide the best understanding of how the component
behaves in actual service. This ensures that the model will be as realistic as possible. The
following questions must be answered to produce a model that encompasses the appropriate
parameters for the fatigue life calculation of a large diameter, tubular threaded connection,
used in a TLP application.
What is the General Geomet ry and Configuration o f TLP Tether Connect i ons?
The typical configuration of a tether connect i on for a TLP is illustrated in Fig. 1. These
threaded connections consist of an externally threaded pipe, called the pin, and an internally
threaded pipe, called the box. For bot h component s, the free end of the threaded length is
called the toe, and the end of the threads closest to the pipe is called the heel. The external
diameter of the box may be as large as 1.2 m. In many cases, the threads ar e tapered, as
shown in Fig. 1, to facilitate ease of assembly. To prevent seawater from entering the pipe
and the threads during service, the end of the box is tightened against the shoulder on the
pin. This initial tightening is called preloading, or pretorquing.
What is the Loading on Tether Connect i ons?
There are three sources of load that result in stresses at the roots of the threads: the
externally applied load that the connection was designed to carry, the internally generated
load that results from preloading the connection, and thermal loading that is caused by
differences in temperature throughout the connection. For this analysis, thermal loads will
not be considered. Preloading the connection results in an initial static loading condition that
must be accounted for in the fatigue analysis.
In this case, the preload is applied at the end of the box, against the shoulder on the pin
(Fig. 1). The connection is assembled, and the t wo component s are screwed together until
the box end is tightened against the preload shoulder, the same idea as "finger tightening"
Preload Shoulder
\ ~ Pin
FIG. 1--Typical tether connection f or a tension leg platform application.

a bolt and nut assembly. Then, the two halves of the connect i on are further tightened by a
specified torque. The resulting interference at the preload shoulder causes the box of the
connection to be nomi nal l y placed in compression, and the pin to go into tension. The
preload, as defined in this paper, is the magnitude of the resulting force at the preload
shoulder.
The loading on the legs of a TLP is axial. An example of a TLP dynamic load history
that has been normalized with respect to the maxi mum load is shown in Fig. 2. Since TLP' s
are semi-submerged structures that are held in place by the tethers, the upwards buoyant
force causes a load history that cycles about a mean tensile load. The load fluctuations are
caused by environmental loading, such as wi nd and waves. The nature of the l oad history
depends considerably on the geographic location of the drilling platform. For example, plat-
forms that are located in the North Sea experience more severe loading spectra than those
located in the Gul f of Mexico. Load histories for this t ype of fatigue analysis must be
obtained either by direct measurement or by careful simulation.
What is the Mai n Mechani sm and Locat i on f o r Fatigue Failure o f Tubular Threaded
Connect i ons ?
As is the case for most types of threaded connections, tubular connections usually fail due
to cracks that form in the roots of the threads. This location is most susceptible to fatigue
failure because of the stress concentrations associated with the notch-like geomet ry of the
thread roots.
As a result of the non-uni form load distribution along the engaged length of threads, the
magnitude of the peak stress varies from thread pitch to thread pitch. Fatigue failures gen-
erally occur in the root of the thread with the highest cycl i c stresses. This thread, termed
the critical thread, is usually the one that carries the highest static load.
1.0
o.B
0.6
o
i 0.4
Z
0.2
0.0
' I ' I ' I ' I
20 40 60 80
P o i n t N u m b e r
FIG. 2--Example of a normalized load history f or a tension leg platform.

What Type of Fatigue Analysis is Appropriate?
Since the fatigue analysis for this probl em involves notches, variable amplitude loading,
and prestressed component s, the strain-life method has been chosen for this case. This
met hod accounts for any notch root plasticity as well as any variations of the local mean or
residual stresses. The strain-life method requires the notch root strain history, which may be
obtained from the local elastic stress history by using either Neuber' s rule [5], or the equiv-
alent strain energy density (ESED) met hod [6].
Summary of Requirements
Based on the above information, the model must fulfill the following requirements:
1. it must predict the distribution of l oad in a threaded connection for different levels of
external load and different amounts of initial preload, and
2. it must be able to predict the elastic stresses in the thread roots that result from those
loads.
By meeting the above two requirements, standard strain-life analysis methods for notches
can be used to estimate the fatigue life of the connection. Of course, this leads to some basic
assumptions and limitations to the analysis as well. The l oad on the threaded components
must always remain nominally elastic; plasticity in the notch root must be very localized. It
is assumed that small amounts of localized plasticity will not significantly affect the elastic
behavior of the connection.
Development of the Model
There are four steps needed to simulate the behavior of the threaded connection: deriving
the elastic stress concentration factor based on the thread loads, deriving the thread load
distribution in the connection, determining the local elasto-plastic strain response in the
thread roots, and then calculating the fatigue initiation life using the Manson-Coffin strain-
life curve for the material.
The fatigue analysis model for this case is developed for a tubular threaded connection.
The geomet ry of the connection is the same as that studied with photoelastic experiments
by Broadbent and Fessler [7,8], providing a means of compari ng the calculated load distri-
bution and stress analysis results to actual experimental values. This aids in establishing that
the model correctly predicts the behavi or of the connection.
Parametric Study of a Tubular Threaded Connection
The threaded tubular connection to be considered for fatigue testing is shown in Fig. 3.
It is required to establish the expected effects that the fol l owi ng design changes would have
on fatigue life:
1. increasing the level of preload f r om zero, and
2. increasing the length (number of engaged threads) of the connection while maintaining
a constant preload.
The connection is to be fabricated from 4340 steel; the mechanical properties are given
in Table 1.

[ 20 mm
FIG. 3--Tubular threaded connection analyzed in this case study.
260 mw
Der i vi ng t he St ress Concent rat i on Fact or
Stress concent rat i on factor, K,, is defi ned for a component as t he rat i o of the l ocal peak
stress to the nomi nal stress. For t ubul ar t hreaded connect i ons, it is common for the peak
stress to be t aken at the r oot of t he thread, and t he nomi nal stress to be the average stress
in the pi pe wal l . However, this defi ni t i on of stress concent rat i on fact or is not ri gorous enough
si nce it does not consi der any i nt ernal l oadi ng due to prel oad.
A compl et e defi ni t i on of the stress concent r at i on fact or must account for all of the l oads
act i ng on the structure. Consi der t he case of a t ubul ar t hreaded connect i on that is not l oaded
ext ernal l y, but subj ect to a shoul der pr el oad due to pret orqui ng. The nomi nal stress in the
pi pe sect i on due to ext ernal l oad is equal to zer o whi l e cl earl y t here are non-zero st resses
in the mat eri al wi t hi n the connector. This l oad confi gurat i on results i n an infinite val ue for
K,, whi ch does not provi de a useful represent at i on of the stresses in t he thread root.
The first st ep in devel opi ng the stress concent rat i on factor is to cl ear l y define the l oads at
each t hread root. Thi s can be char act er i zed by i nt roduci ng the l oad rat i o, m, whi ch is the
rat i o of t he l oad in the cr oss- sect i on at the t hr ead r oot to the l oad on t he t hread
where
_ P . . . . i ( 1 )
mi -- Pi~
TABLE 1--Mechanical properties of 4340 steel.
Elastic modulus, E
Cyclic strength coefficient, K'
Cyclic strain hardening exponent, n'
Fatigue strength coefficient, ~
Fatigue strength exponent, b
Fatigue ductility coefficient, e;.
Fatigue ductility exponent, c
Transition fatigue life, 2N t
Yield strength, %,
Tensile strength, S,
True fracture strength, ~f
True fracture ductility, Cy
208.9 GPa
1760.9 MPa
0.146
1713.3 MPa
-0. 095
0.83
-0. 65
4100 cycles
1103.2 MPa
1172.1 MPa
1634.0 MPa
0.83

rn~ = the load ratio at thread i,
P . . . . ~ = the load on the cross section at the root of thread i, and
P~ = the load acting on thread i.
These are the two modes of loading that contribute to the total stress in the root of the
thread. One component of the total stress is caused by the tensile load acting on the notched
body, P.o,.,i, and the other results from the local bending load, Pi, on the thread. Consider
these two load cases, as illustrated in Fig. 4. A segment of the threaded connection, shown
in Fig. 4a, has a membrane load, P,., applied across its section. The resulting maximum
principal stress component at any given axial position, x, along the surface of the thread root
is gin(x). Therefore, a corresponding normalized stress distribution, K,.(x), may be defined as
K m ( x ) - I~m(X) (2)
Sm
where
K,.(x) = the normalized membrane stress distribution on the surface of the thread root,
cr,.(x) = the maximum principal stress component in the thread root due to the membrane
load, and
Sm = the nominal stress in the thread root section due to the applied membrane load.
The nominal stress in the root section, Sin, is the average stress on the cross-sectional area,
A,
~'--X
(a)
k-------X
A
I
( b )
FIG. 4--Schematic diagram of the two loading modes; (a) membrane load, and (b) thread bending
load.

Sm Pm
= ~ (31
Si mi l arl y, Fi g. 4b shows the same segment of t hreads wi t h one t hr ead l oad, Pb, appl i ed.
For this case, the pr i nci pal stress di st ri but i on, %(x), may be expr essed as a nor mal i zed stress
di st ri but i on, Kh(x), by usi ng the nomi nal section stress caused by t he t hr ead l oad
Ko( x ) - ~rb(x)
Sb (4)
where
Kb(X) = nor mal i zed stress di st ri but i on as a funct i on of posi t i on, x, that results from the
appl i ed bendi ng l oad;
%(x) = the di st ri but i on of the maxi mum pri nci pal stress component ; and
Sb = the nomi nal stress in t he t hread root sect i on due to the appl i ed t hread load.
In this case, the average stress in t he t hread root sect i on, Sb, is caused by the l oad act i ng on
the t hread and is defi ned by
Sb ~- Pb
~- ( 5 )
The concept of t he b e n d i n g stress di st ri but i on, used in Eq 4, is a not i onal one because
the l oad act i ng on the one t hread pi t ch does creat e bot h bendi ng and membr ane stresses.
For a t hreaded component that is nomi nal l y l oaded in tension, it is al ways the case that m
-> 1. By the defi ni t i on of the l oad rat i o, Fig. 4b shows the case for m = 1.
For an el ast i c stress anal ysi s, t he pr i nci pl e of superposi t i on may be used to det ermi ne the
t ot al stress in the t hread root. The di st ri but i on of t ot al stress in t he t hread root, (r(x), may
be expressed as the sum of the membr ane and bendi ng stress di st ri but i ons
or(x) = % ( x ) + %( x ) (6)
Each stress di st ri but i on t erm in Eq 6 can be expr essed by the nomi nal stress and the
nor mal i zed pri nci pal stress di st ri but i ons gi ven by Eqs 2 and 4 such that
S K, ( x ) = SbKb(X) + S,~Km(X) (7)
Subst i t ut i ng for the nomi nal stresses wi t h the appl i ed l oads gi ves
( P, . + Pb) Kt ( x ) = PbKo(x____~ ) + P,~Km(X)
(8)
A A A
that can be si mpl i fi ed by mul t i pl yi ng bot h sides by the cross sect i onal area, A, and di vi di ng
by P, . + Pb
1
K, ( x ) - - - [PbKo(x) + P, , K, , ( x ) ] (9)
Pm + Pb

Fi nal l y, the l oad rat i o m for this superposi t i on of t he l oad cases in Fi gs. 4a and b, is gi ven
by
em + Pb
m = - - (10)
Pb
S u b s t i t u t i n g Pb in t erms of Pm and m i nt o Eq 9 gi ves t he rel at i onshi p bet ween K, and m
K,(x) = 1 [Ks(x) _ Kin(x)] + gm(X ) (11)
m
In Eq 11, Kb(X ) and K,,(x) are i ndependent of t he appl i ed l oad, and depend onl y on t he
t hr ead geomet ry. However, t he t ot al nor mal i zed stress di st ri but i on, K,(x), depends on t he
current di st ri but i on of l oad in t he connect i on.
The stress di st ri but i ons in Eq 11 can be obt ai ned f r om a smal l l ocal model of four t hreau
pi t ches and t wo di fferent l oad cases, shown in Fi g. 5. An axi symmet r i c finite el ement model
is used to r epr esent t he t hreaded connect i on. Thi s means that the t hr ead hel i x is model ed by
a seri es of ri ngs on t he cyl i ndr i cal body. Thi s si mpl i fi cat i on is a gener al l y accept ed assump-
t i on for t hread anal ysi s that is sui t abl e when the di amet er of the threads is l arge compar ed
to t he pi t ch and t he hel i x angle. The first l oad case, shown in Fi g. 5a, is a t ensi l e membr ane
l oad appl i ed evenl y across the face of the model . The stress di st ri but i on al ong the surface
of t he t hread root , nor mal i zed wi t h r espect to the aver age stress in t he sect i on for this l oad
P
I l l l l l l l ~ l
I ~ ~ I I I I I I I I I I I I I I I I IV' I-X-K/ l q ~
(a)
r
b
z
P" P, P2 P2
(b)
FIG. 5--Axisymmetric finite element models of four pitches from the pin of the connector; (a)
membrane load condition, and (b) thread bending load condition.

case, gi ves t he Kin(x) curve for this t hr ead r oot geomet ry. The second l oad case, shown in
Fi g. 5b, is l oadi ng appl i ed to t he f ace of each t hr ead in the model . The magni t ude of the
axi al component of each l oad i s the same for al l of t he threads. Not e t hat this l oad case
does not r epr esent t he bendi ng l oad case used for Eq 4. I t pr ovi des t he nor mal i zed t ot al
stress di st ri but i on, K,(x), for t he speci fi c case of m = 2, whi ch wi l l be referred t o as K: (x).
Even t hough Kb(x) cannot be det er mi ned di r ect l y f r om ei t her of t hese t wo l oad cases, t hey
pr ovi de enough i nformat i on to sol ve for K, (x) by usi ng Eq 11. Thus, the general expr essi on
for K,(x), der i ved f r om the numer i cal solution, can be expressed as
K,(x) = 1 [2K:(x) + (m - 2)K.,(x)] (12)
m
The stress concent rat i on factor, Kt, is the maxi mum val ue of t he K,(x) di st ri but i on for a
gi ven val ue of t he l oad ratio, m. The nor mal i zed stress curves obt ai ned f r om each l oad case
are shown in Fi g. 6, and Fi g. 7 present s K, as a funct i on of the l oad rat i o, m.
Br oadbent and Fessl er [8] pr esent ed nor mal i zed peak stresses f r om t hei r phot oel ast i c anal -
ysi s of the connector. In Fig. 8, t hese are compar ed t o t he peak stresses t hat were pr edi ct ed
wi t h the stress concent rat i on fact or rel at i onshi p (Eq 12), and nor mal i zed by usi ng the same
met hod as Br oadbent and Fessler. The l oad rat i o val ues were cal cul at ed from the pr edi ct ed
t hread l oad di st ri but i on, whi ch is der i ved in t he next section. It is i mpor t ant to underst and
t hat t he ends of t he t hread hel i x i n t he phot oel ast i c model are const ruct ed wi t h a t hread run-
out. Run-out is a decrease in t he t hr ead wi dt h to zero (l i ke a wedge) over one hal f of a
pitch. The effect of this is a great l y r educed t hread stiffness that causes t he thread l oads and
stresses to i ncr ease at the ends of t he engaged length. In pract i ce, unl i ke t he t hreads model ed
by Br oadbent and Fessl er, the run-out is r emoved so that onl y ful l y f or med t hreads carry
any load.
1 0 . 0 - -
~ 8.0
"~ e . 0
,t
4.0
i
~ 8,0
0
Z
0.0
o
o K ~ ( x )
o
o K2(~)
o
o
a
o
o
o o o o o 8 o o
o o o o o
o o
o o
o o o o _ B F j
' I ' I ' I ' I P I
0 , 0 0 . 2 0 . 4 0 . 6 0 . 8 1 . 0
Normalized Axial Position in the
Thread Root
FIG. 6--Normalized thread root surface stresses obtained from the finite element solution.

1 6 . 0 - -
1 6 . 0
~ ,
14.0
0~ 12.0
10.0
- I~nadantm~ of
8.0 ua on (12)
o
~ -
~.~ 6.0
~ -
~ 4.0
2 . 0
0 . 0
' i ' I ' I ~ I ' I ' I
0.0 2.0 4. 0 6.0 8.0 10.0 12.0
Load Ratio, m
F I G . 7- - St r es s concentration f act or as a f unct i on o f the l oad ratio.
L o a d Di s t r i but i on An a l y s i s
It has been shown t hat the l ocal el ast i c stress concent rat i on for the t hread root s can be
cal cul at ed once t he l oad ratio, m, is known. Therefore, t he second part of t he model must
al l ow the l oad di st ri but i on to be cal cul at ed so that t he i ndi vi dual t hread l oads can be found.
Ther e have been many anal yt i cal sol ut i ons to this pr obl em; however, the met hod pr oposed
8 - -
o~ 6
t,O
F~
4
r ~
ed
z
Z
I
[
0
Ca l c ul a t e d Pe a k St r e s s
" l " Br oa dbe nt & Fe s s l e r
~-~ Thread
[ Run-out
I o o < > o
I <> / ~ L
I I
ii._ )
I I
I
' 1 ' I I I '
I
1 2: 5 6
' I ' I '
3 4
Thr e a d Pi t c h
FIG. 8- - Nor mal i z ed t hread root stresses calculated f r om Eq 12 compared to the photo-elastic results
of Broadbent and Fessler f rom Re f 8.

by several researchers in t he past [9-11] is best sui t ed for comput er fat i gue cal cul at i ons.
Thi s met hod uses a l i near spri ng net wor k anal ogue to r epr esent t he deformat i on of t he
t hr eaded connect i on. The basi c concept of t he anal ogue model is to separat e t he di fferent
component s of def or mat i on in t he t hr eaded connect i on i nt o t hose t hat can be represent ed by
usi ng si mpl e anal yt i cal techniques. The component def or mat i ons are t hen mat hemat i cal l y
r eassembl ed so t hat t he t ot al def or med confi gurat i on may be det er mi ned by sat i sfyi ng force
equi l i bri um.
The el ast i c def or mat i ons are model ed by l i near spri ng el ement s. Fi gur e 9 shows spri ng
el ement s super i mposed ont o a schemat i c of a connect i on wi t h t hree engaged t hread pi t ches
so that t he physi cal meani ng of each el ement may be i l l ust rat ed. The t wo modes of defor-
mat i on consi dered are t he axi al def or mat i ons of the cyl i ndr i cal core sect i ons in the box and
pin, and the l umped shear and bendi ng def or mat i on of each t hread pi t ch on an el ast i c support.
For bendi ng and shear, one spri ng el ement is used to r epr esent the t ot al deformat i on of t wo
mat i ng threads. Al l of t he el ement s in t he model are assumed to be r i gi dl y connect ed so that
onl y one- di mensi onal axi al def or mat i on is consi dered.
The stiffness val ues for the spri ngs in t he net work are obt ai ned by usi ng anal yt i cal and
numer i cal met hods. The axi al stiffness val ues are cal cul at ed anal yt i cal l y wi t h the f ol l owi ng
equat i on
AE
k A=- -
P
where
kA = t he axi al stiffness,
A = t he cross sect i onal area at t he r oot of the thread,
E = Young' s modul us, and
p = t he t hread pitch.
(13)
The bendi ng stiffness val ues are obt ai ned f r om the same finite el ement model that was used
to det er mi ne t he stress concent rat i on fact or for the t hread roots. For t he l oad case where al l
of the t hreads in t he smal l model are l oaded, the di spl acement s of nodes on t he pi t ch l i ne
of t he t hreads are used in conj unct i on wi t h the axi al stiffness values cal cul at ed using Eq 13
to sol ve for the val ues of ks. This is i l l ust rat ed in Fi g. 10. For exampl e, t he value of kB3 is
easi l y found f r om t he nodal di spl acement u3, and the known forces and axi al stiffness val ues
by
F 2 F 3F 4 F
u3 = ~ + ~A3 + ~AZ + kA---~ (14)
P
m <
k i ; i ; i : i ; P
9 ?i2i2~'2">?i '2'>i'i'i"i"i'i'i'i'i'i'i')i%i.i-)i.>i
(.i))))
22~2~2k`~22~2~2~2~22~2~P~222~2~222~i~2~2~222~2~22~i2~)~22~2i~2~2~2~2~222~
FIG. 9--Schematic of a connector with three engaged thread pitches and the corresponding spring
element model.

12 3
u 4
m
_
k B 2 k B 4
k A l k A 2 k A 3 k A 4
FIG. l O--Spring network used f or calculating thread bending stiffness from the finite element results.
The load distribution in the connection is obtained by solving a matrix solution for static
equilibrium of the spring net work for the entire connection, with the appropriate load and
displacement conditions. For the connection anal yzed experimentally by Broadbent and Fes-
sler [7], the load distribution is given in Fig. 11. It is compared t o the load distribution
calculated by using the spring network analogue. The calculated results are presented for a
connection with six engaged thread pitches. The photoelastic results of Broadbent and Fessler
were for a connection having between six and seven engaged threads including the thread
run-out. (See the discussion on run-out in the previous section.) Their results were obtained
by sectioning the 3D model at 90-degree intervals and summing the shear stresses across
the base of the thread to obtain the average load at that point. Therefore, there are up to
20- -
1.6
Z
0.8
0.4
( ~ B r o a d b e n t & F e s s l e r D a t a
S p r i n E M o d e l , 6 Threads
o
o
~ 1 7 6 0 o
o o o o o o
0. 0 0. 2 0. 4 0, 6 0. 8 1.O
Nor ma l i z e d Pos i t i on
FIG. 11---Analogue model solutions f or the load distribution in a connector with six engaged thread
pitches compared to the photo-elastic results of Broadbent and Fessler f rom Ref 7.

four data points per thread pitch around the thread helix. The spring de me nt model can onl y
represent entirely engaged, fully formed thread pitches; the calculated loads represent the
average load over one thread pitch. Results for a connect or with onl y a preload applied to
it are presented in Fig. 12.
Fatigue Life Calculations for the Parametric Study
For every case in the parametric study, the local elastic stress history was calculated for
each peak of the loading history, shown in Fig. 2. A maxi mum load 6783.5 kN was used
for all of the fatigue calculations. This load represents a nominal stress in the heel of the
pin that is approximately 50% of the yield stress of the material. Not e that in all cases, the
maxi mum applied load exceeds the magni t ude of the preload. For tubular threaded connec-
tions, the load at the preload shoulder may be lower than the maxi mum applied load. This
is different from the design requirements of bolted joints for which the effect of the preload
is diminished by exceeding the initial cl ampi ng load; this occurs in direct proportion to the
applied load. In tubular threaded connections, as external load is applied the load at the
pr doa d shoulder decreases at a rate that is lower than that of the increasing applied load.
The effect of preload in tubular connections is diminished by reducing the load at the preload
shoulder to zero.
In tubular threaded connections, the magni t ude of the applied load that is necessary to
cause the preload shoulder load to go to zero is termed the critical load; this load may be
many times greater than the initial preload applied to the connection. For this example, the
critical load is approximately eight times the magnitude of the preload.
The local elasto-plastic strain history was determined by the application of Neuber' s rule
[5] and the hysteresis l oop was obtained by doubling the cycl i c stress-strain curve. The state
of stress on the surface of the thread roots is biaxial, with stress component s in the axial
and hoop directions.
3 . 0 - -
o 2 . 0 -
o
Z
0 . 0
Br o a dbe nl ; & Fea~[et- i v
i
s ~ uod~. 7 Th r ~ L, ]
Io
I
I
l - - Run- out
v + [
+ o [
I
I
I I I [ I I I [
~. 2 3 4 5 6 7 fl
T h r e a d N u m b e r
FIG. 12--Analogue model solution of the load distribution for preload only compared to the photo-
elastic results of Broadbent and Fessler from Ref 7.

Therefore, a plane stress assumption is not appropriate for this analysis. Plane strain as-
sumptions are better for this application since t hey more closely approximate the actual stress
state in the thread roots. Al t hough the strain in the hoop direction is not equal t o zero, it is
small compared to the strains in the axial direction.
Plane strain was assumed for this analysis and applied with Neuber' s rule and the plane-
strain modification to the stress-strain curve suggested by Dowl i ng et al. [12]. Neuber' s rule
is usually presented with the term K, AS for the change in elastic stress. It is important to
note that in preloaded tubular threaded connections, K, varies nonlinearly with applied load.
Therefore, Neuber' s rule must be used in the following form
( a a ) 2
- cr~ ( 1 5 )
E
where Air is the total change of the local elastic peak stress f or one load reversal.
The local elasto-plastic strain history was used to determine the fatigue damage from the
strain-life curve (the parameters are given in Table 1) with Mor r ow' s correction for mean
stress [13]
A ~ Or} - - or m
-2 =" E (Ns)b + e}(N:)~
(16)
where o- m is the mean stress.
~ Rainflow counted cycles were used with a linear damage cumulation rule (Miner' s rule)
to estimate the expected fatigue lives.
Results
The effect of preload on the fatigue life of the connect or is illustrated in Fig. 13. Absolute
values of predicted life are purposely left out to emphasize that this model is intended to
estimate the relative effect of parameters on fatigue as an aid to designing actual fatigue
tests. Only after actual testing is conducted, could the calculated fatigue life predictions be
quot ed as absolute values.
Figure 13 shows that increasing the preload protects the threads on the pin by decreasing
the amplitude of the cyclic stresses in the root of the thread (Fig. 14). This is effective up
to a certain level at which the increasing mean stress predominates the fatigue damage and
the total life starts to drop of f again.
Figure 15 illustrates the effect of increasing the number of threads in the connection. It
can be seen that adding more threads results in higher fatigue lives, but with diminishing
returns. Increasing the number of threads reduces the overall load carried by each thread
pitch. However, the amount of load reduction also decreases until adding more threads does
not yield any significant benefit.
S u m m a r y a n d C o n c l u s i o n s
The key aspects of the solution technique may be summari zed as follows:
1. An axisyrnmetric finite element model of a small segment of the threaded connection
is solved for t wo different load cases: tensile membrane loading and direct thread

2.0
1.8
0
O~ 1. 2
0.8
0.4
0.0
o
o
0
o
)o
I
0.00
I
q~
o
o o
o
0
Magnltude of preload
for whlch the crl t ~s l load
is OqU~ tO t,,hS ma ~mUl n
append loa~
I ' I ' I ' [
1. 00 2. 00 3. 00 4. 00
P r e l o a d (kiN)
FIG. 13--Ratio of the calculated fatigue life at the root of the critical thread to the calculated fatigue
life for a connector with no preload. This ratio is plotted as a function of the preload level.
loading. The results are used to obtain the stress concentration factor relationship and
the bending stiffness of the threads.
2. The axial and bending stiffness values are used in a spring net work model of the entire
connection to determine the thread load distribution as a function of the applied load
and the preload.
3. For each peak in the applied load history, the load ratio at the thread of interest is
calculated and used to find the stress concentration factor. The local elastic stress history
is derived from the load history.
4. Not ch strain analysis such as Neuber' s rule or the ESED met hod are used to generate
the local elasto-plastic strain history at the thread root.
5. An appropriate cycle count i ng and damage hypothesis are applied with the strain-life
curve t o calculate the fatigue life of the threaded connection.
This efficient method for estimating the fatigue life of tubular threaded connections pro-
vides a convenient means for estimating the i mpact of design parameter changes on the
overall fatigue performance of the connection. Understanding the behavi or of the connections
is essential in designing meaningful fatigue tests for full-scale connections. The results of
the model are subject to the same caveats and limitations of the strain-life analysis that is
used. Use of the model for estimating the actual service life of a large threaded connection
is not valid without actual experimental tests to verify the results from the model.

o
t~
g
,-.q
2. 0E+9 - -
1. 6E+9
1. 2E+9
8 . 0 E + 8
4. 0E+8
O. OE+O
[ ] Pr e l oa d - 4. 0 MN
~. Pr e l oa d - 0. 8 IdN
- O -' Pr e l oa d - 0. 0 MN
6-
0 I0 2:0
Poi nt Numbe r
V
!
30
FIG. 14- - Local elastic stress history at the root o f the critical thread f or three different levels o f
prel oad (only the f i rst 30 poi nt s are pl ot t ed f or clarity).
2 . 4 - -
0
22- -
2. 0- -
1. 8- -
t . 6- -
1. 4- -
L. 2- -
L O
08
0.6
I I I I I I I q
6 8 10 12 14 16 18 20
Numbe r of Enga ge d Thr e a d Pi t c he s
FIG. 15- - Rat i o of the calculated f at i gue life at the root of the critical thread to the calculated f at i gue
life f or a connection with six engaged thread pitches. The ratio is pl ot t ed as a f unct i on of the thread
length (in number of pitches). The prel oad is 1.0 MN in al l cases.

Student Questions
1. There are many different ways that this model can be made more or less conservative.
List the aspects of the model that make it conservative. Suggest ways to make the
model less conservative, (and, therefore, "more accurate").
2. Could this model be used to model other types of threaded connections, such as nut /
bolt connections, or the end closures to pressure vessels? Which assumptions used in
the development of this model would need to be re-examined if the model were to be
applied to other situations?
3. Give possible reasons why even with a pin protecting preload that leaves the threads
at the heel of the box completely unprotected and subject to the full amplitude of the
applied loading history, that connections still fail predominantly at the first thread pitch
in the pin.
4. While the model indicates that the optimum preload would provide a critical load equal
to the maximum applied load, explain why this is not possible in practice. What would
be a suitable preload for this connection, assuming that the calculated fatigue life were
verified in testing?
5. If at first, cracks were to form and start growing in a thread root that was in the middle
or towards the toe of the pin, explain why the connection would be likely to fall at the
root of the first thread in the pin anyway?
References
[1] Kenny, B. and Patterson, E. A., "Load and Stress Distribution in Screw Thread," Experimental
Mechanics, Vol. 25, No. 3, 1985, pp. 205-213.
[2] Seika, M., Sasaki, S., and Hosono, K., "Measurement of Stress Concentrations in Threaded Con-
nections," Bulletin of the JSME, Vol. 17, No. 111, 1974, pp. 1151-1156.
[3] Brown, A. F. C. and Hickson, V. M., "A Photo-elastic Study of Stresses in Screw Threads,"
Proceedings of the Institution of Mechanical Engineers, Institution of Mechanical Engineers, Lon-
don, 1952, pp. 605-612.
[4] Hetenyi, M., "The Distribution of Stress in Threaded Connections," Experimental Stress Analysis,
Vol. 1, No. 1, 1943, pp. 147-156.
[5] Neuber, H., "Theory of Stress Concentration for Shear Strained Prismatical Bodies with Arbitrary
Non-Linear Stress Strain Law," Journal of Applied Mechanics, Dec. 1961, pp. 544-550.
[6] Glinka, G., "Energy Density Approach to Calculation of Inelastic Stress-Strain Near Notches and
Cracks," Engineering Fracture Mechanics, Vol. 22, No. 3, 1985, pp. 485-508.
[7] Broadbent, T. P. and Fessler, H., "Distribution of Thread Load in Screwed Tubular Connections,"
Fatigue of Offshore Structures, W. D. Dover and G. Glinka, Eds., Engineering Materials Advisory
Service (EMAS), Cradley Heath, England, 1988.
[8] Broadbent, T. P. and Fessler, H., "Stress Analysis of Screwed Tubular Joints," Fatigue and Crack
Growth in Offshore Structures, Proceedings of the Institute of Mechanical Engineers, Institute of
Mechanical Engineers, London, 1988, pp. 171-186.
[9] Bluhm, J. L. and Flanagan, J. H., "A Procedure for the Elastic Stress Analysis of Threaded Con-
nections Including the Use of an Electrical Analogue," Experimental Stress Analysis, Vol. XV,
1957, p. 85.
[10] Miller, D. L., Marshek, K. M., and Naji, M. R., "Determination of Load Distribution in a Threaded
Connection," Mechanism and Machine Theory, Vol. 18, No. 6, 1983, pp. 421-430.
[11] Newport, A., Topp, D. A., and Glinka, G.,"The Analysis of Elastic Stress Distribution in Threaded
Tether Connections," Journal of Strain Analysis, Vol. 22, No. 4, 1987, pp. 229-235.
[12] Dowling, N. E., Brose, W. R., and Wilson, W. K., "Notched Member Fatigue Life Predictions by
the Local Strain Approach," Fatigue Under Complex Loading: Analyses and Experiments, Society
of Automotive Engineers (SAE), Warrendale, PA, 1977, pp. 55-84.
[13] Morrow, J., "Cyclic Plastic Strain Energy and Fatigue of Metals," Internal Friction, Damping and
Cyclic Plasticity, ASTM STP 378, American Society for Testing and Materials, Philadelphia, 1965,
p. 45.

He r be r t J. Sut herl and, ~ Paul S. Veers, 1 and Thomas D. As hwi l P
Fatigue Life Prediction for Wind Turbines:
A Case St udyon Loading Spectra and
Parameter Sensitivity
REFERENCE: Sutherland, H. J., Veers, R S., and Ashwill, T. D., "Fatigue Life Prediction
for Wind Turbines: A Case Study on Loading Spectra and Parameter Sensitivity," Case
Studies for Fatigue Education, ASTM STP 1250, Ralph I. Stephens, Ed., American Society for
ABSTRACT: Wind turbines are fatigue-critical machines used to produce electrical energy
from the wind. These rotating machines are subjected to combinations of wind, gravity, and
gyroscopic loadings that are highly irregular in nature. Historical examples of fatigue problems
in both research and commercial wind-turbine development are presented. Some example data
on wind-turbine environments, loadings, and material properties are also shown. Before a
description of how the authors have chosen to attack the cumulative damage assessment, ques-
tions are presented for the reader' s reflection. The solution technique used by the authors to
define the loading spectrum for wind turbines is then presented. Special emphasis is placed on
the development of a loading spectrum for use in the fatigue analysis. Less attention is paid
to methods of cumulative damage assessment; Miner' s rule and constant amplitude S-n data
are used. A case study then applies the procedures to an actual wind-turbine blade joint. The
wind turbine is the 34-m diameter vertical axis wind turbine (VAWT) erected by Sandia Na-
tional Laboratories near Bushland, TX. The case study examines parameter sensitivities for
realistic uncertainties in inputs defining the turbine environment, stress response, and material
properties. The fatigue lifetimes are calculated using a fatigue analysis program, called LIFE2,
which was developed at Sandia. The LIFE2 code, described in some detail in the Appendix,
is a PC-based, menu-driven package that leads the user through the steps required to charac-
terize the loading and material properties, then uses Miner' s rule or a linear-crack propagation
rule to numerically calculate the time to failure. Only S-n based cumulative damage applications
are illustrated here.
The LIFE2 code is available to educational institutions for use as a case study in describing
complicated loading histories and for use by students in examining, hands on, parameter sen-
sitivity of fatigue life analysis
KEYWORDS: fatigue, wind turbines, loading spectrum, damage assessment, Miner' s rule,
uncertainty, variability
Hi s t ory
Mos t l arge, mode r n wi ndmi l l s ( cur r ent l y cal l ed wi nd t urbi nes) ar e used to pr oduce el ec-
t ri cal powe r f or ut i l i t y appl i cat i ons. The l ar gest concent r at i on o f t hi s cl ass o f t ur bi nes i n t he
wor l d ( about t wo- t hi r ds o f t he wor l d' s capaci t y) is cont ai ned in t hr ee Cal i f or ni a sites. Cal -
i f or ni a has over 16 000 t ur bi nes wi t h a t ot al r at ed capaci t y o f be t we e n 1500 and 1600 MW,
gener at i ng appr oxi mat el y 1% o f Cal i f or ni a' s t ot al el ect r i cal usage. I n 1992, t hey gener at ed
1 Distinguished member of technical staff, senior member of technical staff, and senior member of
technical staff, respectively, Wind Energy Technology Department, Sandia National Laboratories, Al-
buquerque, NM 87185-0708.
174

SUTHERLAND ET AL. ON WIND TURBINES 175
2.8 billion kWH of electricity, whi ch is sufficient to supply the residential electricity needs
of a city the size of San Francisco, CA or Washington, D.C. for one year. In that year alone,
they averted emission of more than 2.8 billion lbs of carbon dioxide, a major greenhouse
gas, and 17 million lbs of other pollutants that woul d otherwise be produced by fossil-fuels
[1,2]. While additional developments are continuing throughout the United States, additions
to European wi nd installations are projected to surpass U.S. capacity before the turn of the
century, when Europe will have two-thirds of the worl d' s wi nd power [3].
Wind turbines may be divided into two generic groups based on their axis of rotation:
horizontal axis wi nd turbine (HAWT) blades rotate about a horizontal axis and vertical axis
wi nd turbine (VAWT) blades rotate about a vertical axis. A typical HAWT, the Northern
Power Systems 100-kW turbine, is shown in Fig. 1 [4]. An example of a VAWT is shown
in Fig. 2. This turbine, the Sandi a/ DOE 34-m Test Bed, is a 500-kW research turbine, located
near Bushland, TX, and built as part of the U.S. Department of Energy (DOE) wind energy
research program [5]. Other turbines have more or fewer blades (some have only one) and
HAWTs may have either an up-wi nd or down-wi nd blade orientation. The VAWT Test Bed
is used in an example presented here; however, the discussions and analysis techniques
presented in this paper are applicable to wind turbines of either configuration.
Fatigue Problems with Wind Turbines
The first truly modem wind turbine, the Smith-Putnam megawat t scale HAWT, was built
in New England during World War II under wartime motivation to save oil. The machine
experienced continual problems with blade fatigue and was never able to achieve its promised
productivity. When wind energy again gained promi nence in the early 70' s (following the
Arab oil embargo), there was a t endency to view implementation of the t echnol ogy as ba-
sically a probl em in aerodynamics. The first research machine installed at Sandia National
Laboratories was operated on the r oof of a three-story building in the center of a populated
area. No structural measurements were collected during initial performance testing. Onl y
later, when a strain gage was monitored, was it discovered that the blades were loaded nearly
to their yield strength on every rotation. The testing resumed with reinforced blades in a
more remote site.
A large HAWT research machi ne of mid 70' s vintage, the NASA 100 kW Mod-0, also
experienced fatigue problems. Maintenance workers first saw evidence of hydraulic fluid
leaking from blade cracks during a visual inspection from the ground. A fairly complicated
joint, between a steel spar and the al umi num stressed skin of the blade, had numerous cracks.
Crack patching efforts onl y extended the life marginally, providing time to obtain perfor-
mance data. Subsequent analyses revealed that the loads on the turbine due to wind shear
(the variation in wind with height) and tower shadow (wind bl ockage by the tower) had been
grossly underestimated 2.
Even nominally simple component s with simple designs and loadings can experience fa-
tigue problems. The multi-megawatt (2.5 MW) MOd-2 turbines located near Goodnoe Hills,
WA had a simple, hollow, circular mai n shaft (the shaft that connects the blades to the gear
box). The predominant load on the shaft was due to the weight of the rotor, that is, simple
gravity loading on a rotating shaft. Access ports into the center of the shaft were thoroughly
analyzed before installation. Unfortunately, unknown to the designers and analysts, instal-
lation crews drilled tiny holes near the access ports for a set of mount i ng brackets. A crack
started in one of these holes, propagated to th e larger ports, and was almost hal f way around
2 Spera, David, private correspondence, 1993.

FIG. 1--The NPS HAW'E.
the shaft before a maintenance technician, working on the parked rotor for completely un-
related reasons, noticed the huge crack. 2
Problems were also experienced in the commercial sector during this period. One entre-
preneurial start-up company of the early 80' s was sponsored by investors who were seeking
to harvest tax credits. These credits required that the turbines be installed before the end of
the calendar year. The company' s sole engineer was trying desperately to get some operating
data (performance, stress levels, and so forth) off the prototype machine on a windy New
Mexico mesa while supervising the production and installation of a group of machines near
Palm Springs, CA. Equipment malfunctions and telemetry problems on the prototype made

f
\
FIG. 2--The Sandia 34-m VAWT Test Bed.
time series data hard to get, but data in the form of hardware kept appearing on the ground.
Every time the prototype was operated, broken bolts were strewn about the turbine. Subse-
quent operational and modal test results indicated that the rotor had not one, but two natural
frequencies of vibration coincident with the rotational speed of the turbine. Either one of
these resonant conditions would have been sufficient to shake the turbine apart in a few
hours of operation?
Virtually all of the turbines built in California in the early 80' s have experienced fatigue
problems in "energetic" sites (sites with an average wind speed of 7 m/ s or more). Blades
3 The production machines had to be erected before the end of the year for tax purposes. Although
retrofit with stiffening struts to eliminate the resonances, they were plagued with problems and never
became productive machines. The company soon went bankrupt. After standing idle for a decade, the
machines were removed.

have been repaired or replaced on most of the turbines. Inspection and maintenance programs
have progressed t remendousl y as the operators gain the wi sdom of thousands of hours of
operating experience on thousands of turbines. Turbines installed more recently have dem-
onstrated tremendous improvements in availability. In addition, research programs at the
national laboratories are measuri ng material fatigue properties, defining turbine loadings, and
creating fatigue lifetime estimation tools for use in the design of the new generation of
machines currently on the drawing board at a handful of U.S. companies.
Why so Many Fatigue Problems?
A wind turbine residing in a farmer' s pasture on a midwestern plain or in a mountain pass
near a hot California desert is subject to the whi ms of Mot her Nature. It must rotate when
the wind is blowing anywhere from 5 to 25 m/ s (12 to 60 mph), enduring many cycles of
vibration. The machine must also be able to stop in any wi nd condition and survive bone-
rattling gusts (in a parked condition) greater than 55 m/ s (120 mph).
At rotation rates from 30 to 60 rpm, it does not take long to accumulate large numbers
of oscillations. The blades of a 30-year life turbine typically must withstand at least 109
cycles, quite large for any rotating piece of equipment. The stress concentrations at "hot
spot s" are often not known, and the fatigue properties of typical turbine materials are not
well characterized.
Wind turbines were installed in large numbers in California in the early 1980' s. The driver
for this installation frenzy was the tax advantage to the investor of an installed machine,
regardless of the eventual revenue stream. 4 Some particularly poorl y engineered machines
were blown down by the first major storms through the mount ai n passes where the wind
farms are located. But most machines had no trouble with the severely stormy environments;
they just tended to break under the continual cyclic loading of operation.
Even the most successful of the early turbine designers had overlooked a significant com-
ponent of the loads. The designers assumed the loads could be approximated using a spatially
uniform incoming wind, with variations in the average proceeding very slowly. However,
unlike aircraft, which fly in normally uni form flows high above the ground, wi nd turbines
always operate in the earth' s turbulent boundary layer. Thus, the incoming wind fluctuates
significantly in both time and space. Turbulence was originally t hought to produce dynamic
loads at sufficiently low frequencies to be neglected. But because the blades produce rela-
tively large lift forces as they move t hrough the wind, the turbulence-induced loads are
magnified and shifted to higher frequencies (just as the speed of a car driving over a rough
road affects the frequency of the loads). The broad-band nature of turbulence also excites
the natural frequencies of the turbine more than had been expected.
Economics
As illustrated in the examples cited above, wi nd turbines are subjected to at least a billion
fatigue cycles during the course of a service lifetime. I n comparison, transport aircraft are
typically designed for bet ween one and ten million stress cycles. The turbine must therefore
endure the lifetime stress cycles of an aircraft every three months of its service lifetime. The
ability of turbine designers to address a difficult design probl em with "hi gh t ech" solutions
is greatly restricted by economics. Wind turbines are designed to produce electrical power
4 The wind industry is currently opposed to any tax incentives based on installation, but instead favors
production based measures to rectify inequities in relative treatment of competing electricity sources.

f
0
L
0 800 1000
I , , I I
2 0 0 400 6 0 0
Time, sec
FIG. 3--Typical wind speed time series data.
for the grid and therefore compete economi cal l y with conventional sources of electrical
power generation. As an example, t he use of aerospace class materials to reduce weight and
t o enhance the reliability and structural properties of turbine blades is almost impossible
because of the cost. The wind industry typically builds blades using fiberglass composites
from boat building t echnol ogy or al umi num alloys typically reserved for architectural trim.
T h e P r o b l e m
Overview
After the previous bri ef perspective on the difficulties and constraints of the wind business,
we now turn our attention to the probl em of predicting fatigue lives for wind turbine com-
ponents. First, raw data from operating wind turbines are presented to provide an example
of typical environments, loadings, and material properties for wind turbines. The reader is
then asked to stop and reflect on questions about how to approach the cumulative damage
assessment under such circumstances. The solution technique developed by the authors is
then presented with special emphasis on development of the fatigue loading spectrum. A
comput er code called LI FE2 [6-11], capable of calculating fatigue lifetimes for wind turbine
components, is described in some detail in Appendi x A. The last maj or section of this paper
uses the LI FE2 code to conduct a case study on the variability of fatigue life predictions
due to realistic uncertainties in the input parameters required by this analysis technique?
Raw Data Examples
The wind itself, of course, is random. An example of a wind time series from the Texas
Panhandle is shown in Fig. 3. This data record lasts approximately 15 min (1000 s), during
which time the wind speed varies between 8 and 15 m/ s. The so-called turbulence intensity,
a measure of the amount of fluctuation in wind speed, is comput ed by dividing the standard
deviation by the mean. The average wi nd speed in Fig. 3 is 11.4 m/ s, and the turbulence
5 The LIFE2 code is available by request from the authors.

intensity i s 11.7%. Most current wind farms in the United States are located in mountain
passes where the winds are much more turbulent, with intensities between 15 and 30%.
Turbine blades are the most fatigue-sensitive component of the turbine because of the
magnitude and number of stress cycles they experience. Research turbines, prototypes, and
even some production turbines are strain-gaged t o measure the response of the blades in
various conditions [12]. Figure 4 shows the response of the 34-m Test Bed blades in bot h
the flatwise bending (about the chord line of the airfoil) and lead-lag bending (about an axis
25
20
5
~g
o
~ -5
-lO
.1 -15
1 1 0 , I , I , L , I ,
-20
r ~
- [
m
-30
-40
-50
-60
0
200 400 600 800
Ti me , sec
1000
FIG. 4- - Test Bed blades stresses, flatWise and lead-lag in 11 m/ s winds.

r
o .
~q
14
12
I0
A A ,A A A , ^ , A A , ^ , ,
, [ w v y ' v v v V
" t , , , , I , , , , I , , , , t , , , ,
600 605 610 615 620
Ti me, sec
per pendi cul ar to t he chord). 6 It is apparent t hat the charact er of the bl ade response is di fferent
f r om that of t he wi nd input. (The wi nd speed is measur ed about t wo rot or di amet ers upwi nd
of t he t urbi ne. ) Twent y- second segment s of Fig. 4 are expanded in Fi gs. 5 and 6, showi ng
t he fl at wi se and l ead- l ag responses, respect i vel y, in mor e det ai l . Even for such a short t i me
segment , the wi nd speed fluctuates consi derabl y, bet ween 9.5 to 13.5 m/ s . Thi s is not at yp-
i cal . The t urbi ne bl ades r espond in a mor e nar r ow- band fashi on than t he wi nd in bot h t he
l ead- l ag and fl at wi se di rect i ons, and t he stress ampl i t udes do not consi st ent l y correl at e wi t h
t he wi nd peaks.
Exampl es of t urbi ne response whi l e operat i ng in wi nds ar ound 19 m/ s are shown in Fi g.
7. The stress l evel s whi l e t he rot or is oper at i ng at r el at i vel y l ow wi nds of 6 m/ s are shown
in Fi g. 8. The stress l evel s whi l e the rot or is par ked (not rot at i ng) in 14 m/ s wi nds are
shown in Fi g. 9.
Fat i gue propert i es can be difficult to find for t hose mat er i al s not al ready used in aerospace
or gr ound vehi cl e appl i cat i ons (as is t ypi cal for most wi nd t urbi ne bl ades). The so-cal l ed
6 The 34-m Test Bed has been equipped with a large array of sensors to monitor all aspects of the
machine's performance. Current instrumentation includes 72 strain signals from the rotor, 25 environ-
mental signals, 22 turbine performance signals, and 29 electrical performance signals. An overview of
the instrumentation is provided in Ref 5.

- d
r
w )
r~
r
14
10
- 2 8 . . . . I . . . . I . . . . I . . . .
-30
-40
"5064 605 610 615 620
Ti me , sec
FIG. 6--Twenty second time series of flatwise stresses.
endurance limit (stress amplitude with zero mean at 107 cycles to failure) is the most likely
property to be found in handbooks. An S-n curve woul d provide the next level of detail, i f
it can be found. In many cases, coupon testing will be required to characterize the specific
material used for the blades. Figure 10 shows the raw data that resulted from coupon tests
of (1) an extruded aluminum alloy (6063) [13] and (2) a unidirectional fiberglass composi t e
[14], two typical materials used in wind-turbine blades. The al umi num data are shown in
tabular form in Appendi x B.
Q u e s t i o n s
Before proceeding, it may be useful to think t hrough one' s own met hod of attacking the
problem of calculating fatigue life under the conditions laid out above. The reader is therefore
asked to pause here, before becomi ng biased by the solution technique laid out in subsequent
sections, and reflect on the following questions.
9 How would you formulate the cumulative damage problem for wind-turbine blades?
How do you categorize events? How do you assign damage to events? What parameters
are required? Whi ch ones do you think are most important?
9 How can one categorize wi nd speed with its large variability into something that can
be used in a fatigue analysis? How can turbulence intensity be included?

"~ 2 5
r ~
Gg
r ~
r ~
2 0
15
l:[
10
m
1000
9 How woul d you find the hi ghest stressed l ocat i on(s) on a wi nd t urbi ne bl ade?
9 How do you charact eri ze i rregul ar l oadi ngs l i ke t hose shown in Fi gs. 4 to 9? Do t hey
need to be consi dered in t he anal ysi s?
9 How i mpor t ant is the act ual sequence of the peaks and val l eys in t he l oadi ng?
9 Why is t he charact er of t he bl ade response so much di fferent f r om the wi nd input?
9 How can t i me hi st ori es of t urbi ne response be used in the fat i gue anal ysi s pr ocedur e?
9 What oper at i ng dat a on t he wi nd t urbi ne woul d you col l ect and how woul d you reduce
(anal yze) t he dat a?

25
20
15
10
l e
I 0
5
0
-5
- I 0
-15
, d
-20
20
m
, I , i , I , I ,
-2C
i , i , i , i
-40
"600 200 400 600 800
Ti me, sec
FIG. 8- - Fl at wi s e and wind-log stresses in 6 m/ s winds.
1000
9 Coul d t he publ i shed S- n curve of a si mi l ar but di fferent al l oy (or composi t e l ay- up)
al ong wi t h a publ i shed endurance l i mi t be used t o creat e an S- n curve for the 6063
al umi num (or fi bergl ass bl ade)? I f so, how?
9 How coul d one fit a curve or set of curves to such a wi de spread of S- n data, shown
in Fi g. 10 and Appendi x B? How woul d you combi ne mean and ampl i t ude i nformat i on
in an anal yt i cal model ? How woul d you expect curve fitting or model i ng di fferences to
affect t he pr edi ct ed fat i gue l i fe of a component ?
9 What sensi t i vi t y do you expect in fat i gue l i fe pr edi ct i ons? (that is, what is t he di fference

r
g~
oi l
" f
2O
10
r ~
e .
r ~
4,)
, d
w~
5
10
10
5
0
- 5
-10
- 15
- 2 0
10
' - " ] - - - " - " - r , r - - " l r w- . w
. , , aJ L . . . . . . b ._. _ Am hJtl" ~ L. , ~, _. L~, , . L, . ~
0 'J-,w - - v, ~- ~' ' f f ~' " 1 v, r - " l t ~r l , , r - . l " ~q' r l v" ' ~ 1 r - '
i I I I
2 0 0 4 0 0 600 800 1000
Time, sec
-10
- 2 0
-30
-40
0
FIG. 9--Parked blade stresses, both flatwise and lead-lag, in 14 m/ s winds.
in predi ct i ons due to 5 to 10% changes in i nput par amet er s?)
The r eader is l eft wi t h many questions. I nf or mat i on is pr ovi ded bel ow to i l l ust rat e how
we answered some, but not all, of the questions t hat are posed here.
Sol ut i on T e c h n i q u e
Appr oach
Due to the compl exi t y of const ant l y varyi ng st resses exci t ed by const ant l y varyi ng wi nd
speeds, a general , anal yt i cal sol ut i on to the pr edi ct i on of fat i gue l i fet i me for a wi nd t urbi ne

r ~
e .
. = .
t . .
<
v-q
O
2.3
2 . 2 ~ : O . . ' ~ . 9
2 . , ~ . ~ .
No ~ Mo a n s ~ o s s 9 "~-~'"~ ";~,.~'~,.
1.9 ~ 100 MPa l l ~ . : .... ""~"4~ ' 5 "~"
t - - I - - "" . "41~. ~ " ' - . 1.8
69MPa 9 . . , . _ ' " ' -
1.7 - ""[~'" ~ D ~ 35 MPa
t - - 4 ' -
1.6 - 0 MPa
. , @. .
1 . 5 , I , I , I , I ,
4 5 6 7 8
Logl0(Cycles to Failure)
1.2
b
9 Pol yest er Mat r i x wi t h
1.0 Uni Di r ect i onal Fabr i c
~- 9 Vi nyl est er Mat r i x wi t h
r ~ ~ Uni Di r ect i onal Fabr i c
~ 0 . 8 ~ * Pol yes t er Mat r i x wi t h
"*" J ~ Non- Wove n Uni axi al Fi ber s
- o 0 4
Z 0 . 2
1E+0 1E+I 1E+2 IE+3 1 E + 4 1 E + 5 IE+6 IE+7 1E+8 IE+
Cycl es to Fai l ure
FIG. lO--Fatigue test results. (a) 6063 aluminum. (b) unidirectional fiberglass composite.
does not a ppe a r t o be t r act abl e unl ess si gni f i cant as s umpt i ons ar e i nt r oduced. The l oadi ng
s pe c t r um p r o b l e m is br oken up i nt o a ser i es o f smal l er , t r act abl e p r o b l e m set s t hat ar e bas ed
upon t he oper at i onal st at e o f t he t urbi ne. The a ppr oa c h c a n be s umma r i z e d b y st ar t i ng wi t h
t he det ai l ed st ress s pect r um defi ni t i ons and i nt egr at i ng ove r s ucces s i vel y br oa de r s egment s
o f t he ma c h i n e ' s oper at i onal states. Fi rst , t he st r esses at each nomi na l wi nd s peed ar e def i ned
b y a hi s t ogr a m ( or pr obabi l i t y densi t y f unct i on) o f st ress cycl es. The annual d a ma g e at a
par t i cul ar wi nd speed is det er mi ned b y i nt egr at i ng ove r t he st r ess cycl e di st ri but i on, wei ght -
i ng b y t he a mount o f t i me spent at t hat wi nd speed, and usi ng Mi n e r ' s r ul e t o s um da ma ge
ba s e d on const ant a mpl i t ude S- n data. Th e t ot al da ma ge dur i ng oper at i on i s cal cul at ed b y
Copyri ght by ASTM Int ' l (al l ri ght s reserved); Thu May 8 10: 42: 12 EDT 2014
Downl oaded/ pri nt ed by
Uni versi dad De Chi l e pursuant t o Li cense Agreement . No furt her reproduct i ons aut hori zed.

SUTHERLAND El" AL, ON WIND TURBINES 187
integrating over all wi nd speeds. Finally, damage during other transient events, such as start-
up or buffeting while parked, are added t o estimate the total damage over an average year,
that is, the annual damage rate. Lifetime is the reciprocal of the average damage rate. This
approach has the distinct advantage of allowing the analyst to repeat the calculation for
different wi nd sites, whi ch is of interest since a turbine will rarely be designed for a single
site.
The calculations are done with the aid of the previously mentioned LI FE2 code (see
Appendi x A). It leads the user t hrough the probl em set-up in four steps. First, the wi nd
speed distribution for the turbine site is described by an average annual distribution. The
second input describes the material fatigue properties required by the damage rule being
used to predict the service lifetime of the component. The third input is a joint distribution
of mean stress and stress amplitude (stress states) for all of the various operational states of
the turbine. These "cycl e count mat ri ces" can be defined for each operational state using
time series data, obtained from simulated or measured time series, using a rainflow counting
algorithm, [15] or from anal yt i cal / numeri cal models. The fourth input describes the opera-
tional parameters for the turbine and the stress concentrations factor(s) for the turbine com-
ponent. (Wind speed distribution, operating stress-state distribution, and material data are
described in more detail below.) Finally, the damage caused by cycles at each "stress state"
can be summed, first over the distribution of operational stress states at each wind speed,
and then over all wi nd speeds and other turbine states. The LIFE2 code actually contains
two fatigue analysis approaches; the first applies Mi ner' s rule to S-n data and the second is
a linear crack propagation model [9,16].
Assumptions
In the case of wi nd turbines, any design that has a chance of success must be able to
endure a large number of cycles. Therefore, the material response must be nominally elastic.
Either fatigue life or crack growth analysis can, and has been, used to estimate wind turbine
fatigue lifetimes. We will restrict the discussion here to an S-n analysis using Mi ner' s rule
[16]. While this approach has many flaws, wi nd turbine applications create fewer problems
than most with Mi ner' s rule because of the high-cycle, limited-plasticity nature of the load-
ing, keeping load sequence effects relatively small. (Reference 19 has a discussion of the
relative errors to be expected in applying simple damage rules in wind turbine applications.)
Therefore, loading statistics that preserve the average numbers of cycle amplitudes are suf-
ficient for our purposes. Keep in mi nd that this approach is a rough approximation in the
wi nd turbine case and may be less suited to other applications. However, the case study
results that follow suggest that errors in the fatigue model i ng may be swamped by the effects
of uncertainty in the input parameters.
The internal stresses in the turbine blades are assumed t o be described by beam bending
with local stress concentrations at bol t holes and other nonuniformities. All the measurements
from the operating turbine are collected as beam-bendi ng loads and converted to nominal,
maximum-fiber stresses using Mc/ L Local stresses are estimated by applying stress concen-
tration factors to the nominal stress.
The Wind Regime
Typically, a wi nd turbine is designed for a generic site such as California mountain passes,
the Great Plains of the United States, the North Sea coast of Europe, and so forth. I n the
initial design phase, the annual wind speed distribution for these sites is modeled using a
Weibull or a Rayl ei gh probability density function [17]. The Weibull distribution is defined

by the average wi nd speed and a "shape factor. " The Rayl ei gh di st ri but i on is a speci al case
of the Wei bul l wi t h a shape fact or of t wo. As the desi gn, t est i ng, and i nst al l at i on of a t urbi ne
cont i nue, t he generi c descri pt i ons can be r epl aced wi t h si t e-speci fi c data, usual l y in the f or m
of a hi st ogr am of the annual wi nd speed di st ri but i on. 7 The percent age of t i me spent in any
range of wi nd speeds is defi ned by t he i nt egral of t he di st ri but i on over the range (or the sum
over the nor mal i zed hi st ogr am cells).
Const i t ut i ve Propert i es
A st ress-based fatigue l i fe approach usi ng Mi ner ' s rul e to accumul at e damage from each
stress cycl e requi res that t he number of cycl es to fai l ure be descr i bed as a funct i on of cycl e
mean and ampl i t ude. For many mat eri al s t hi s function, t ypi cal l y cal l ed an S- n di agram, may
be posed usi ng Goodman or Ger ber model s [16]. In bot h of t hese model s, the mean stress
component , Sin, and the al t ernat i ng stress component , Sa, of t he cycl e are mapped into an
effect i ve stress state, Se~ t hat cor r esponds t o the S-n di agr am at zero mean stress ( Appendi x
B).
Typi cal l y, the S-n behavi or can be found in mat eri al handbooks or vi a a l i t erat ure survey.
However, for wi nd t urbi nes t here are t wo si gni fi cant difficulties. The first is that the prepon-
derance of avai l abl e dat a is for aerospace mat eri al s. These dat a are of little val ue here because
the hi gh cost of t hese mat eri al s has rest ri ct ed t hei r use in wi nd turbines. The second is t hat
many fat i gue cri t i cal desi gns do not need to resi st 109 cycl es. Thus, t he avai l abl e S-n dat a
bases t ypi cal l y cont ai n l i t t l e or no very hi gh- cycl e data. For t hese reasons, mat eri al fat i gue
propert i es have been der i ved from ext ensi ve coupon t est pr ogr ams sponsor ed by the DOE
wi nd energy pr ogr am t hrough the nat i onal l aborat ori es. The raw dat a on ext ruded 6063
al umi num are t abul at ed in Appendi x B. These dat a are used, in conj unct i on wi t h a Goodman
mean stress formul at i on, to eval uat e bl ade j oi nt lives in this 34- m Test Bed case study.
Operat i ng St resses
The wi nd dr i vi ng a wi nd t urbi ne is const ant l y fluctuating, even in the st eadi est of sites.
Even for the br i ef per i ods when a measur ed wi nd speed is seen to be rel at i vel y constant,
stresses on a wi nd t urbi ne rot or can be seen to be fluctuating "r andoml y. " Si mi l ar responses
are found in many appl i cat i ons, i ncl udi ng gr ound vehi cl es, aircraft, off-shore structures, and
bri dges.
Rai nfl ow count i ng [15] is t ypi cal l y used to find cl osed hyst eresi s l oops in the mat eri al
response, whi ch are defi ned to be the cycl es. Cycl e means and ampl i t udes are col l ect ed i nt o
hi st ogr am bi ns and di spl ayed as ei t her frequency of occurrence or exceedance di agrams. The
first is equi val ent to a pr obabi l i t y densi t y funct i on and the second is l i ke an inverse cumu-
lative di st ri but i on funct i on from pr obabi l i t y theory. Fr equency- of - occur r ence exampl es of
j oi nt di st ri but i ons of cycl e means and ampl i t udes for t he HAWT in Fi g. 1 are shown in Fi g.
11.
Mal col m [18] has used VAWT exampl es to i l l ust rat e how the shape of t he st ress-range
di st ri but i on, when nor mal i zed by the r oot mean square ( RMS) of the stress t i me series, can
be i ndependent of wi nd speed. Once t he di st ri but i on t ype has been est abl i shed, onl y t he
7 The annual wind speed distribution is also one of the prime variables that determine the annual
power production of a wind turbine. This wind speed distribution enables investors to project the "return-
on-investment" from power sales over the turbine's lifetime. Thus, site wind speed statistics are not only
a major factor in determining the service lifetime, they are the determining f act or f or the economics of
turbine placement.

a
Cycle Counts
"' ~#~ ~e a n stress
Cycle Counts
b
" ~ # e ~ I Mean stress
FI G. l l - - J o i n t histogram of blade stress means and amplitudes f or the NPS HAW'E ( a) root f l ap
bending stress. ( b) root edgewise bending stress.

functional relationship between the RMS and the wind speed needs to be defined. (Defining
the high amplitude "tail" of the distribution can be quite difficult and is a topic of current
research interest.) In addition, the magnitude of the mean stress fluctuations (during opera-
tion) is generally small when compared to the material strength and can therefore be ne-
glected. However, the shift in mean stress between operating and parked states may produce
significant damage in the turbine blades, like the G-A-G cycle in aircraft. The histogram in
Fig. 12 is typical of the distribution of rainflow counted ranges when the mean stress vari-
ations are ignored.
Veers [19,20] has shown that most VAWT operating stresses may be modeled as a narrow-
band Gaussian process. The resulting Rayleigh distribution of stress amplitudes and ranges
(Fig. 12) is a one-parameter distribution described by the RMS of the time series. To use
this model for the operational stresses, only the RMS stress, mean stress, and cycle rate must
be defined as functions of wind speed. This model works best for flatwise VAWT response
where the stresses are largely a random vibration response to turbulence. But the model is
often poor in lead-lag response because of a substantial sinusoidal gravity load (in HAWTs)
and cyclic torque (in VAWTs).
Case Study
Introduction
This case study is presented to illustrate the variation in fatigue analysis results for an
actual machine using analytical, measured, and reference-book estimates of the quantities
FIG. 12--Histogram of rainflow counted stress amplitudes for the Test Bed.

governi ng the fatigue analysis. The wi nd turbine under consideration is the Test Bed shown
in Fig. 2. This fatigue analysis was originally conduct ed by Ashwill et al. [21]. As discussed
above, some of the analysis presented here depends on the design characteristics of this
particular turbine; however, the discussions and analysis techniques are applicable to all wind
turbines. For this case study, we onl y consider the stresses imposed on the turbine during
operation, that is, the operational stresses.
This example follows the sequence established above and implemented in the LI FE2 code
(Appendix A). We describe the Test Bed, the four input parameter sets, and finally, we
calculate the predicted service lifetime for the turbine blade. The latter calculations are
presented as a parametric study for variations in the input quantities. The LIFE2 code was
used for all fatigue life calculations.
Sandia/DOE 34-m VAWT Test Bed
The Test Bed (Fig. 2) is a 34-m diameter VAWT erected by Sandia National Laboratories
near Bushland, TX (just west of Amarillo) for research purposes. The turbine and site have
been equipped with a large array of sensors that permit the characterization of the turbine
under field conditions [5,22,23]. The extensive experimental data base allows us to illustrate
the fatigue analysis of a wi nd turbine component and to examine the sensitivity of lifetime
predictions to various estimates of the input data.
Structural dynami c analysis confirmed by test measurements [12] indicated the highest
stressed region of the turbine was at the upper blade-to-tower joint (called the upper root).
(Figure 13a shows the mirror-image lower root.) This example examines the stresses i mposed
on the blade at this joint. The blade comi ng into the joint has an airfoil shape with a 48-in.
cord (Fig. 13b), and a bolted "cl am shell" joint connecting it to the central column. The
stresses in both the flatwise bending direction (about the cord of the airfoil) and the lead-
lag bending direction (about an axis perpendicular to the chord) are critical in fatigue.
The Wind Regime
For the Test Bed example, we examine the differences created by using three different
annual wi nd speed distributions. The first distribution is a generic description of the wind
speeds in the Texas panhandle region of the U.S. Great Plains, which could be obtained
from a wi nd energy atlas [24]. A Rayl ei gh distribution with an average wind speed of 6.3
m/ s (14 mph) is typical of the area. The second distribution, focusing more directly on the
specific site, is a histogram O f 17 years of data at the Amari l l o Airport 8 and has an average
of approximately 6.6 m/ s (15 mph). The third distribution, obtained over 6 years (1983 to
1988) at the Bushland site, 8 has a 5.8 m/ s (13 mph) average. All three distributions are
shown in Fig. 14.
There are three maj or differences in these wind speed distributions that affect the service
lifetime predictions. The first is that the Rayl ei gh distribution has a long tail extending out
to very hi gh winds, whereas the Bushland and Amarillo distributions contain little occurrence
of winds above 20 m/ s. The second is that the Amarillo distribution has a hump in moderat e
winds that is not seen in the Bushland data. The third difference is reflected in the average
wi nd speeds that range f r om 6.6 to 5.8 m/ s. For turbine operation between 6 and 20 m/ s,
the Amarillo site is the most energetic and the Bushland site is the most benign.
8 Clark, N., USDA Agricultural Research Center, Bushland, TX, private correspondence, 1989.

~ _ , | .
mU~DE CLAI P
L_ J . ,o,~ ,o,,o,,.,
a
i.,J
b . , , . o o , . , " 1
FIG. 13--The upper root of the 34-m Test Bed. (a) blade-to-tower joint. (b) airfoil cross section near
the tower.
Const i t ut i ve Propert i es
The Test Bed bl ades wer e ext ruded f r om 6063 al umi num alloy. Ini t i al l y, onl y l i mi t ed
fat i gue dat a for this mat er i al were avai l abl e in the literature. S-n propert i es had to be ex-
t r apol at ed f r om this l i mi t ed dat a set and knowl edge of t he propert i es of ot her al umi num
al l oys [25]. The resul t i ng zero mean stress S- n curve is shown in Fi g. 15. Thi s curve was
expanded t o i ncl ude mean stresses usi ng Goodman' s rul e wi t h an ul t i mat e stress of 241 MPa
(35 ksi ). These dat a (Fi g. 15) are descr i bed here as t he "r ef er ence dat a. "
Because the reference dat a were suspect , a seri es of S- n mat er i al tests was conduct ed on
t he ext ruded al umi num bl ade mat er i al [13]. The t est dat a at vari ous mean and al t ernat i ng

SUTHERLAND E l AL. ON WIND TURBINES 193
0.08
t 9 Rayl ei gh Di st r i but i on
0. 07 L " " "~' An~arillo Ai r por t
0 . 0 6 9 " . . . . . .
" ~ I . ; t ~ Bus hl and Test Site
0 . 0 4 I - I l . ~ "
0.03 I - I ; : : \ '
F / ! : , . \ '
0.010.00 ~ : "" *" 9 "
0 5 10 15 20 25
Wi n d Spe e d, m/ s
FIG. 14--Typical wind speed distributions.
stress levels shown in Fig. 10a have been converted to "effect i ve" stress levels using a
Goodman rule, based on the measured ultimate stress of 244 MPa (35.4 ksi) [8].
Three formulations of the S-n behavior were derived from the above data set. The first is
the mean or "least squares cur ve" (LSC) fit to the data, which approximates a 50% survival
level. The second, based on a Weibull fit to the variations about the LSC, is an estimate of
the 95% survival level. A third eliminates the change in slope at about 10 million cycles in
the LSC fit and linearly extends the curve out to the hi gh cycl e region [261. (This is equivalent
to the extrapolation of l ow-cycl e S-n data to high cycles.) These formulations (Fig. 15) will
be called the LSC, the 95%, and the high cycle extrapolation fits, respectively.
Stress States
For this example, we examine only the operational stresses on the turbine, neglecting
cycles accumulated while parked in high winds and the start-stop cycles. The narrowband
Gaussian model for stress response is used here. 9 The stress states have been measured and
9 The rainflow counting technique may be used to create histograms of stress states for this turbine
since a large amount of time series data is available. However, the narrow band Gaussian model was
chosen for this case study for its simplicity and because it yields better insight into the dependence of
the operational stress levels on wind speed.

2 " 5 1 I
2 " 4 k - - . . . . . L e a s t o S q n a --r e s " C u r v e F i t I
s. . _ " ~ 95 Yo Survival Level 1
2 . 3 1 ~ . . . . . . . . . . . . . . . ". " \ . . . . . . I
" " i " I
: , - . I
ll:;
3 4 5 6 7 8 9
L o g l O ( C y c l e s t o F a i l u r e )
FIG. 15--Constitutive properties for the 6063 aluminum.
compared to analytical predictions with good results [12,22]. As mentioned above, the high-
est stressed region of the blade is where the 48-in. chord blade section attaches to the tower.
Predi ct ed and Measured RMS St r es s es - - The stress states at many locations along the
blades were predicted before the turbine was built using a beam finite-element model adapted
at Sandia to perform frequency-response analysis for VAWT structures [27]. The analysis
gives the magnitudes of stress responses at several driving frequencies. The RMS for the
sum of the sinusoids at all the driving frequencies is used as an estimate of the RMS of the
random response. The stress states for the upper root were measured using strain gages l~
located 36 in. from the blade root [28]. Based on a finite element analysis of this region,
the measured stresses had to be multiplied by factors of 1.08 for lead-lag and 1.23 for flatwise
to adjust them to the maximum nominal (section) stress in the root joint [12,21 and 22]. The
RMS is calculated from the time histories of the local stresses after removal of the mean.
The measured and predicted flatwise RMS stresses, for fixed speed operation at 28, 34, and
10 Strain gages are used in pairs on the blades as bending load sensors. The gages are wired in
Wheatstone bridges to add bending and cancel axial and torsional loadings. Known bending moments
were applied to the blades before the turbine was assembled to calibrate the moment measuring sensors.
Nominal stress levels at the maximum fiber are then estimated by the beam bending formula Mc/1 are
the standard blade-bending output from the instrumentation system.

~ 10
0 ! !
2 5
_ Constant Speed Operation
- _L m_
/
- 34 rpm / / / Z' "
- 38 rpm , , , / / " "
- . . . . , , ~ . . " .." .'"
- Variable Speed Operation
/. ~176176176
_ ,.~:"
. . ' 2 "
~ ~ i
J t "~176 j . '
j , f o- . s"
O
~ ~176176176176
. . . . ! . . . . I i 9 9 9 I a t a a !
0 5 1 0 1 5 2 0
Wind Speed, m/s
FIG. 16 Measured flatwise RMS stresses at the upper root.
38 rpm, are shown in Fi gs. 16 and 17, respect i vel y. Compar i son of t he measured and pre-
di ct ed RMS st resses for vari abl e speed operat i on is shown in Fi g. 18. The l ead-l ag stresses
are not shown here, but are r epr oduced in Ref 22.
St ress Concent rat i on Fa c t o r - - A stress concent rat i on fact or ( SCF) was appl i ed to the max-
i mum nomi nal bendi ng stress to account for stress ri sers in the bol t ed r oot j oi nt . The exact
SCF for the upper r oot j oi nt is unknown, but is expect ed to be t hree or more. Ear l y fatigue
anal yses suggest ed that t hese stress l evel s were sufficient for the appl i cat i on (that is, a one
of a ki nd research machi ne wi t h frequent i nspect i ons) so no det ai l ed stress anal ysi s or com-
ponent testing was done.
Cycl e Ra t e - - T h e rate at whi ch stress cycl es are accumul at ed is a necessar y par amet er for
det ermi ni ng ser vi ce l i fet i me when usi ng the nar r ow- band Gaussi an model [20]. For t hese
predi ct i ons, the est i mat ed rat e for t he pr edi ct ed RMS stresses was appr oxi mat ed at t hree
t i mes the rot at i onal rat e of the t urbi ne for the l ead- l ag stress cycl es and t wo t i mes for flatwise
stress cycl es. 1~ For the measurement s, this rate was det er mi ned usi ng t he average "mean-
cr ossi ng" rat e [29]. The di fferences were smal l , except at 28 rpm, wher e the measured
flatwise cycl e rat e was appr oxi mat el y 50% hi gher t han the est i mat ed rate.
l~ These cycle rates are selected based on the interaction of the frequency content of the aerodynamic
loading and the structural dynamics of the turbine rotor.

15
/
I. Constant Speed Operation /
t . . . . , ' "
34rpm / / /
38rpm / /
10 1- - " Variable Speed Operation ~ i i i i i I"
~ 9 -
s . . . . . f ,
. 0 . 9
I 9 , , , I 9 9 9 9 I 9 9 9 , I 9 9 9 9
0 5 10 15 20 25
Wi nd Speed, m/s
F I G. 17- - P r edi cted fl atwi se RMS stresses at the upper root.
Operational States
This variable-speed turbine has been operated at fixed speeds throughout its operating
range of 18 to 38 rpm and in a continuously variable-speed mode. Variable speed is used to
track the highest aerodynamic efficiency, increasing rotor speed with wind speed until the
maximum design speed is reached. The turbine operates at maximum design speed until
increasing wind would overpower the generator. The turbine speed is then reduced to limit
power. The mapping between turbine speed and wind speed is shown in Fig. 19 [30].
Baseline Life Estimates
To establish a reference frame for this parametric sensitivity study, two baseline sets of
parameters are used to calculate reference service lifetimes for the upper root joint. The first
baseline case represents the limited data available early in the design process. The wind
speed distribution is taken to be Rayleigh distributed with a 6.3 m/ s mean. The stress states
are assumed to follow the narrow-band Gaussian model with analytically predicted RMS
versus wind speed behavior. The reference S-n curve is used to define the material fatigue
properties. The second baseline uses measured quantities from turbine operating experience.
The wind speed distribution is for the Bushland site. The RMS stress levels are as measured

15
M e a su re d S tre sse s
. , 7
A n a ly tic a l S tre sse s / , ' / /
i
,o / /
/ /
. . . . | - | V a r i a b | l e S p e e d
0
0 5 10 15 20 25
Wind Speed, m/s
FIG. 18--Comparison of the measured and predicted RMS flatwise stresses with variable speed
i 3 5
30
20
| 9 9 9 9 9 9 9 9
operation.
4 0
5 10 15 20
Wind Speed, m/s
FIG. 19--Mapping between turbine speed and wind speed f or variable speed operation. Copyright by ASTM Int'l (all rights reserved); Thu May 8 10:42:12 EDT 2014

TABLE 1--Key for the tables.
Wnd Annual Wind Speed Distribution
9 R--Rayleigh distribution with 14 mph average
9 A--distribution for the Amarillo airport, 17 years of data
9 B--distribution at the Bushland test site
Fat--S-n Fatigue Data for Extruded 6063 Aluminum
9 Ref--reference data
9 LSC--least square-curve-fit to experimental data
9 95%---curve fit of experimental-data at 95% survival level
9 Hgh---extrapolation of the high-stress experimental data
OPS---Operation Stress States
9 Ana--analytical predictions for the stress states
9 Mea--measured stress states
9 Flat--flatwise stresses
9 L-L--lead-lag stresses
and adjusted to the maximum nominal root stress. The S-n data are the LSC fit to the coupon
tests. Three fixed operating speeds, 28, 34, and 38 rpm are used, as well as a variable speed
case. The SCF is taken to be 3.0 and cycle rates are estimated for both cases. A key to the
abbreviations used in the case study is given in Table 1. The lifetimes, in years, calculated
using the LIFE2 code, are shown in Table 2.
P a r a m e t e r S e n s i t i v i t y
Wind Regime Variation
Table 3 lists the estimated fatigue lives, as in Table 2, but applying each of the three
candidate wind speed distributions to each baseline case. Notice that the Amarillo and Bush-
land wind distributions produce lead-lag predictions that vary by a factor of 10.
S-n Data
The four different S-n estimates from Fig. 15 are applied to the baseline cases in Table
4. Although different treatments of the test data result in differences up to a factor of 2, the
reference data is more conservative by factors of 3 to 6.
Stress States
Table 5 lists results for the difference between analytical estimates of the RMS stress and
measured RMS stresses. Although the fit between analysis and measurement shown in Fig.
TABLE 2--Reference service lifetimes (in years)for the case study.
Flatwise Stress States Lead-Lag Stress States
Wnd Fat OPS 28 rpm 34 rpm 38 rpm Var 28 rpm 34 rpm 38 rpm Var
R REF Ana 1360 33.1 2.21 11.9 61.1 205 000 0.91 19.40
B LSC Mea 966 504 286 391 296 000 9 470 000 73.9 494

SUTHERLAND ET AL. ON WIND TURBINES
TABLE 3--Ef f ect of the wind regime on lifetime (in years).
Lifetime
Wnd Fat OPS Flat L-L
R Ref Ana 11.9 19.4
A Ref Ana 7.86 10.1
B Ref Ana 29.4 109
R LSC Mea 150 125
A LSC Mea I00 71.0
B LSC Mea 391 494
199
18 is as good as any analyst could expect, the fatigue life predictions vary by about a factor
of 3.
St ress Concent rat i on Fact or ( SCF)
Finally, the SCF is varied only slightly from 3.0 to 3.5 to illustrate its impact of fatigue
lifetime; results are shown in Table 6. Notice that the change in lifetimes ranges from a low
of a factor of 4 to a high of a factor of 20. This sensitivity is the highest found in the case
study and is, interestingly, on the parameter about which the least is known about the joint.
It should be noted that for the shortest lifetimes predicted in the study, some of the assump-
tions that rely on the high-cycle, nominally elastic nature of the material response may be
violated.
Questions
As we noted in the beginning of this paper, many questions about the determination of
service lifetimes for wind turbine components have not been answered. In addition to those
questions raised previously, the reader should consider these additional questions.
9 With the variability in lifetime estimates shown in the case study, how would you
present your fatigue life predictions to the corporate executives who need to decide
whether to fund an initial production run of these turbines? What additional work do
you recommend to reduce the financial risk?
TABLE 4--Ef f ect of the constitutive formulation on lifetime
(in years).
Lifetime
R Ref Ana 11.9 19.4
R LSC Ana 59.7 149
R 95% Ana 35.7 70.3
R Hgh Ana 52.8 135
B Ref Mea 104 83.7
B LSC Mea 391 494
B 95% Mea 243 249
B Hgh Mea 307 430

TABLE 5--Ef f ect of the operational stresses on lifetime
(in years).
Lifetime
R Ref Ana 11.9 19.4
R Ref Mea 33.5 15.7
B LSC Ana 121 602
B LSC Mea 391 494
9 Cri t i que t he anal ysi s pr ecedur e pr esent ed here. What assumpt i ons have the pot ent i al to
pr oduce t he great est error in l i f e pr edi ct i on? How woul d you i mpr ove t he procedure?
What st eps woul d you t ake first? Second?
9 Coul d you come up wi t h a cl osed f or m sol ut i on for t he (approxi mat e) t i me to fai l ure?
What assumpt i ons do you consi der appropri at e? How much error woul d t hey pr oduce?
Is it accept abl e? Is i t conservat i ve?
9 Woul d you expect di fferent l i f et i mes for i dent i cal machi nes in di fferent l ocat i ons? Why
or why not ? I f so, by how much?
Ob s e r v a t i o n s o n t he Ut i l i t y o f t he Re s u l t s
As obser ved in t he case study pr esent ed here, fat i gue l i fe est i mat es for wi nd turbines can
(and t ypi cal l y do) have a l arge amount of uncert ai nt y associ at ed wi t h them. Mi ner ' s rul e
may be off by a fact or of 2 or more, hi gh or low, in the predi ct i on of servi ce lifetimes. As
i l l ust rat ed by this case study, measur ed wi nd speed di st ri but i ons from rel at i vel y cl ose sites
can change l i fe est i mat es by a fact or of 4 up to a fact or of 8. And, a change of 15 percent
in t he r oot stress can t ransl at e i nt o mor e than an order of magni t ude change in t he l i fe
est i mat e. Wi t h such uncert ai nt y in t he est i mat i on, one must wonder whet her t he performance
of a fatigue anal ysi s is even j ust i f i ed in t he desi gn phase of a wi nd turbine.
The goal of fat i gue-cri t i cal engi neer i ng is to desi gn component s t hat can endure t hei r
l oadi ng envi ronment for t he desi gn l i f et i me and to be economi cal l y compet i t i ve. (As di s-
cussed above, economi cs is one of the mai n drivers in a vi abl e desi gn for wi nd t urbi nes. )
Wi t h t he uncert ai nt i es ci t ed here, fat i gue anal yses find t hei r great est ut i l i t y in pr ovi di ng order-
of - magni t ude l i fe est i mat es for t urbi ne component s and rel at i ve l i fe est i mat es for compet i ng
desi gn opt i ons; that is, t hey i dent i fy desi gn defi ci enci es and per mi t t r ade- of f anal yses of
TABLE 6--Ef f ect of the stress concentration factor on lifetime
(in years).
Lifetime
Wnd Fat OPS SCF Flat L-L
R Ref Ana 3.0 11.9 19.4
R Ref Ana 3.5 1.53 0.52
B LSC Mea 3.0 391 494
B LSC Mea 3.5 108 37.6

various design options. Thus, even with all of its problems, the estimated service lifetime is
important to the design process.
As noted by Veers et al. [31], maybe the designer should be asking the question: "What
is the probability that the turbine design will meet its designed service lifetime?" rather than
asking "What is the service lifetime of this turbine?" The former question requires the
designer to evaluate the uncertainties and inherent randomness of the input variables used
in the fatigue analysis. For this case study, a median lifetime of hundreds of years still yields
a probability for premature failure of over 2% for a 20-year lifetime [31]. Thus, the designer
must "over-design" the turbine to account for the lack of knowledge of the load spectrum
and material properties for this analysis. Only with an improved state of knowledge from
turbine testing, turbine operating experience, and improved material characterizations will
the designer be able to refine the turbine design and increase its reliability.
Concl usi ons
Wind turbines are subjected to a severe and unrelenting environment driving the materials
to their limits of fatigue endurance. The loadings are random in nature and continuously
fluctuating in both cyclic amplitude and global intensity. Formulating the problem requires
breaking it down into manageable pieces while making simplifying assumptions to permit
tractable solutions. The procedure developed at Sandia National Laboratories is presente d
here as a case study. It is neither perfect nor exhaustive, but serves to illustrate how sense
can be made out of complete randomness. The LIFE2 fatigue and fracture analysis code
used for the calculations in the case study is explained in an Appendix A and is available
from the authors, complete with user' s manuals and supporting documentation. The case
study illustrates the tremendous variability in life predictions that can occur with relatively
modest changes in turbine placement, stress analysis results, or assumptions on uncertain
inputs. With the LIFE2 code [6-11], additional studies or specific problem assignments can
be formulated to lead students through the process of cumulative damage summation and to
demonstrate the range of life estimates that will result from parameter variations.
Acknowledgments
This work is supported by the U.S. Department of Energy at Sandia National Laboratories
under contract DE-AC04-94AL85000.
APPENDI X A
The LIFE2 Code
The LIFE2 code [6-11] is a fatigue/fracture mechanics code that is specialized to the
analysis of wind turbine components. It is a PC-based, menu-driven FORTRAN code that
is written in a top-down modular format. The package leads the user through the input
definitions required to predict the service lifetime of a turbine component. In the current
formulation, the service lifetime of turbine components may be predicted using either Miner' s
rule with constant amplitude S-n data or a linear-elastic crack propagation rule [8]. The code
is described in Refs 6 to 9, and a complete set of user' s manuals are provided in Refs 7, 8,
10, and 11. The code does have the ability to plot input and output variables.
The LIFE2 code requires four sets of input for the prediction of the service lifetime for a
turbine component: (1) the annual wind speed distribution at the site under investigation, (2)

an S - n di agram (fatigue data) or crack growth rate data (linear crack propagation) for the
material comprising the component under investigation, (3) the stress cycles of the total
stress states on the component as a function of the operating state of the turbine and the
average wi nd speed, and (4) the operating parameters and stress concentration factor(s) for
the component.
Nume r i c a l Fo r mu l a t i o n
The LI FE2 code is based on the assumption that the total set of fatigue cycles on the
turbine is the sum of a set of cycl e distributions (histograms) characterized by wind speed
and operating conditions (that is, parked or rotating). The analysis proceeds by integrating
the fatigue cycles imposed on the turbine using a damage accumul at i on model.
To illustrate this formulation, we use Mi ner' s rule t o describe the accumulation of damage
i n' a component during normal operation of the turbine. Mi ner' s rule states that the damage
accumulated from n stress cycles at stress state S is determined by the following relation
n(S)
D = i ( A- l )
N ( S )
where N is the number of stress cycles at stress state S required to fall the material. The
damage is then integrated (summed) over all stress states to determine the total damage
accumulated in the component. The component fails when the damage D sums to one.
Typically, the stress cycles at stress state S are written as a function of the cycl es' mean and
alternating stress components. Thus, the integration of the damage relation over all stress
states may be written as
( ~f oo
r t ( S m, Sa)
D I v = ~ 1 d S m d S a (A-2)
N(Sm, Sa)
JO J - o o
For the wind turbine, we assume that each stress cycl e may be written as function of the
input wind speed; thus, Eq A-2 is the damage at a given wind speed, V. When integrated
over the distribution of wi nd speeds imposed upon the turbine, the total damage becomes
fv,. -~ f v , . " f ~ f f ~ n(S'n' s a ' V ) d S m d S ' d V
O = , D]v p v ( V ) d V = , P v N(Sm ' Sa )
(A-3)
where Vci and Vco a r e the cut-in and cut-out wind speeds; that is, the wi nd speed at which
the turbine is turned on and the wind speed above which the turbine is no longer operated.
The wind speed distribution is typically described using the probability density function P v
for the annual wi nd speed distribution. Thus, for a time of At, the damage accumulated by
the operating wi nd turbine equals
fv,. ~ f f f f = n(Sm' S~' V ' A T ) d S m d S , , d V
D ( A t ) = , P v N(Sm, Sa)
(A-4)
and the predicted service lifetime for the turbine component, T f in years, is given by
1
T s -- ~ for At = 1 year (A-5)
The LIFE2 code comput es these relations numerically. Similar descriptions are developed
for other aspects of turbine operation and wind conditions (for example, the stress cycles
produced by starting and stopping the turbine and by high winds "buffet i ng" a stopped

SUTHERLAND ET AL. ON WIND TURBINES 2 0 3
t urbi ne) and for ot her damage cri t eri a (l i near crack propagat i on). A compl et e descr i pt i on of
the al gori t hms used in the code are gi ven in Ref 8.
Input Variables
As ment i oned above, t he LI FE2 code requi res f our sets of i nput i nformat i on to det er mi ne
servi ce l i fet i mes. They are (1) t he annual wi nd speed di st ri but i on, (2) the const i t ut i ve equa-
tion for the damage l aw bei ng used, (3) t he cycl i c stresses as a funct i on of wi nd speed and
oper at i ng condi t i ons of t he turbine, and (4) t he paramet ers descr i bi ng t he oper at i on of the
turbine. Each of t hese i nput s is descr i bed in det ai l in t he case st udy present ed above. Each
i nput may be descr i bed usi ng t abul ar dat a or speci al i zed i nput functions.
St ress St at es
The most difficult dat a set to obt ai n for i nput i nt o the LI FE2 code is the cycl e count
funct i on n(S=, So) (Eqs 1 t hrough 5). For the LI FE2 code, t hi s function t akes t he form of a
series of cycl e count mat ri ces. Each mat r i x is used to descr i be a si ngl e aspect of t urbi ne
operat i on.
The first cl ass of the cycl e count mat r i ces used by the LI FE2 code is for t he "oper at i onal
st resses, " namel y, t he stresses that occur duri ng operat i on of t he t urbi ne under nor mal wi nd
condi t i ons. These cycl e counts for this oper at i on state are di vi ded into a series of cycl e count
mat ri ces; each defi nes t he stress cycl es as a function o f wi nd speed. The cycl e count mat ri ces
must be defi ned for sufficient wi nd speed to cover the ent i re operat i ng range of t he turbine.
The second and t hi rd cl asses of mat ri ces descri be speci al events and l oads t hat occur when
t he t urbi ne is not operat i ng. The second cl ass cont ai ns cycl e count mat ri ces for st art -st op
cycl es, emer gency stops, l oss- of - gr i d stops, and so forth. The t hi rd cl ass cont ai ns mat ri ces
for hi gh- wi nd events; t hat is, the stress cycl es i nduced by "buf f et i ng" of the t urbi ne by hi gh
wi nds when it is parked.
Dat a I nput
Al l t hree cl asses of cycl e count mat ri ces may be ent ered i nt o t he LI FE2 code usi ng several
i nput techniques. For al l of the cases, i f the cycl e counts are known f r om ot her anal yses or
measurement s, t hey may be ent ered di r ect l y into the code. Ti me series dat a f r om ei t her
measurement s or si mul at i ons from t i me- domai n structural anal ysi s codes 12 may be used also.
For this case, the stress hi st ori es are eht ered vi a a dat a file. The LI FE2 code uses a rai nfl ow
count i ng al gori t hm [15] t o process t he dat a i nt o t he appr opr i at e cycl e count mat r i ces [9-11].
Fr equency domai n i nput dat a for the oper at i onal stresses may be used by t he LI FE2 code
as wel l [32,33]. The spect ral stress r esponse of the structure is ent ered vi a a dat a file. The
LI FE2 code uses an i nverse fast fouri er t ransform ( FFF) to t ransform the frequency domai n
dat a into the t i me domai n. The rai nfl ow count i ng al gori t hm, ci t ed above, is t hen used to
cycl e count the data.
Nar r ow Band Gaus s i an Mo d e l
Thi s model uses a Rayl ei gh di st ri but i on defined by t he r oot - mean- squar e ( RMS) of the
stress t i me series to defi ne cycl e ampl i t udes. To use this model in t he LI FE2 code for the
12 As discussed in the opening remarks, the analysis of wind turbines using structural analysis codes
is a tedious and difficult problem. The input loads to the codes must be determined first using aerody-
namics codes that include the effects of turbulence in the incoming wind. From these loads the structural
dynamics codes must determine the stresses in the structure. For VAWT structures, successful simulations
have been conducted in the frequency domain for "quasi-steady" input conditions [21]. Transient events
[34] and the analysis of HAWT structures [35] require time domain solutions.

oper at i onal stresses, the RMS par amet er must be defi ned as a funct i on of wi nd speed. The
code accept s t hi s funct i on in t abul ar form. To compl et e the model i ng of the cycl e counts,
the mean stress and t he cycl e rat e are assumed to be a const ant for al l cycl es in the
di st ri but i on.
Computational Modules
The code is wri t t en in a t op- down modul ar f or mat wi t h an "execut i ve modul e" that con-
t rol s t he overal l operat i on of t he code and t he five mai n comput at i onal modul es. The first
four defi ne t he i nput par amet er s and dat a sets, and t he fifth modul e performs t he damage
cal cul at i on. Auxi l i ar y modul es per mi t dat a to be pl ot t ed and st ored for future reference. Dat a
may be pl ot t ed on a vi deo di spl ay or sent to a printer. Al l common graphi cs drivers are
support ed. A menu syst em is used to faci l i t at e movement about t he code and t o i nput dat a
i nt o the code.
Platform
The LI FE2 code is a PC- compat i bl e FORTRAN code. Bot h t he source code and an exe-
cut abl e file are avai l abl e. The code adapt s, vi a i nput vari abl es, to vari ous vi deo and pri nt er
confi gurat i ons. The code requi res 640 kbyt es of memory, and a har d di sk is r ecommended.
Current l y, the LI FE2 execut abl e files occupy appr oxi mat el y 0.5 Mbyt e. The source files
occupy al most 1 Mbyt e. Fi l es to descri be a t ypi cal pr obl em occupy several Mbyt es.
APPENDIX B
Material Data for 6063 Aluminum
The fat i gue anal ysi s of t urbi ne structures usi ng Mi ner ' s rul e requi res a det ai l ed knowl edge
of the fat i gue propert i es of the mat er i al under anal ysi s. For 6063-T5 al umi num, the mat eri al
used in the bl ades of t he 34- m Test Bed VAWT, an appropri at e dat a base was not avai l abl e
i n t he l i t erat ure. The necessary dat a base was obt ai ned by Van Den Avyl e and Sut herl and
[13] and Mohamadi an and Gr aham 13 usi ng st andard ASTM t echni ques.
The 6063-T5 al umi num had a measured yi el d stress of 205 MPa and a measured ul t i mat e
stress of 244 MPa. The S-n dat a base, requi red for Mi ner ' s rul e damage law, was obt ai ned
usi ng appr oxi mat el y 100 bend speci mens cycl ed at five al t ernat i ng stress ampl i t udes and at
four mean stress levels. A summar y of the raw dat a from t hese exper i ment s is l i st ed in Table
B- l , whi ch fol l ows. Thi s t abl e cont ai ns t wo dat a sets: t he first set was obt ai ned by Van Den
Avyl e and Sut herl and and the second was obt ai ned by Mohamadi an and Graham. The former
dat a are not ed by t he number ed speci mens and the l at t er are not ed by speci mens l abel ed
wi t h t he l et t er " S" for Sout hern University. The final 8 dat a ent ri es in this t abl e l i st dat a
from speci mens that di d not fai l duri ng the experi ment s. These "r un- out " fat i gue tests were
t ermi nat ed at 5 x 108 (500 000 000) stress cycl es.
For our anal ysi s of t hese dat a, we chose to fit t he dat a wi t h a Goodman Fi t usi ng the
ul t i mat e st rengt h of t he mat er i al S,. Thi s rul e states that the fat i gue l i fe at al t ernat i ng stress
S a and mean stress S m is equal to t he fatigue l i fe at an equi val ent zer o- mean- st r ess al t ernat i ng
stress state of Se~r t hrough the rel at i on
13 Mohamadian, H. P. and Graham, I. J., Southern University, Baton Rouge, LA, private corre-
spondence.

SUTHERLAND ET AL. ON WIND TURBINES 2 0 5
ID
74
8O
75
73
25
23
24
85
26
S
S
S
S
8
10
9
6
S
S
S
S
2
3
5
S
S
S
4
9O
89
68
87
82
84
81
83
66
68
65
67
27
28
29
3O
15
17
18
16
S
S
S
S
13
Mean
S n u
, ~ ( M ~=~
0.00
a
-0.34
i
0.00
i
0.00
i
-0.34
i
0.00
i
-1.03
!
0.03
!
0.34
|
0.00
i
0.00
i
0.00
=
0.03
|
0.34
!
0.00
i
-0.34
=
0.70
i
0.03
i
0.00
l
0.00
l
0.03
]
-0.34
I
-5.00
i
2.43
i
0.00
i
0.03
i
0.03
i
0.03
o
42.01
=
42.69
i
44.68
i
42.69
l
33.36
i
33.61
1
33.61
i
33.95
|
33.61
i
34.76
i
34.08
o
34.47
|
34.25
i
34.32
I
34.80
i
34.50
i ,
34.06
I
34.93
I
33.61
i
34.13
I
34.47
I
34.47
|
34.47
a
i 34.47
L 34.4o
T A B L E B - l - - S - n data f or 6063-75
Cyclic
( M P , , )
82.92
83.47
64.36
84.92
99.36
99.44
99.63
100.70
101.13
103.42
103.42
103.42
103.42
C y c m to
Fi JJm
11,99,4:103
22,963,803
39,155,3O0
17t149,200
3,492,700
4,420,9O0
81375,200
5,550,000
12,679,800
13189O,000
141182,030
161832,000
18,256,(XX)
116. 15 1, 594, 700
116. 58 1, 378, 200
116.93
119.18
11612,000
1,571,0(X)
137.90 1,1 63.030
137. 90 1, 532, 000
137. g0 1, 565, 000
137. 90 1, 884, 000
1 38.24 297,900
'144.6g 498,500
157.43 165,100
172.37 50,800
172.37 51,600
172.37 59,500
174.91 591700
65. 91 105,443,900
66.40 97, 376, 300
6 7 . 6 7 193,322,400
6980 134,251,700
70. 44 37,282,900
7 0 . 5 8 69, 989, 103
70. 58 100,409,400
7 0 . 9 2 39,839,300
8 0 . 6 7 9~529,400
81.32 18,437,900
8 1 . 7 4 19,026,200
81.77 5, 081, 000
9 9 . 3 1 2, 222, 400
100. 17 1, 965, 500
100. 92 1, 555, 200
102. 10 1, 983, 500
115.80 844,500
116.42 960,500
116.58 576,500
116.87 682.2O0
137.90 159,200
137.90 183,300
137.90 194,500
137.90 213,000
150.21 249,100
aluminum.
11 33.74
]
12 34.72
i
14 34.63
i
20 32. 74
|
19 34.68
i
21 34.32
|
22 35.05
e
91 71.97
i
71 67.15
i
72 67.24
i
69 68.69
=
45 69.60
i
47 69.22
]
48 69.79
i
48 70.17
i
44 68.55
|
42 69.11
|
41 68.67
=
43 69.11
31 68.35
|
35 69.44
38 69.79
|
151.48
151.62
151.91
171.g0
172.36
172.95
174.58
53.05
63.78
63.87
64.62
80.04
80.30
80.95
81.30
99.25
99.36
9857
100.05
117.37
118.29
118.64
128r808
134r700
179t200
80~600
70,400
64f3o0
63,300
211 ~987,300
18,978,500
Io~ggGr3OO
51,097,000
1,983,000
2r114,600
4,201,100
3,689,900
1r110r3oO
803r000
844,000
722,800
343,400
252,200
405,300
34 : 7 0 . 7 4 120. 26 412,800
38 I 68.33 130. 82 170~600
m
37 69. 04 131. 18 124,100
40 69. 06 132. 23 187,500
|
39 70. 26 133. 50 267,603
S 68. 95 165. 47 71,000
S , 68.95 172. 37 51,300
S , 68.95 172. 37 61,800
S 68.95 172. 37 70,900
|
64 100.97 6 0 . 5 8 10, 335, 100
| i
53 102.80 61.13 16, 116, 300
| =
60 102.53 6 2 . 0 7 3, 194, 800
i
52 102.39 6 2 . 6 7 49, 925, 900
= |
59 103.70 7 6 . 0 5 4, 060, 100
= |
62 , 103.63 7 6 . 0 8 1, 487, 100 i
61 , 104.59 7 6 . 2 9 2, 806, 500
63 , 105.60 7 6 . 8 4 2, 015, 200
55 , 102.87 87.09 1, 476, 500 !
58 , 103.70 8 7 . 1 1 1, 824, 500
56 103. 77 87.12 454,700
57 , 101.77 87.33 687f900 ,
54 101. 57 101. 57 429,500
1 =
51 , 102. 48 102. 46 5321200 ,
49 103. 64 103. 64 343,400
1 |
50 104. 25 104. 25 340,300
| ]
86 0.34 70. 81 500,000,000
93 68.82 53. 32 500,000,030
| |
94 66.83 51. 90 500,000,000
92 68.50 52. 30 500,000,000
| |
96 88.78 49. 31 500,000,000
| |
97 , 89.87 49. 91 500,0301000 ,
98 , 89.68 49.93 50010001000 ,
95 94.46 50. 58 500,000,000

Wh e n t hese dat a are mapped i nt o t he equi val ent stress state, t her e are t wo di st i nct r egi ons
to t he curve. We chose to fit each s egment wi t h a st rai ght l i ne on a l og- l og pl ot of t he form:
1Oglo (Serf) = a + b lOglo (n)
(B-2)
References
[1] Phillips, J. A., "Wi nd Power' s Coming of Age," The Electricity Journal, Vol. 5, No. 3 April 1992,
pp. 22-32.
[2] Annual Energy Outlook, 1993 with Projections to 2010, DOE/EIA-0383(93), Energy Information
Administrations, Jan. 1993.
[3] "A Growth Market in Wind Power," EPRI Journal, Vol. 17, No. 8, Dec. 1992, pp. 4-15.
[4] Coleman, C. and McNiff, B., Final Report: Dynamic Response Testing of the Northwind 100 Wind
Turbine, Subcontractor Report, SERI Cooperative Research Agreement DE-FC02-86CH10311, So-
lar Energy Research Institute, Golden CO, Dec. 1989.
[5] Ashwill, T. D., Berg, D. E., Gallo, L. R., Grover, R. D., Klimas, P. C., Ralph, M. E., Rumsey,
M. A., Stephenson, W. A., and Sutherland, H. J., "The Sandia 34-Meter VAWT Test Bed," in
Proceedings of WindPower '87, SERI/CP-217-3315, American Wind Energy Association (AWEA),
Washington, D.C., 1987, pp. 298-308.
[6] Sutherland, H. J. and Schluter, L. L., "The LIFE2 Computer Code, Numerical Formulation and
Input Parameters," in Proceedings of WindPower '89, SERI/TP-257-3628, American Wind Energy
Association, Washington, D.C., 1989, pp. 37--42.
[7] Schluter, L. L. and Sutherland, H. J., Reference Manual f or the LIFE2 Computer Code, SAND89-
1396, Sandia National Laboratories, Albuquerque, NM, Sept. 1989.
[8] Sutherland, H. J., Analytical Framework f or the LIFE2 Computer Code, SAND89-1397, Sandia
National Laboratories, Albuquerque, NM, Sept. 1989.
[9] Schluter, L. L. and Sutherland, H. J., "Rainflow Counting Algorithm for the LIFE2 Fatigue Analysis
Code," in Ninth ASME Wind Energy Symposium, SED-Vol 9, D. E. Berg, Ed., American Society
of Mechanical Engineers (ASME), New York, 1990, pp. 121-123.
[10] Schluter, L. L. and Sutherland, H. J., User's Guide f or LIFE2's Rainflow Counting Algorithm,
SAND90-2259, Sandia National Laboratories, Albuquerque, NM, Jan. 1991.
[11] Schluter, L. L., Programmer's Guide f or LIFE2's Rainflow Counting Algorithm, SAND90-2260,
Sandia National Laboratories, Albuquerque, NM, Jan. 1991.
[12] Ashwill, T. D., "Initial Structural Response Measurements for the Sandia 34-Meter VAWT Test
Bed," in Eighth ASME Wind Energy Symposium, SED-VoL 7, D. E. Berg and E C. Klimas, Eds.,
American Society of Mechanical Engineers (ASME), New York, 1989, pp. 285-292.
[13] Van Den Avyle, J. A. and Sutheriand, H. J., "Fatigue Characterization of a VAWT Blade Material,"
in Eighth ASME Wind Energy Symposium, SED-Vol. 7, D. E. Berg and E C. Klimas, Eds., American
Society of Mechanical Engineers (ASME), New York, 1989, pp. 125-129.
[14] Mandell, J. F., Reed, R. M., Samborsky, D. D., and Qiong, P., "Fatigue Performance of Wind
Turbine Blade Composite Materials," SED-Vol. 14, in Wind Energy--1993, S. Hock, Ed., ASME,
New York, 1993, pp. 191-198.
[15] Downing, S. D. and Socie, D. E, "Simple Rainflow Counting Algorithms," International Journal
of Fatigue, Vol. 4, No. 1, 1982, pp. 31-40.
[16] Osgood, C. C., Fatigue Design, 2nd Edition, Pergamon, Oxford, 1982.
[17] Barchet, W. R., "Wi nd Energy Data Base," in Proceedings of the Fifth Biennial Wind Energy
Conference and Workshop I1, SERI/CP-635-1340, Solar Energy Research Institute, Golden, CO,
1981.
[18] Malcolm, D. J., "Prediction of Peak Fatigue Stresses in a Darrieus Rotor Wind Turine under
Turbulent Winds," in Ninth ASME Wind Energy Symposium, SED-Vol 9, ASME, 1990, pp. 125-
136.
[19] Veers, P. S., "Simplified Fatigue Damage and Crack Growth Calculations for Wind Turbines," in
Eighth ASME Wind Energy Symposium, SED-Vol. 7, D. E. Berg and P. C. Klimas, Eds., ASME,
New York, 1989, pp. 133-140.
[20] Veers, P. S., A General Method f or Fatigue Analysis of Vertical Axis Wind Turbine Blades, SAND89-
2543, Sandia National Laboratories, Albuquerque, NM, 1983,
[21] Ashwill, T. D., Suthefland, H. J., and Veers, E S., "Fatigue Analysis of the Sandia 34-Meter Vertical
Axis Wind Turbine," in Ninth ASME Wind Energy Symposium, SED-Vol. 9, ASME, New York,
1990, pp. 145-151.

[22] Ashwill, T. D. and Veers, E S., "Structural Response Measurements and Predictions for the Sandia
34-Meter Test Bed, " in Ninth ASME Wind Energy Symposium, SED-Vol. 9, ASME, New York,
1990, pp. 137-144.
[23] Selected Papers on Wind Energy Technology, January 1989-January 1990, P. S. Veers, Ed.,
SAND90-1615, Sandia National Laboratories, Albuquerque, NM, 1990.
[24] Elliott, D. L., Holladay, C. G., Barchet, W. R., Foote, H. P., and Sandusky, W. E, "Wind Energy
Resource Atlas of the United States," DOE/CH10093-4, DE86004442, Pacific Northwest Labora-
tory, March 1987.
[25] Sutherland, H. J., Ashwill, T. D., and Slack, N., "The LIFE Computer Code: Fatigue Life Prediction
for Vertical Axis Wind Turbine Components," SAND87-0792, Sandia National Laboratories, Al-
buquerque, NM, 1987.
[26] Mitchell, M. R., "Fundamentals of Modem Fatigue Analysis," in Fatigue and Microstructure,
American Society for Metals, Metals Park, OH, 1979, pp. 385-437.
[27] Lobitz, D. W. and Sullivan, W. N., Comparison of Finite Element Predictions and Experimental
Data for the Forced Response of the DOE 100 kW Vertical Axis Wind Turbine, SAND82-2534,
Sandia National Laboratories, Albuquerque, NM, 1984.
[28] Sutherland, H. J. and Stephenson, W. A., Rotor Instrumentation Circuits for the Sandia 34-Meter
Vertical Axis Wind Turbine, SAND88-1144, Sandia National Laboratories, Albuquerque, NM, 1988.
[29] Crandall, S. H. and Mark, W. D., Random Vibration in Mechanical Systems, Academic Press, New
York, NY, 1963.
[30] Ralph, M. E., "Control of the Variable Speed Generator on the Sandia 34-Metre Vertical Axis Wind
Turbine," in Proceedings of WindPower '89, SERI/TP-257-3628, Solar Energy Research Institute,
Golden, CO, Sept. 1989, pp. 99-104.
[31] Veers, P. S., Sutherland, H. J., and Ashwill, T. D., "Fatigue Life Variability and Reliability Analysis
of a Wind Turbine Blade," Probabilistic Mechanics and Structural and Geotechnical Reliability,
Y. K. Lin, Ed., ASCE, American Society of Civil Engineers, New York, July 1992, pp. 424-427.
[32] Sutherland, H. J. and Osgood, R. M., "Frequency-Domain Synthesis of the Fatigue Load Spectrum
for the NPS 100-kW Wind Turbine," in Proceedings of WindPower '92, American Wind Energy
Association, (AWEA), Washington, DC, Oct. 1992, pp. 321-328.
[33] Sutherland, H. J., "Effect of the Flap and Edgewise Bending Moment Phase Relationships on the
Fatigue Loads of a Typical HAWT," in Wind Energy--1993, SED-Vol. 14, S. Hock, Ed., ASME,
New York, 1993, pp. 181-187.
[34] Dohrmann, C. R. and Veers, P. S., "Ti me Domain Structural Response Calculations for Vertical
Axis Wind Turbines," in Eighth ASME Wind Energy Symposium, SED-Vol 7, D. E. Berg and P. C.
Klimas, Eds., ASME, New York, 1989, pp. 107-114.
[35] Wright, A. D., Buhl, M. L., and Thresher, R. W., FlAP Code Development and Validation,
SERI/TR-217-3125, Solar Energy Research Institute, Golden, CO, 1988.

George H. Kyanka l
Fatigue Cases Involving
and a Wood Composite
the Use of Wood
RE F E RE NCE : Kyanka, G. H. , " Fa t i g u e Ca s e s I nv o l v i ng t he Us e o f Wo o d a n d a Wo o d
Co mpo s i t e , " Case Studies for Fatigue Education, ASTM STP 1250, Ralph I. Stephens, Ed.,
ABS TRACT: Many wood and wood-composite structural elements are exposed to fatigue
loading situations in service. Because wood is composed of a complex cellulosic fiber-based
structure at the microscopic level, fatigue cracking phenomena are not commonly observed.
Fatigue-related failures can look just like static failures. Fatigue cracking does occur under
certain types of loading, however, and some examples are presented.
The special properties of wood are briefly examined to show why fatigue fracture/cumulative
damage is so difficult to interpret in wood and wood products. Some historical insights are
covered. The development of special tests to gather fatigue data for wood is discussed to further
point out the unique concerns of designers working with the material.
Two case studies involving parallel to the grain fatigue fractures are presented. The case of
a failed shotgun stock represents a regular repeated load situation on a component that was
redesigned. Broken necks in bowling pins are due to a more complex dynamic loading regime,
but exhibit similar cracking patterns in a laminated composite product.
KEYWORDS: wood, cellular structure, orthotropic properties, cellulose, grain direction
Wood is, by vol ume, t he most wi del y us ed const r uct i on mat er i al i n the wor l d i n sol i d and
conver t ed forms. Most engi neer i ng st udent s, however, are t aught very little about t he struc-
t ural per f or mance of wood i n service. Very l i t t l e i nf or mat i on on propert i es such as fat i gue
per f or mance is f ound i n c o mmo n texts and r ef er ences for engi neer i ng mat eri al s. Even suc-
cessful desi gner s such as Dr. Fokker, t he f amed ai rcraft engi neer of Wor l d War I, expressed
t he i dea t hat wood is, i n fact, i mmu n e to fat i gue [1]. The t rut h is t hat wood, l i ke ot her
mat er i al s, can sust ai n st ruct ural damage due to repet i t i ve l oadi ng, and can ul t i mat el y fail i n
fat i gue due to such l oads [2]. The mode of fract ure is of t en mi sunder st ood, si nce fat i gue
fract ures i n wood and wood composi t es can l ook j us t l i ke st at i c l oad i nduced fractures. To
expl ai n how t hi s can be t rue, it is necessar y to l ook at some uni que st ruct ural charact eri st i cs
at t he mi cr oscopi c l evel and above.
The st ruct ure of wood at t he vi si bl e l evel is ort hot ropi c, t he t hree pr i nci pal axes bei ng t he
l ongi t udi nal , radi al , and t angent i al . Co mmo n usage is that t hese are al ong t he grai n or across
t he grai n of wood. The gr ai n (Fig. 1) is def i ned by rows of cel l s of var yi ng sizes and
or i ent at i on (Fig. 2) that are ar r anged i n repet i t i ve pat t erns. Each cell is made up of l ayers
of mi cr of i br i l s of cel l ul ose bound i n a mat r i x of l i gni n, a compl ex pol ymer i c adhesi ve ma-
t eri al . Si nce all of t he component s are pol ymer s, t hey exhi bi t a combi nat i on of el ast i c and
vi scoel ast i c behavi or. The assembl age is qui t e compl i cat ed, as seen i n Fig. 3.
The pr esence of a l arge numbe r of hol es, densi t y gradi ent s, and vari abl e cel l sizes creates
Professor, SUNY College of Environmental Science and Forestry, Faculty of Wood Products Engi-
neering, Syracuse, NY 13210-2786.
208

KYANKA ON THE USE OF WOOD AND A WOOD COMPOSITE 209
FIG. 1--Fi ber direction and principal planes in wood.
a system of crack absorbing and deflecting mechanisms such that most crack fronts only
move a small distance before stopping or changing form. In crystalline metals, the front
assembles itself and progresses as loading continues. In wood, much local damage may occur
before anything visible appears. Wood can provide for six different crack propagation pat-
terns in fatigue, leading to a Mode I fracture (Fig. 4). Therefore, it is necessary to evaluate
each direction for its significance in a fatigue situation.
The LR and LT cracks are those associated with longitudinal, or along the grain, loads.
These are the ones that are easiest to arrest and that are normal to the strongest direction in
wood. The broken tennis racquet in Fig. 5 shows how hard it is to create failures in these
planes. After 1.6 million serves, it shows a large number of cracks with no complete sepa-
ration. This type of fracture would occur if it was caused by a static bending load normal
to the racquet face. There is no identifiable point or points of origin for the cracks.
The TR and RT modes of fracture in wood are easier to create due to the cellular structure
of wood, but are not of great concern since wood is rarely loaded to any large stress levels
in these directions. One major exception to this occurs when large torsional loads are put
on long poles, such as those used to carry electrical service lines. Such twisting can produce
stresses that progressively break down the wood along concentric "shells" from the center
to the outside diameter. This can lead to failures in high winds due to loss of bending strength
in the pole. A few such cases have been documented, but this mode of fracture is not common
[3].
The easiest direction for a propagating crack to move is one along the RL or TL directions,
or somewhere between them. These are the crack planes used to easily split firewood. The
wood fibers are parallel to the crack and the large cell openings can serve to channel the
crack along. There are fewer arrest sites in these directions, and the properties of wood
strength perpendicular to its grain direction are its weakest. Most wood design manuals refer

FIG. 2--Cross section of wood showing fibers and arrangement (maple magnified approximately
250x).
to wood in tension perpendicular to grain as a condition to be avoided in designing. Some-
times the stresses cannot be avoided, and this is where our case studies are taken from.
It is interesting to note that fatigue data for wood has been largely generated by tests
made specifically for wood. One such test is cantilever bending of a beam by an electric
motor and eccentric drive arm. The test does not result in fracture of the beam, but rather
is called a failure when the damage in the wood reaches a point such that the elastic curve
of the bent beam changes due to cell wall damage [4]. This is in keeping with the earlier
discussion of how hard it is to cause LR or LT cracks to propagate.
Case Studies
Case 1
As part of manufacturer' s product verification tests for new shotgun models, most man-
ufacturers require a number of bench test firings. The guns are held in a vise and mechan-
ically triggered. After the number of required firings the model is then checked for wear and

FIG. 3--Radial view of persimmon wood (approximately 200
so forth. The fatigue incident occurred when a gun snapped off the front (fore) stock during
test and flipped upward upon a firing. The lower half of the stock was still in the vise.
Subsequent testing of similar guns was halted and the used models were checked closely to
see if there were similar problems. Several stocks were noted to have cracks running along
the grain (TL or RL), starting at the front edge of the wood (Figs. 6 and 7).
The model being tested was a semiautomatic 12-gage shotgun. The cartridges used were
ejected and replaced through the use of the gas pressure from the previous round. These
gases were routed through a cylinder that ran under the barrel. This cylinder entered the
forestock where the cracks were noted.

FIG. 4--Planes of crack propagation in wood.
FIG. 5--Wood tennis racquet broken in fatigue tester after 1.6 million serves.

FIG. 6--Walnut gun stock showing fatigue crack.
FIG. 7--View of fatigue crack in Fig. 6 showing propagation into stock.

The rear (toward the user) end of t he f or est ock was not t i ght l y at t ached to t he gas cylinder.
There was no ot her t ype of force appl i ed to the st ock ot her than the pressure and recoi l
effects. The st ock was made from Bl ack Wal nut (Juglans nigra) suppl i ed from mi dwest
pl ant at i on stock.
Anal ysi s of the pr obl em showed t he gas charge pr oduced swel l i ng of t he cyl i nder sufficient
to pr oduce a Mode I fat i gue crack t hat coul d move into t he wood. The t i ght fit around t he
cyl i nder al l owed the cr acked wood to ret urn after each shot and a brass ret ai ni ng screw
hel ped to hol d t he wood t i ght l y enough to al l ow each subsequent surge of pressure to creat e
nor mal stresses at t he cr ack tip and move t he crack along.
Si nce si mi l ar desi gns had been used on ot her model s wi t hout pr obl ems, i t was apparent
that further anal ysi s was needed. The wood used was sapwood. Hear t wood is the wood
nearest to the bark and is also accept abl e for use. Sapwood had t r adi t i onal l y been used in
pr i or model s, but needed to be col or ed (it is whi t e in the t ree) and is usual l y l ow in specific
gravity. Physi cal eval uat i on of the wood t ypi cal l y suppl i ed for this pr oduct showed that it
had low specific gravi t y compar ed to t he average for walnut. Al l wood st rengt h propert i es
are rel at ed to wood densi t y to a power n rangi ng from 1.1 to 2.0 [5]. Thi s combi nat i on of
l ower-densi t y wood, sl i ght l y t hi nner cross sect i ons in t he st ock for the new desi gn, and t i ght
fit at the gas cyl i nder l ed to failure. The desi gn was re-eval uat ed and made accept abl e by
usi ng a soft gasket at the front of t he st ock to dampen pressure surges, i ncreasi ng the st ock
t hi ckness by 20%, and usi ng a l arger r ei nf or cement screw. There was no at t empt to get
hi gher-densi t y wood si nce current wood suppl i es from mi dwest pl ant at i ons were all found
to be from smaller, fast er-grown trees that pr oduce l ower - densi t y wood. Thi s t ype of pr obl em
can be expect ed to become mor e common unless desi gners compensat e for decreasi ng den-
si t y in t hei r wood product s.
Case 2
Bowl i ng pi ns t ake a severe beat i ng in servi ce and may fract ure before t hei r desi red servi ce
life. Pi ns are gi ven a l i fe expect ancy in "l i nes. " A l i ne is a compl et e t en-frame game by a
bowler. Two t housand l i nes is desi r ed for a pin. At that t i me, the finish wi l l be cracked and
the wood crushed at cont act areas, but, i f possi bl e, still in one piece. Dur i ng 2000 lines, a
pi n may undergo bet ween 100 000 and 200 000 i mpact s agai nst the floor, wal l s, and ot her
pins.
A pi n is shaped so t hat the mass of t he head and the base resonat e t hrough the neck to
gi ve the fami l i ar and pl easant sounds pr oduced when t he pi ns are st ruck by a ball. The
frequency is in the mi d- 2000 Hz range. A pr opagat i ng cr ack in the neck r egi on can al t er t he
resonance and cause the pi n to sound "dead, " or make a cl anki ng noi se when struck. In t he
ext reme, the head can br eak off. Fi gur es 8 and 9 show a pi n that exhi bi t s smal l cracks in
the body, movi ng up the neck area. A compl et e neck fract ure is shown in Fi g. 9.
An i ncrease in pr emat ur e pi n br eakage occurred and was i nvest i gat ed to det ermi ne why
rat es of br eakage and ret urns i ncreased wi t h no basi c desi gn change. The revi ew i nvol ved a
t horough eval uat i on of the raw mat eri al , har d mapl e (Acer saccharum), and t he sequence of
fabri cat i on processes used to pr oduce t he pin.
Pi ns are made from l ami nat ed boar ds to expose the i mpact surfaces to forces that are
di st ri but ed to a combi nat i on of r adi al and t angent i al faces, t hereby l oweri ng t he opport uni t y
for a deep spl i t to devel op. The l ami nat i ng process uses over 20 pi eces for a pi n and is
i mport ant for its abi l i t y to cont rol pi n wei ght by usi ng dr i l l ed l ayers i nsi de t he pi n (Fig. 8).
The boards are gl ued wi t h a ur ea- f or mal dehyde resin.
The failures were f ound to be heavi l y concent rat ed at gl ue lines, i ndi cat i ng a weaker-t han-
nor mal bond. Invest i gat i on of the manufact uri ng process showed that t he onl y change was

FIG. 8--Bowling pin cross-section showing small cracks along grain.
a new surface planer that machined the faces of the used boards. The new planer used a
large abrasive belt rather than rotating knife blades. A look at the planed surfaces showed
heavy fiber damage and loose debris on surfaces. The problem was that adhesive was being
used to coat a very large and loose surface. The adhesive bond formed was degraded and
the pin was weakened at the glue lines. Figure 10 shows a machine-sanded surface. (Compare
Fig. 10 to Fig. 2, which was prepared by knife cutting the surfaces).

FIG. 9--Fractured bowling pin head showing cracks through wood and on glue line.
Some General Comment s on Wood Fatigue
In loading conditions where tension perpendicular to the grain is not a factor, wood has
extremely good resistance to fatigue failure [6]. In bending, the primary changes in wood
structure after many reverse load cycles is the occurrence of compression wrinkles in the
cell walls. This microbuckling can accumulate and resemble stacked dislocations in metal
crystals. These buckled zones lose their compressive strength and transfer their loads to
adjacent fibers. Even when cumulative damage is large, wood components, such as beams,
only weaken by becoming effectively reduced in the cross-sectional area, where the damaged
fibers are located. The result is a more flexible beam with a reduced static load capacity,
without a propagating crack leading to fracture. A simple experiment to demonstrate this is
to run a long-term flexure test under constant amplitude and monitor the power input to the
test apparatus. As damage occurs, the power to flex the beam drops off. After long test
periods, a microscope can be used to look at the highest stressed wood (the compression
zone is best) to observe fiber damage, such as cell-wall buckling and collapse.
A good experiment for students is to observe the development of damage in a cantilever
beam specimen undergoing flexural cycling. The experimental set-up used here is one where
the beam is loaded by a rotating cam on an electric motor in contact with a roller follower
on the cantilever end. A deflection proportional to 80% of the static modulus of rupture is
the starting point. After the specimen undergoes 10 000 cycle load sequences, a thin section
is removed from the face of the beam at the base attachment. As the beam is cycled, the
student can observe these thin sections in a light microscope and note the progress of a
wrinkled zone from the compression face into the beam cross section. A sketch of the system
is shown in Fig. 11.
Observing macroscopic fatigue failures in wood is very difficult since they are obvious
only under very specific load situations, such as tension perpendicular to grain and large
torsional loads, which are unusual in wood product design. Composites made of wood, such
as strandboard, paper, or plywood are less likely to show fatigue since their highly irregular
microstructures are more able to diffuse cracks [6].

FIG. lO--Sanded (abrasive planed) surface of Maple before laminating into bowling pin (compare to
Fig. 2, also maple).
RIGID CLAMP d- 80% MOR
I" , ~
R
SPECIMEN
~.003" THICK
[RAZOR CUT]
DRIVE CAM
FIG. 11--Cantilever beam tester for laboratory demonstration.

Composites of wood base include paper, whi ch is made of individual wood fibers, wafer
or flake board products, and laminated or veneered products. Observing and verifying fatigue
failures in these materials is even more difficult than in solid wood, because the fiber axis
system is random. Fatigue has been noted in the manufacture and use of paper products
under high tensile stress [5]. Engineers designing with wood chip or fiber composites would
be prudent to develop appropriate tests t o evaluate their designs under simulated use
conditions.
Al t hough the early engineers who said that fatigue is not found in wood structural com-
ponents were wrong, it is easy to see how they were perhaps fooled by the forgiving nature
of wood in use. Finding well documented and verifiable fatigue failures is unusual in wood,
but possible. It is most likely to occur in products where the loading is in directions that
stress the weaker planes. Stresses perpendicular to the grain are a special concern and the
main reason why design handbooks for wood caution against any applications leading to
such conditions.
Conclusions
It might appear from the preceding information that there is little need to study wood
fatigue since it is relatively rare. In fact, the situation is changi ng rapidly in the use of wood
and a few recent developments indicate that fatigue in wood may become a more serious
c onc e r n.
The design of wood structures has traditionally been based on very conservative design
stresses that are assigned by applying large strength-reducing ratios to the measured strength
of clear wood. The resultant low stresses have kept wood members well below load levels
where fatigue damage has any effect. In the past two years, a new test program using actual
breaking loads within specific grades of lumber has resulted in a revision of design stresses.
Some allowable stresses are now higher within a grade [7].
There is considerable effort being made to use wood in the design of small to medium
hi ghway bridges in order to upgrade the hi ghway infrastructure. The designs used will use
wood structural elements at higher stresses than previous designs. Engineers are concerned
about fatigue for these applications [8].
Anot her factor that maintains fatigue as a concern is the gradual reduction in specific
gravity of all common- wood species [9]. The walnut used in the earlier gun stock example,
is not an isolated situation. More wood is plantation grown, heavily fertilized, and intensely
managed to produce larger volumes of timber, with resulting lower density. All wood strength
properties are directly related to density, by a variable power relationship [5]. There are data
[10] that shows that the average density of commercial lumber has been decreasing for at
least the past 50 years. The design values published for timber have been decreasing over
this same period of time. Therefore, the threshold where fatigue is a concern is being lowered
[I01.
When the increase in now used design stresses is combi ned with a raw material of de-
creasing strength qualities, it is clear that fatigue will be an issue that should be considered
and watched during design. It may very well continue to be a secondary concern in most
wood designs, but certainly not one to be routinely neglected.
References
[1] Barlow, T. M., "The Weight of Aircraft," Aircraft Engineering, Vol. 1, 1929, p. 29.
[2] Lewis, W. C., "Design Considerations for Fatigue in Timber Structures," Journal of Structural
Division, May 1960, pp. 15-23.

KYANKA ON THE USE OF WOOD AND A WOOD COMPOSITE 2 1 9
[3] "Duration of Load and Fatigue in Wood Structures," Proceedings, Paper 1361, ST-5, Vol. 83,
Subcommittee on Timber Structures, American Society of Chemical Engineers (ASCE), Sept. 1957,
pp. 1-12.
[4] Lewis, W. C., "Fatigue of Wood and Glued Joints Used in Laminated Construction," Proceedings
of the Forest Products Resource Society, Vol. 5, Forest Products Resource Society, 1951, pp. 221-
229.
[5] Bodig, J. and Jayne, B. A., Mechanics of Wood and Wood Composites, Van Nostrand-Reinhold,
New York, 1982, pp. 307-312.
[6] Kyanka, G. H., "Fatigue Properties of Wood and Wood Composites," International Journal of
Fracture, Vol. 16, No. 6, Dec. 1980, pp. 609-616.
[7] National Design Specification for Wood Construction, National Forest Products Association, Wash-
ington, DC, 1991.
[8] Ritter, M. A., Timber Bridges, Publication EM 7700-8, U.S. Department of Agriculture, Washington,
DC, 1990.
[9] Green, D. W. and Kretschman, D. E., "'Stress Class Systems," USDA Forest Products Laboratory
Research Paper, FPL-RP-500, 1990.
[10] Ethington, R. L., Galligan, W. L., Montrey, H. M., and Freas, A. D. , "Evolution of Allowable
Stresses in Shear for Lumber," General Technical Report FPL 23, U.S. Department of Agriculture,
Madison, WI, 1979.

I
-
-
i

v
J

;
z

l
m

!

r
m

L
L
I

b
-
J

!

,
,
.
o

J
]

!

~
r

C
o
p
y
r
i
g
h
t

b
y

A
S
T
M

I
n
t
'
l

(
a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
)
;

T
h
u

M
a
y

8

1
0
:
4
2
:
1
2

E
D
T

2
0
1
4
D
o
w
n
l
o
a
d
e
d
/
p
r
i
n
t
e
d

b
y
U
n
i
v
e
r
s
i
d
a
d

D
e

C
h
i
l
e

p
u
r
s
u
a
n
t

t
o

L
i
c
e
n
s
e

A
g
r
e
e
m
e
n
t
.

N
o

f
u
r
t
h
e
r

r
e
p
r
o
d
u
c
t
i
o
n
s

a
u
t
h
o
r
i
z
e
d
.


Downloaded/printed by Universidad de Chile Pursuant To License Agreement. No Further Reproductions Authorized

Загружено:

Сведения о документе

Исходное описание:

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Downloaded/printed by Universidad de Chile Pursuant To License Agreement. No Further Reproductions Authorized

Загружено:

Авторское право:

Доступные форматы

Copyright by ASTM Int'l (all rights reserved); Thu May 8 10:42:12 EDT 2014

Вам также может понравиться