100% found this document useful (1 vote)

156 views270 pages

STP 612-1976

The document is the introduction to a symposium on thermal fatigue of materials and components. It discusses that thermal fatigue problems involve both thermal and mechanical loadings on materials. It notes that recent advances in stress analysis methods and testing equipment have enabled more detailed study of thermal fatigue beyond simpler isothermal testing. The symposium focuses on stress analysis, life prediction techniques, studying material behavior under thermal fatigue, and thermal-mechanical fatigue testing systems. It aims to review the current state of understanding of thermal fatigue.

Uploaded by

emre

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

100% found this document useful (1 vote)

156 views270 pages

STP 612-1976

Uploaded by

emre

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 270

THERMAL FATIGUE OF

MATERIALS AND
COMPONENTS

A symposium
sponsored by ASTM
Committee E-9 on Thermal Fatigue of
Materials and Components
AMERICAN SOCIETY FOR
TESTING AND MATERIALS
Philadelphia, Pa., 17-18 Nov. 1975

ASTM SPECIAL TECHNICAL PUBLICATION 612

D. A. Spera, NASA Lewis Research Center, and
D. F. Mowbray, General Electric Co., editors

List price $27.00

04-612000-30

jAMERICAN SOCIETY FOR TESTING AND MATERIALS

'l916 Race Street, Philadelphia, Pa. 19103
•

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27 13:13:08 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
© by AMERICAN SOCIETY FOR TESTING AND MATERIALS 1976
Library of Congress Catalog Card Number: 76-21535

NOTE
The Society is not responsible, as a body,
for the statements and opinions
advanced in this publication.

Printed in Baltimore, Md.

Nov. 1976

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27 13:13:08 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
Foreword
The Symposium on Thermal Fatigue of Materials and Components was
presented at the meeting held in New Orleans, 17-18 Nov. 1975. The
symposium was sponsored by The American Society for Testing and
Materials through its Committee E-9 on Fatigue. D. A. Spera, National
Aeronautics and Space Administration Lewis Research Center, and D. F.
Mowbray, General Electric Co., presided as symposium co-chairmen.

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27 13:13:08 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
Related
ASTM Publications
Fatigue Crack Growth Under Spectrum Loads, STP 595 (1976), $34.50
(04-595000-30)
Manual on Statistical Planning and Analysis for Fatigue Experiments,
STP 588 (1976), $15.00 (04-588000-30)
Handbook on Fatigue Testing, STP 566 (1974), $17.25 (04-566000-30)

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27 13:13:08 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
A Note of Appreciation
to Reviewers
This publication is made possible by the authors and, also, the un-
heralded efforts of the reviewers. This body of technical experts whose
dedication, sacrifice of time and effort, and collective wisdom in review-
ing the papers must be acknowledged. The quality level of ASTM publica-
tions is a direct function of their respected opinions. On behalf of ASTM
we acknowledge with appreciation their contribution.

ASTM Committee on Publications

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27 13:13:08 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
Editorial Staff
Jane B. Wheeler, Managing Editor
Helen M. Hoersch, Associate Editor
Ellen J. McGlinchey, Assistant Editor
Kathleen P. Turner, Assistant Editor
Sheila G. Pulver, Editorial Assistant

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27 13:13:08 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
Contents
Introduction
What is Thermal Fatigue?— D. A. SPERA 3
Nonlinear Analysis of a Tapered Disk Thermal Fatigue Specimen—
D. F. MOWBRAY AND J. E. MCCONNELEE 10

Thermal Stress Concentration Factors in Large Shafts—D. C. GONYEA 30

Low-Cycle Fatigue Analysis of the Turbine Disk for the National
Aeronautics and Space Administration High-Temperature
Turbine Rig—s. TEPPER 38
Thermal Fatigue and Its Failure Prediction for Brittle Ceramics—
D. P. H. HASSELMAN, R. BADALIANCE, AND E. P. CHEN 55
Description of a Computerized Method for Predicting Thermal
Fatigue Life of Metals—D. A. SPERA AND E. C. COX 69
A Study of Thermal Fatigue Mechanisms—M. A. H. HOWES 86
Thermal-Stress Fatigue Behavior of Twenty-Six Superalloys—
p. T. BIZON AND D. A. SPERA 106
Effect of Microstructure on the Thermal Fatigue Resistance of a
Cast Cobalt-Base Alloy, Mar-M509—c. c. BECK AND
A. T. S A N T H A N A M 123

Some Thermal Fatigue Characteristics of Mild Steel for Heat

Exchangers—j. s. LAUB 141
Low-Cycle Thermal Mechanical Fatigue Testing—s. w. HOPKINS 157
Thermal-Mechanical, Low-Cycle Fatigue of AISI 1010 Steel—c. E. JASKE 170
Thermomechanical Fatigue Crack Propagation in an Anisotropic
(Directionally Solidified) Nickel-Base Superalloy—A. E. GEMMA,
B. S. LANGER, AND G. R. LEVERANT 199
Vacuum Thermal-Mechanical Fatigue Behavior of Two Iron-Base
Alloys—K. D. SHEFFLER 214
Instability Effects in Thermal Fatigue—L. F. COFFIN 227
Life Prediction of Thermal-Mechanical Fatigue Using Strainrange
Partitioning—G. R. HALFORD AND S. S. MANSON 239

Summary 255
Index 260

Introduction

Most low-cycle fatigue problems in high-temperature machinery in-

volve thermal as well as mechanical loadings. By thermal loadings it is
meant that the material is subjected to cyclic temperature simultaneous
with the cyclic stress. Analysis of these loadings and consideration of
the attendant fatigue damage becomes very complex and often gross
simplifications are introduced. Usually this involves the use of isothermal
data and life prediction techniques evolved from isothermal testing. In
fact, fatigue studies in the laboratory have generally bypassed real thermal
fatigue loadings in favor of easier isothermal testing.
It would appear that recent rapid advances in analysis methods (for
example, finite element computer programs) and testing equipment (for
example, servohydrauhc test systems) obviate some of the reasons for
resorting to isothermal fatigue in the analysis of thermal fatigue prob-
lems. Indeed, there is now increased emphasis on studying thermal
fatigue. In light of these developments, a symposium on the topic of
thermal fatigue seemed both appropriate and timely. ASTM Committee
E-9 on Fatigue through its Subcommittee E09.08 on Fatigue Under
Cyclic Strain undertook the symposium organization in hope that it would
serve as a review of the current state of the art in thermal fatigue.
The symposium, as it has developed, focuses on four important aspects
of thermal fatigue: (a) stress and deformation analysis; (b) Ufe prediction
techniques; (c) materials behavior; and (d) thermal-mechanical testing.
The papers on stress analysis span the domain of complexity from strictly
elastic to elastic-plastic-creep analysis. Life prediction techniques include
applications of the creep damage and strain range partitioning approaches
to thermal fatigue analysis. Papers on material behavior studies span the
range of materials from the most common structural material, low-carbon
steel, to advanced superalloys and ceramics. A number of papers deal
with the subject of thermal-mechanical fatigue test systems. These sys-
tems allow one to study thermal fatigue with control not possible just a
few years ago. As a result, we can expect to see considerable emphasis
on this topic in the future. The present papers describe the equipment
and testing procedures developed and demonstrate valuable data. In addi-
tion to the technical papers, the first paper in the volume (the opening
address) gives a very interesting historical sketch of the development of
thermal fatigue testing and analysis.
The results of the symposium as presented in this volume should be

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
2 THERMAL FATIGUE OF MATERIALS AND COMPONENTS

of use to several groups within the scientific community, most notably

to the mechanical design engineer contending with the analysis of thermal
fatigue problems, the materials engineer faced with selecting the proper
materials to resist thermal fatigue loadings, and the laboratory test en-
gineer who is establishing facilities for conducting thermal fatigue tests.
It is our belief that the results presented in this volume will also serve as
a stimulus for future meetings on the topic of thermal fatigue.

D. F. Mowbray
Mechanics of Materials Unit, Materials and
Processes Laboratory, General Electric
Company, Schenectady, N.Y.; symposium
cochairman and also coeditor of this
publication.

What is Thermal Fatigue?

REFERENCE: Spera, D. A., "What is Thermal Fatigue?" Thermal Fatigue of

Materials and Components, ASTM STP 612, D. A. Spera and D. F. Mowbray, Eds.,
American Society for Testing and Materials, 1976, pp. 3-9.

ABSTRACT: Definitions are suggested for terms such as thermal fatigue, thermal-
mechanical fatigue, and thermal-stress fatigue. Historical developments in the field of
thermal fatigue from 1838 to the present are reviewed. It is hypothesized that
we have advanced from qualitative to quantitative understanding of the thermal
fatigue process and can now make life predictions with factor-of-two reliability
for conventional metals.

KEY WORDS: thermal fatigue, fatigue (materials), definitions, fatigue life

The purpose of the symposium which produced this volume was to ex-
change ideas on methods for determining and improving the resistance of
materials and mechanical components to that combination of failure
mechanisms which we know as thermal fatigue. To various people at
various times, thermal fatigue is also known as thermal stress fatigue,
thermal strain fatigue, creep fatigue, thermal cracking, thermal shock,
thermal rupture, thermal endurance, low-cycle thermal fatigue, heat
checking, craze cracking, fire cracking, and just plain high-temperature
fatigue. Considering this assortment of names, it is appropriate for us to
discuss the terminology we will need in order to communicate effectively
with one another on this complicated subject.
First, a few simple definitions will be suggested which may be useful,
particularly to those who have had Uttle contact with the study of thermal
fatigue. If these definitions help us, we may wish to continue using them.
If not, let us try to improve them. I believe they are relevant to present
studies of thermal fatigue phenomena. And what is just as important,
they contain a certain measure of respect for the historical development
of our field. After giving the definitions, I will briefly review the de-
velopment of terms during the past 137 years of our technical history.

' Consultant, Fatigue Branch, Lewis Research Center, Cleveland, Ohio 44135.

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
THERMAL FATIGUE OF MATERIALS AND COMPONENTS

Suggested Definitions
First, let me suggest this definition for thermal fatigue: "Thermal
fatigue is the gradual deterioration and eventual cracking of a material
by alternate heating and cooling during which free thermal expansion is
partially or completely constrained." Constraint of thermal expansion
causes thermal stresses which may eventually initiate and propagate
fatigue cracks. Thermal fatigue may be classified under the more general
heading of low-cycle fatigue, because thermal fatigue cracks usually start
in less than 50 000 cycles. In addition, a thermal fatigue cycle usually con-
tains significant inelastic strain. Thus, I suggest we start defining our
terms by dividing the subject of low-cycle fatigue into two branches, de-
fining one as "thermal fatigue" if the temperature is not constant with
time, and the other as "isothermal fatigue" if it is. As an example,
thermal fatigue might result from the starting and stopping of a piece of
high-temperature equipment, while isothermal fatigue might be the con-
sequence of vibration during steady-state operation.
Constraint of free thermal expansion and contraction is a necessary
ingredient in the thermal fatigue process. For convenience, constraints
may be grouped into two general categories: external and internal. Ex-
ternal constraints are those provided by boundary forces applied to the
surfaces of the body which is being heated and cooled. This type of con-
straint is more typical of specimen testing than it is of components in
service. Designers of high-temperature equipment usually take consider-
able care to provide for overall thermal expansion and contraction
through the use of clearances, sliding supports, bellows, and other
mechanical devices. Thus, thermal fatigue with external constraint is
primarily a laboratory testing practice in which external forces on a test
specimen are used to simulate internal thermal stresses in an actual com-
ponent.
In a typical thermal fatigue test with external constraint, a uniaxial
specimen—usually tubular or hourglass in shape—is heated and cooled
uniformly across its test section while constraining forces are applied
through end grips. This test technique is referred to as "thermal-mechani-
cal fatigue testing." End constraints may be as simple as a semirigid
frame which only allows the specimen to lengthen and shorten in ac-
cordance with the elastic compliance of the frame. This simple device is
often called a Coffin-type thermal fatigue apparatus, after its developer.
Dr. L. F. Coffin, Jr., [1].^ Today, the simple constraining frame has
given way to the modern servocontroUed fatigue machine in which
temperature and strain cycles may be applied independently according to
predetermined programs. Whether the apparatus is simple or complex,
thermal-mechanical fatigue is always characterized by external constraint
' The italic numbers in brackets refer to the list of references appended to this paper.

forces. Therefore, let me suggest this second definition: "Thermal-

mechanical fatigue is thermal fatigue in which constraints on thermal
expansion are external."
The second general category of thermal fatigue failures are those with
internal constraints. Internal constraints on the expansion and contraction
of an element of material are provided by adjacent material elements
which are either at a different temperature or are made of a different
material (for example, fiber-reinforced composites), or both. Internal
constraints are typical of high-temperature machine components which
are subjected to rapid heating and cooling, usually by some type of work-
ing fluid. Large thermal gradients may be produced in these components
which will, in turn, produce self-equilibrating thermal stresses. In the
laboratory, relatively high rates of surface heating and coohng are
usually necessary to reproduce these internal constraints. Specimens are
often shaped to simulate a significant feature of the component being
studied and to accentuate thermal gradents. For example, thermal fatigue
cracking of the leading edge of a gas-turbine blade may be simulated in
the laboratory by a wedge specimen flame-heated along its apex edge.
Often, the component is tested directly, since it provides its own in-
ternal constraints. Because self-equiUbrating thermal stresses are
characteristic of thermal fatigue with internal constraints, we may add
this third definition: "Thermal-stress fatigue is thermal fatigue in which
constraints on thermal expansion are internal."
Figure 1 is a summary of the suggested terminology. We may divide the
general field of low-cycle fatigue into thermal and isothermal categories,
depending on whether temperature is cyclic or constant, respectively. The

LOW-CYCLE FATIGUE:
LIFE LESS THAN 50 000 CYCLES
SIGNinCANT INELASTIC STRAIN

THERMAL FATIGUE: ISOTHERMAL FATIGUE:

CYCLIC TEMP CONSTANT TEMP

THERMAL-MECHANICAL THERMAL-STRESS
FATIGUE: FATIGUE:
EXTERNAL CONSTRAINT INTERNAL CONSTRAINT

FIG. 1—Proposed terminology for high temperature fatigue.

thermal fatigue category may then be subdivided into thermal-mechanical

fatigue and thermal-stress fatigue, depending on whether constraints are
external or internal, respectively.

Historical Review
Having suggested some definitions, let us briefly review the develop-
ment of the field of thermal fatigue and the origin of some of the terms
just defined. Much of the information which follows was taken from a
1956 literature survey prepared by Harry Majors, Jr., of the University of
Alabama 12]. The study of material failure as a result of thermal cycling
goes back to 1838, at least. In that year, Duhamel published the first
equations for calculating the thermal stresses in a body subjected to non-
uniform heating [3]. However, it was not until 1894 that fracture was
combined with thermal stress in a quantitative way. This was done in
Germany by Winkelmann and Schott who studied what we now call
thermal shock of ceramic materials, using only rapid heating [4]. Eight
years later, Hovestadt and Everhart of the United States published an
analysis for thermal shock of glass through rapid cooling [5]. And that
was the trend of the thermal stress failure literature for the next 30 or
more years: thermal shock of materials by either rapid heating alone or
rapid cooling alone. Most—but not all—of the materials studied were
ceramics and glass. The earliest reported study of the thermal shock be-
havior of ductile metals was probably that by BoUenrath and co-workers
in 1938, again in Germany [6].
It was not until 1935, almost a century after Duhamel's publication on
thermal stress, that alternate heating and cooling were discussed in the
literature as a cause of cyclic thermal strains. This discussion appeared in
a text on plasticity in crystals, written by Erich Schmid and Walter
Boas [7]. These authors noted that, when a noncubic crystal is alternately
heated and cooled, it gradually increases in size, provided no phase
changes take place. Theoretically, a high-temperature component made of
such an anisotropic material would produce within itself ever-increasing
thermal stresses and eventual cracking. And this cracking would occur
even if the component were heated and cooled slowly, without tempera-
ture gradients and without external constraints. Nine years later. Boas
and a co-worker by the name of R. W. Honeycombe showed that this
was indeed the case for bearings made from a particular tin alloy. Their
paper, published in 1944, was given the rather liberal title of "Thermal
Fatigue of Metals" [8]. This was probably the first time this phrase ap-
peared in the literature.
As we have seen, then, the original definition of thermal fatigue,
credited to Walter Boas in 1944, was very restrictive. It applied only to
failure in a noncubic lattice material slowly heated and cooled without

any external constraint. During the next decade, this Hmited definition
would be gradually expanded because engineers had a need for a term to
describe high-temperature material failures that were cyclic in nature.
We can guess that it was difficult at first to combine two concepts that
had developed separately: the concept of thermal stress and the concept
of fatigue failure. As an example, let me quote from an old report from
the archives of my laboratory at the Lewis Research Center of the Na-
tional Aeronautics and Space Administration. At the time this report was
written in 1947, our laboratory was called the Aircraft Engine Research
Laboratory of the National Advisory Committee for Aeronautics
(NACA). The title of this report is "Investigation of Rim Cracking in
Turbine Wheels with Welded Blades," and the principal author was
Morton Millenson. To explain the cause of cracking in certain gas turbine
disks, Millenson pointed to cyclic plastic strains, as follows: "Subsequent
cycles of starting and stopping will cause alternate compressive and tensile
flow and progressively weaken the material until cracking occurs in a
manner somewhat similar to the fatigue of metals " I can guess that
Millenson's coauthor would have used a more definite phrase than
"somewhat similar to the fatigue of metals" had it been solely up to him.
The coauthor of this 1947 report [9] was a young man by the name of
S. S. Manson. It would not be long before Manson conceived and di-
rected fatigue research programs at our laboratory and established an
international reputation in the field of high-temperature material be-
havior.
The terminology was changing and developing, however, as it always
does to fill needs. The earUest mention of thermal fatigue in our files at
the Lewis Research Center was in 1949, when John Weeton, a colleague
of Millenson and Manson, used the term to describe one of the failure
mechanisms he observed in combuster liners for gas turbines [10]. In
1950, the term thermal cracking was used by Harry Wetenkamp and his
co-workers at the University of Illinois in their study of the failure of
railway car wheels caused by the heating action of brakes [77]. In 1952,
Helmut Thielsch wrote a survey article for the Welding Research Council
on the subjects of thermal fatigue and thermal shock [72]. Later that
same year, in a lecture deUvered at the University of Michigan, Manson
spoke on the behavior of materials under conditions of thermal stress
[75]. He stated that thermal shock is the failure of a material after a
single cycle of thermal stress while thermal fatigue requires repeated
thermal stress cycles.
Today, that lecture at Michigan is remembered, not for the thermal
stress information it contained, but for Manson's proposal that low-cycle
fatigue Ufe depends primarily on the imposed range of inelastic strain. He
suggested a power law which is known throughout the world today as the
Manson-Coffin equation.

I mentioned Coffin earlier as the inventor of an apparatus for thermal-

mechanical fatigue testing. Using this apparatus to conduct a broad series
of fatigue tests, he also obtained a power-law relationship between in-
elastic strain range and cycles to failure, and he did so simultaneously
with and independently of Manson's work. His two landmark papers on
the effects of thermal stress on ductile metals were presented in 1953
[J, 14]. They have served as basic references for just about every paper
on thermal fatigue written in the last 20 years. I decided to reread these
two pioneering papers to see how Coffin defined and used the term
thermal fatigue. To my surprise, in an estimated 25 000 words of text, the
phrase "thermal fatigue" never appears once! I asked Coffin about this,
and he was just as surprised as I was, not at the absence of any reference
to thermal fatigue, however, but at the fact that he had written 25 000
words on the subject!
Whatever his reasons for not mentioning thermal fatigue in 1953, Cof-
fin introduced the term "thermal stress fatigue" in 1954. The occasion
was an American Society for Testing and Materials' symposium on the
effect of cyclic heating and stressing on metals, which was the predecessor
of the symposium which produced this volume. In Coffin's paper [15],
the two terms "thermal stress" and "fatigue" were finally wedded, 116
years after Duhamel's original publication.
To summarize the next ten years, let me quote from an excellent British
text on thermal stress, edited in 1964 by P. P. Benham and Russell
Hoyle of the Imperial College of Science and Technology. There is a
chapter in this book on the subjects of thermal fatigue and thermal
shock, written by H. G. Baron of the Royal Armament Research and
Development Establishment [16], Baron ends this chapter with the follow-
ing observation: "The subject of thermal fatigue has advanced in the last
decade (that is, from 1954 to 1964) from a position of almost total ob-
scurity to one in which many of the controlling factors are qualitatively
understood, and in which order-of-magnitude estimates can sometimes be
made of the number of thermal cycles which will produce cracks."
This, then, was the situation as of 1964, as seen by a specialist in our
field: qualitative understanding, sometimes leading to order-of-magnitude
life estimates.

Conclusions
Now another decade has passed, the second since Manson and Cof-
fin introduced life analysis to thermal fatigue. To conclude, let me offer
the following hypothesis. In this second decade, we have advanced from
qualitative to quantitative understanding, and from order-of-magnitude
life estimates to factor-of-two life predictions, at least for conventional

metals. This is an optimistic—but I think realistic—hypothesis which re-

quires further evaluation.

References
[/] Coffin, L. F., Jr. and Wesley, R. P., Transactions, American Society of Mechanical
Engineers, Vol. 76, Aug. 1954, pp. 923-930.
[2] Majors, H., Jr., "Thermal Shock and Fatigue—A Literature Survey," Technical
Report No. 1, Office of Ordnance Research Project No. 1230, Sept. 1956.
[3] Duhamel, J. M. C. in Memoires de I'Institut deFrance, Vol. 5, 1838, p. 440.
[4] Winklemann, A. and Schott, O., Annalen der Physikalischen Chemie, Vol. 51, 1894,
p. 730.
[J] Hovestadt, H. and Everhart, J. D. in Jena Glass, Macmillan, New York, 1902, p. 228.
[6\ BoUenrath, F., Cornelius, H., and Bungardt, W., Jahrbuch der Deutschen Luft-
fahrtforschung, 1938, Vol. 11, pp. 326-338.
[7] Schmid, E. and Boas, W. in Kristallplastizitaet, Verlag Julius Springer, Berlin, 1935,
p. 202.
[S] Boas, W. and Honeycombe, R. W. K., Nature, Vol. 153, April 1944, pp. 494-495.
[P] Millenson, M. B. and Manson, S. S., "Investigation of Rim Cracking in Turbine
Wheels with Welded Blades," NACA RM E6L 17, National Advisory Committee for
Aeronautics, Feb. 1947.
[10] Weeton, J. W., "Mechanisms of Failure of High Nickel-Alloy Turbojet Combustion
Liners," NACA TN 1938, National Advisory Committee for /\eronautics, Oct. 1949.
[77] Wetenkamp, H. R., Sidebottom, O. M., and Schrader, H. J., "The Effect of Brake
Shoe Action on Thermal Cracking and on Failure of Wrought Steel Railway Car
Wheels," Bulletin 387, Engineering Experiment Station, University of lUinois, Urbana,
Vol. 47, No. 77, June 1950.
[72] Thielsch, H., "Thermal Fatigue and Thermal Shock," Welding Research Council
Bulletin, Series No. 10, April 1952.
[13] Manson, S. S., "Behavior of Materials Under Conditions of Thermal Stress," NACA
TN-2933, National Advisory Committee for Aeronautics, July 1953.
[14] Coffin, L. F., Jr., Transactions, American Society of Mechanical Engineers, Vol. 76,
Aug. 1954, pp. 931-950.
[75] Coffin, L. F., Jr. in Symposium on Effect of Cyclic Heating and Stressing on Metals
at Elevated Temperatures, ASTM STP 165, American Society for Testing Materials,
June 1954, pp. 31-52.
[16] Baron, H. G. in Thermal Stress, P. P. Benham and R. Hoyle, Eds., Pitman and
Sons, London, 1964, p. 204.

Nonlinear Analysis of a Tapered Disk

Thermal Fatigue Specimen

REFERENCE: Mowbray, D. F. and McConneiee, J. E., "Nonlinear Analysis of a

Tapered Disk Thermal Fatigue Specimen," Thermal Fatigue of Materials and Com-
ponents, ASTM STP 612, D. A. Spera and D. F. Mowbray, Eds., American Society
for Testing and Materials, 1976, pp. 10-29.

ABSTRACT: Calculations of the stress-strain response of a common tapered disk

thermal fatigue specimen are presented. The calculations were made by complete non-
linear finite element solution and by a simpler method employing the elastic strain-
invariance principle. The results are used to show the effect of several common test
variables on the resulting stress-strain response, to assess the effect of various material
behavior and structural analysis assumptions on the stress-strain response, and to ex-
plore the application of several high-temperature fatigue damage prediction methods
to transient thermal fatigue loading.

KEY WORDS: thermal fatigue, fatigue (materials), disks, thermal analysis, stress
analysis

A common testing approach for generating thermal fatigue data under

transient thermal conditions incorporates the combination of tapered disk
specimens and fluidized baths [1,2]} The specimens are subjected to al-
ternate heating and cooling shocks by immersion in the fluidized baths at
differing temperatures. Results from this type of test have close applica-
bility to turbine thermal fatigue problems because the strain-time-temper-
ature history produced in the specimens is similar to those existing in tur-
bine parts such as blades and stator vanes. Unfortunately, direct measure-
ment of stress or strain or both are not possible so that results obtained
from such tests cannot be expressed in a quantitative manner. Quantifica-
tion requires highly complex analysis when temperature and loading con-
ditions induce nonlinear deformations. The analysis must consider the non-
uniform, three-dimensional geometry in the thermal and stress analysis,
' Manager, Mechanics of Materials Unit, and Manager, Component Analysis and Test
Unit, respectively. Materials and Processes Laboratory, General Electric Co., Schenectady,
N. Y. 12345.
' The italic numbers in brackets refer to the list of references appended to this paper.

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
MOWBRAY AND MC CONNELEE ON TAPERED DISK SPECIMEN 11

factor in material property variations with temperature, and account for

the accumulation of time independent and time-dependent nonlinear de-
formations.
Considerable work has been completed in the past on the analysis of
the tapered disk test [3-5]. However, many approximations were used in
the structural analysis method and material behavior representation. The
present paper offers some expansion on and refinements of these past
efforts. The finite element (FE) method was employed to improve the ac-
curacy of heat transfer and stress analyses. Sensitivity studies were con-
ducted to determine the effect of several different material behavior as-
sumptions on the results.
Two analysis methods were utilized. They differ in the degree of ap-
proximation made and information obtjiined. One was based on an ap-
proach utilizing linear axisymmetric finite element stress analysis to deter-
mine the total strain history of the peripheral fiber. The nonUnear strain
components and stress of this fiber were found by treating it as a uniaxial
bar subjected to the calculated strain history. The second method involved
use of an axisymmetric FE computer program with capability for treating
nonlinear, time-dependent material behavior. This method generated a
complete stress-strain solution. However, it was very costly in terms of
computation time so that the former, simpler approach was utilized in the
bulk of the analysis. In the subsequent discussion, the simpler approach
will be referred to as the uniaxial, elastic strain invariance (UESI) method
and the complete finite element solution as the axisymmetric, FE method.
The results of the analyses are used to
(a) show the effects of several common test variables on the resulting
stress-strain response,
ib) assess the effect of various material behavior and structural analysis
assumptions on the resulting stress-strain response, and
(c) explore the application of several high temperature fatigue damage
prediction methods to transient thermal fatigue loading.

Tapered Disk Specimen and Test

The tapered disk specimen is shown in Fig. 1. The overall diameter and
center thickness are 1.5 times the original dimensions employed by Glenny
et al [1]. The fourfold variation in peripheral radius {Rp) was incorpo-
rated in the test as a means of achieving a variation in cyclic strain range
for fixed minimum and maximum cycle temperatures.
In the testing, the disk specimens are subjected to edternate heating and
cooling shocks by transferring them between hot and cold fluidized baths
of aluminum oxide sand. In the tests analyzed herein, the minimum tem-
perature (rmin) was always held at 70°F for an exposure time {tc) of 4 min.

PART Rp,t.OOI 2R,i.OI

1 .010 2.38
2 ,020 2.40
3 .030 2.42
4 .040 2.44

FIG. 1—Tapered disk thermal fatigue specimen.

The hot bath or maximum temperature (Jmax) was varied between 1500
and 1900°F, with exposure times (4) of 1 or 4 min.
The test material analyzed is the cast cobalt-base superalloy, FSX414.
This alloy has been described in Ref 6. Table 1 gives a summary of static
properties over the temperature range of interest.

TABLE 1 —Mechanical properties for FSX-414.

Tensile Rupture

Tensile 0.2% Yield

Temperature Strength, Strength, Reduction Reduction
°F ksi ksi of Area, % o, ksi r«,h of Area, %

80 106 60 18
800 86 41 18
1000 79 36 19
1200 74 33 21
1400 65 31 23 21.0 2150 39
1600 45 30 28 10.0 1150 12
1800 4.5 1870 4

Analysis Methods
Both analysis methods started with the transient temperature fields de-
termined by FE computation. Temperature-dependent values of thermal

conductivity and specific heat were utilized, and the heating and cooling
shocks were simulated by applying time-independent values of convective
heat transfer coefficient on all exterior surfaces. The brief transfer time
from the hot to cold bath was included by specifying a convective heat
transfer coefficient corresponding to still air during the first 3.5 s of the
cooling shock. The thermal properties employed are given in Table 2.
(The heat transfer coefficients are those measured in the baths.) Axisym-
TABLE 2—Thermal and physical properties for FSX414.

Thermal Coefficient Elastic

Temperature, Conductivity, of Expansion Modulus,
°F Btu/ftVin./h/°F xlO' 10' psi

200 85 7.2 27.4

400 101 7.55 26.4
600 116 7.9 25.3
800 129 8.25 24.3
1000 142 8.55 23.1
1200 154 8.75 22.0
1400 165 9.05 20.7
1600 176 9.3 19.4
1800 187 9.5 17.8
Convective heat transfer coefficients, Btu/°F in.^ h
1. Hot bath, <1688''F 0.80
2. Hot bath, >1688°F 1.04
3. Cold bath 0.80
4. Still air 0.08
Specific heat, Btu/lb-mass°F 0.12
Density Ib-mass/in.' 0.312

metry in the temperature and stress fields was assumed. An example FE

grid is shown in Fig. 2. The elements employed are of the constant stress
type.
Comparisons of calculated time-temperature response and thermocou-
ple measurements at several locations within a disk are shown in Fig. 3. In
general, these comparisons show a good correspondence between calcula-
tions and measurement.
In the initial seconds of each shock, the maximum temperature change
develops at the disk periphery so that the maximum strains also develop
at this location. The difference between the peripheral (T^) and volume
average temperature (Jkyg) is proportional to the peripheral strain. Ex-
ample plots of this difference as a function of time are shown in Fig. 4.
Maxima are reached in 5 to 15 s and complete dissipation of the gradient
in 1 to 2 min.
The heat transfer and stress analysis programs used are all compatible
so that the same finite element models were used in all analyses, with one

/ <
/

/ C B

0 ^ - - ^ ^ ^ ^ ^ —-• ~" "^ H "—

r A

FIG. 2—Example finite element grid (Rp = 0.04 in.).

ISOOr- HEATING
1600- '"^^

"0 10 20 30 40 50 60 70 8090^120 160 200240

IMMERSION TIME, SEC

FIG. 3—Comparison of measured and calculated time-temperature response of a disk

specimen.

Te-Tovq.^F

FIG. 4—Calculated lime-temperature curves for disk specimens.

exception. The exception was involved with the nonUnear FE analysis. In

order to reduce cost for this case, a simpler model was employed after
preliminary compeirisons with results from the full model demonstrated
the adequacy of the reduced model.
Before proceeding with specific descriptions of the two analysis methods,
two assumptions about material behavior required in both analysis methods
will be discussed. One is concerned with the Bauschinger effect and the
second with a method for stress transfer in creep analysis. Figure 5 illus-
trates schematically the nature of the assumptions involved.
With regard to the Bauschinger effect, the stress level at which reversed
plastic flow commences must be specified if strain hardening is included
in the analysis. Three common assumptions are illustrated by Points a, b,
and c in Fig. 5a [7]. Point a corresponds to a kinematic theory, Point c to
the isotropic theory, and Point b implies yielding in tension and compres-
sion to be independent of each other.
The kinematic theory allows for the characteristic in cyclic plasticity of
a translating yield surface, whereas the other two theories do not. Despite
this feature, a recent investigation [8] has shown the isotropic theory to be
better able to predict subsequent yielding behavior. Observation of the
data examined in Ref 8 indicates that the independent tension and yield
theory will provide better predictions than the kinematic or isotropic theo-
ries.
All three assumptions were explored in the current calculations with the
UESI method. Differences in results were not large so that, for computing
convenience, the bulk of the computations were carried out using the in-
dependent tension and compression yield assumption (Point b in Fig. 5a).

FIG. 5—Schematic illustrations of material behavior considerations in nonlinear, cyclic

temperature analysis: (a)Bauschinger effect and (b) stress transfer in creep analysis.

The FE analysis method employed the isotropic hardening assumption

only.
In both analysis methods, creep strain was specified in the form of the
hyperbolic sine law

£c = 2At^^^ sinh Go (1)

where A and G are functions of temperature, evaluated by means of non-

linear regression analysis of available creep data. For the FSX414 alloy,
the following specific functional forms were found to provide a good fit
to available data

A = P - Q/(T + 460)
(2)
G = l/(U - VT)

where P, Q, U, and Fare regression parameters.

The creep relationship being nonlinear requires that one specify a rule
for determining creep rate after a change of stress level occurs. Two com-
monly employed assumptions in the treatment of this problem are called
the strain hardening and time hardening rules. These are indicated in Fig.
5b by Lines oa and ob, respectively. The strain hardening rule is generally
thought to be the most accurate and the most physically acceptable [9]. It
was therefore used in the present analyses. Time hardening was utilized in
selected examples, and the difference in results will be shown.

UESI Method
In this method, the total strain-time-temperature histories were ob-

tained by elastic analysis using the elastic strain-invariance principle [10].

The said principle assumes that strains computed by elastic analysis are the
correct total strains, even though the stresses are affected by plastic flow.
This principle is known to yield accurate predictions of total strain for
purely thermal axisymmetric loading such as in the disk test.
Total strain solutions were obtained by elastic FE analysis at various
times in the temperature transients. The temperature-dependent values of
elastic modulus and coefficient of thermal expansion used in the analyses
are listed in Table 2.
The nonlinear part of the analysis treated the peripheral fiber as a uni-
axial bar subject to prescribed strain-time-temperature histories. It is
noted that similar reasoning is employed with success in fatigue notch
analysis [77]. A typical example of the cycle experienced by the disk pe-
riphery is shown in Fig. 6 where strain is plotted against temperature,
with time indicated by crosshatching.

Rp=O.OIin.
Tmo»=l688''F

STRAIN,
% 0

CROSS-HATCHING INDICATES TIME IN SECONDS

FIG. 6—Example strain-temperature cycle for peripheral fiber.

The method of solution was patterned after that given by Spera [4].
Total strain is considered to be the sum of three components; that is

(3)

where
£, = instantaneous total strain,
£, = elastic component.

£p = time independent plastic component, and

£^ = time-dependent plastic or creep component.
The instantaneous stress, a, can be specified as a function of the elastic
strain via the elastic modulus (E); that is

o = Ee^ (4)

Since e, and E are specified as a function of time, the problem is to deter-

mine the instantaneous values of o, £p and e^.
For solution of Eq 3, £p and Cc were expressed as functions of o.
The time-independent plastic strains (Cp) were determined from the tem-
perature dependent stress-strain curves shown in Fig. 7. In the solution
technique, the curves were represented by a series of six stress-strain data

02 04 06 08 1.0
STRAIN,%

FIG. 7—Stress-strain curves.

points connected by linear lines; curves for intermediate temperatures

were defined by linear interpolation. Transfer from one stress-strain curve
to another (or one temperature to another) was at constant total strain.
Hence, in functional form

£p = f(a,t„T) (5)

As discussed earlier, creep strain was specified in the form of Eq 1.

Since both £p and £c are nonlinear functions of stress, a numerical solu-

tion technique was applied. Following Spera's [4] lead, the method of
Runge and Kutta [12] was utilized to set up a time incremental calculation
procedure.

Axisymmetric FE Method
The finite element computer program used in this analysis is based on
incremental plasticity theory and the von Mises yield condition emd asso-
ciated flow rule. Although capable of treating combined isotropic and
kinematic hardening, the present analysis was restricted to purely iso-
tropic hardening. The same creep relationships were used as with the
UESI method. Creep stress transfer was assumed in accordance with the
strain hardening rule. The program is currently restricted to the use of bi-
linear stress-strain relationships. The bilinear approximations employed in
the analysis are indicated by dotted lines in Fig. 7.
As discussed earlier, the finite element model used in the stress-stredn
calculations was a very much simplified model with only 16 elements and
33 node points. This simplification was used to reduce the cost of the in-
elastic analysis after preliminary comparisons with results from the com-
plete FE model indicated that this reduced model was adequate to accu-
rately calculate the stress-strain response at the periphery of the disk.

Results and Discussion

The shape of the calculated hysteretic loops were found to be relatively
insensitive to analysis method, material behavior assumptions and test
conditions. Figure 8 shows an example of a calculated loop in which the
UESI method was employed. Temperatures achieved at different locations
are indicated by the crosshatching. It is instructive to trace through this
loop to gain a feeling for the complexities involved in transient thermal
fatigue. Starting at Point O, the initial heating shock drives the peripheral
material into compression along Path OA. At Point A, the stress has
achieved the proportional limit, and time-independent plastic flow com-
mences. The stress level does not become increasingly negative beyond
this point because the proportional Umit stresses are decreasing rapidly
with the increasing temperature.
Point B is the maximum negative strain developed in the heating shock,
and thereafter the strain and stress increase. The initial stress increase is
elastic. At some Point B', prior to achieving positive stress values, creep
commences. Because the temperature is increasing along with the stress,
the creep becomes increasingly prominant, and the rise in stress slows
down due to concomitant stress relaxation. At some Point C, the thermal
strain available for increased stress is outweighed by the creep induced re-
laxation, and the stress decreases in magnitude again. For this example.

Tmox=l688''F,Rp=O.OIin.
NUMBERS IN PARENTHESES F(800)
INDICATE TEMPERATURE
IN'F

FIG. 8—Calculated hysteretic loop for peripheral fiber.

the stress continues to relax to Point E after the strain achieves a zero
value (Point D).
The cooling shock introduces tensile strain and intially increasing ten-
sile stress. Since the cooling shock starts at a high temperature and the
stress is increasing, creep occurs initially in the cooling shock. At some
point, designated E' in Fig. 8, the temperature dips below where creep
and stress relaxation will occur. The stress and strain continue to increase
to Point F, which is the maximum positive strain generated by the cooling
shock. The peculiar curvature obtained in loading from Points £ to F is a
result of the initial stress relaxation and rapid increases in the flow stress
with the decreasing temperature. At Point F, the strain begins to de-
crease, and the fiber unloads elastically. The unloading extends into the
compressive stress region to the zero strain position. Point G. Close to the
zero strain point, compressive plastic flow initiates. Upon reintroduction
of the heating shock, time-independent plastic flow continues to the mini-
mum strain position (Point B).
A solution from the FE method was obtained for the test conditions
discussed in Figs. 6 and 8. This result is compared with the corresponding
UESI method solution in Fig. 9. Figure 9a shows strain-temperature his-
tories, and Fig. 9b, hysteretic loops. For consistent comparison, the hyster-
etic loop calculated by the UESI method employed isotropic hardening
and the bilinear stress-strain curves.

0.2
STRAIN,

1800

(a) Strain-temperature cycles.

FIG. 9—Comparisons of calculated results using the UESI and FE methods of analysis.

— UESI METHOD
o FE METHOD

-0.5 -0.4

(b) Hysteretic loops.

FIG. 9—Continued.

Examining first the peripiieral fiber strain-temperature behavior (Fig.

9a), it can be observed that the two calculational methods predict nearly
identical histories. The maximum difference in strain at a given tempera-
ture is about 10 percent, with the more rigorous of the solutions indica-
ting more positive magnitudes. This result would appear to justify the as-
sumption of elastic-strain in variance for this problem.
Differences in the calculated hysteretic loops (Fig. 9b) result mainly
from the slightly different strain-temperature histories; that is, the FE
method calculated loop is of the same proportions but translated in the
positive strain direction. Also, the total inelastic strain calculated is nearly
equal, with, however, less of it in creep strain for the FE method. The
latter shows up as higher stresses in the negative strain region and appears
to result from the slightly lower total compressive strain amplitude. In
conclusion, it can be stated that treatment of the peripheral fiber as a uni-
axial member is a good approximation.
This comparison indicates that the two calculational methods predict
the same behavioral trends. In most temperature regions, identical be-
havior is predicted. On the basis of this result, the simpler UESI method
was utilized to investigate the effects of material behavior assumptions
and test variables.
The calculated stress-strain response of the peripheral fiber is affected
by the choice of assumption used in describing the Bauschinger effect and
stress transfer in creep. Example results are illustrated in Figs. 10 and II.
Figure 10, which shows hysteretic loops calculated assuming the three dif-
Tmoi=l688°F,Rp=0.0l
-ISOTROPIC THEORY
-IND. TENSION 8 COMPR. YIELD
-KINEMATIC THEORY

0.4 0.5

FIG. 10—Hysteretic loops calculated with differing assumptions on the Bauschinger effect
{UESI method).

Tm0K = l688*F,Rp=0.0liii.
• STRAIN HARDENING
-TIME HARDENING

FIG. 11—Hysteretic loops calculated with the strain hardening and time hardening rules
for creep-stress transfer (UESI method).

ferent assumptions on the Bauschinger effect previously discussed, indi-

cates that the isotropic hardening and independent tension and yield as-
sumptions predict the same behavior in the tension domain. However, as
would be expected, the isotropic theory predicts greater compressive stresses.
In contrast, the kinematic theory, which gives rise to a cycle effect, predicts
on shakedown (five to six cycles) higher tensile stresses in the high temper-
ature tension domain but lesser stress elsewhere. It also predicts somewhat
larger plastic strain ranges.
Figure 11 shows loops resulting from the time and strain hardening as-
sumptions in the creep analysis. Time hardening gives rise to less total
creep and, therefore, higher stresses in the high temperature tensile do-
main. The resulting difference in total inelastic strain range is however,
insignificant. In other words, if available elastic strain is not converted to
creep, additional time-independent plastic strain results.
Figure 12 shows examples of loops calculated for other test conditions.
Figure \2a of the figure shows the effect of Rp at constant Tmax and 4 ;
Fig. Mb, the effect of T^^^ at a constant Rp and t^; and Fig. 12c, the dif-
ference between t^ = 1 and 4 min at constant Tmax and Rp. Observe that
the stresses achieved in each half of the cycle vary little with test condi-
tion. The major effect that each variable has on loop size is the total in-
elastic strain.
Ultimately, one would want to use the results of analyses like these for

Tmox = l688''F

Rp=O.OI in
Rp=0.02
Rp=0.04

-0.5 - 0 4

(a) Effect of « , .
FIG. 12—Calculated hysteretic loops for various test conditions (UESI method).

Rp=0.0lin

-0.6 -0.5

(6)Effectof r^ax.
FIG. n—Continued.

Tmo« = l778°F,Rp=0.0lin
•th=4MIN
• th^l MIN

0.3 0.4
STRAIN, %

fatigue life predictions. Although it is not the purpose of the present pa-
per to examine fatigue life of the disk specimens, it is pertinent to exam-
ine the analysis results in light of potential application of current high
temperature life prediction concepts. The following three damage con-
cepts will be considered: (a) creep damage (Spera [4\, (b) strain range
partitioning (Manson et al [13]), and (c) frequency modified fatigue life
(Coffin [14]).
The creep damage hypothesis assumes that high-temperature fatigue
damage is solely the result of creep damage. In his application of it, Spera
proposes summing the damage via the linear life fraction rule, with the as-
sumption that creep in compression is additive to that in tension. For
varying temperature cycles, the creep damage per cycle, <t>, is given by

(6)
0 tr

where
A^ = time increment and
tr = creep-rupture time for the average stress and temperature within
the time increment.

Failure is assumed to occur after A^ cycles when

N^ = l.O (7)

Use of this concept then requires knowledge of the isothermal stress-rup-

ture properties in the temperature range of interest but no fatigue proper-
ties.
The strain range partitioning concept hypothesizes that the damage
process is dependent upon the type of inelastic deformation. The inelastic
strains present in a hysteretic loop are partitioned into the following com-
ponents:
1. £pp, tensile plastic strain reversed by compressive plastic strain (that
is, all nontime-dependent deformation).
2. £p,, tensile plastic strain reversed by compressive creep strain, or £,,,,
tensile creep reversed by compressive plastic strain.
3. £,.,, tensile creep strain reversed by compressive creep strain (that is,
all time-dependent deformation).
Further assumptions in the concept include relating each of these strain
components to life through the Coffin-Manson equation and summing
the damage caused by the various strain components in a linear fashion.
That is

M pp Fpp " =
^
A
•^^
(8)
' ' cp ^cp -*-'

etc., and

1 1 1 1 1
+ -1- + N,, = Nf (9)
A'.. N,.

In these relations, Npp, N^p, etc., are lives corresponding to the properly
subscripted strain components £pp, £,.p, etc.; Ny is the life when more than
one component is acting.
For thermal loading of the type examined in this paper, only the £pp and
£,p components are significant. These correspond to the Cp and £,. compo-
nents calculated. To use the concept, one would then have to know the
isothermal fatigue properties, a, ft, A, and B. It has been proposed, on
the basis of some initial experimental evidence, that these are independent
of temperature [15]; that is, damage from each type of deformation is in-
dependent of temperature. This possibility would greatly simplify a dam-
age analysis.
The frequency modified fatigue life approach is based on the hypothesis

that high-temperature fatigue damage is, in the main, environmental

damage. The Coffin-Manson relationship is modified, based on the fol-
lowing relationship between frequency and time to failure

v^tf = c, (10)

where
V = frequency of cycling,
tf = time to failure, and
k, c = material constants.
It is further assumed that Ci is a constant for each plastic strain range
and that the following relationship exists

C/AE,^ = C2 (11)

where
Lt,p = the total plastic strain range (Atp + AeJ and
ji, C2 = material constants.
Taking vtf = Nf, the following modified Coffin-Manson form can be
arrived at

(NfV'- ^YM,, = c^ (12)

Employment of this concept in transient thermal fatigue analysis requires

obtaining values of k, p, and C2 from isothermal fatigue tests. They are
known to be temperature dependent [16] so that a damage analysis would
have to assume some average or dominant values.
The analysis methods used in this paper readily provide one with these
inelastic strain components. Also, a creep damage computation is easily
facilitated. On examining the magnitudes of these quantities, they appear
to be quite sensitive to the material behavior assumptions embodied in the
stress analysis. As an example. Table 3 presents results for one test condi-
tion. It can be observed that AE^, AE^ , and ^ vary widely, depending upon
the analysis assumptions. As an extreme, the time hardening assumption
predicts the lowest creep strain and largest creep damage. Only A£,p is rel-
atively insensitive to the computation assumptions.
As a final observation, it is noted that none of the concepts examined
explicitly treat the effect of mean stress bias. This may be an important
omission, since mean stress biases are inevitably present in thermal load-
ings. The creep damage approach will predict an effect for some cases,
but it would seemingly be in the wrong direction because compressive
creep is considered additive to tensile.

TABLE 3—Predicted damage variables for the test condition

( T „ „ = J688°Fand Rp = 0.01 in.).

Material Behavior Assumption Variables

Analysis Bauschinger Creep Stress

Method Effect Transfer A£,p Atp At, •ti, %

UESI IT and CY" £-hard 0.495 0.198 0.297 0.22

UESI IT and CY /-hard 0.463 0.278 0.185 0.88
UESI isotropic £-hard 0.492 0.195 0.297 0.22
UESI kinematic £-hard 0.610 0.350 0.260 0.06
FE' isotropic £-hard 0.532 0.301 0.231
UESI' isotropic £-hard 0.515 0.210 0.305 0.31

" Independent tension and compression yield.

'' Bilinear stress-strain curves.

In conclusion, it is suggested that each of these damage concepts could

be applied to the analysis of thermal fatigue upon obtaining the appro-
priate isothermal properties. The strain range partitioning concept appears
to be the easiest to apply. No statement can be made at this time on the
accuracy of each concept. It is apparent that the predictions from any of
the concepts would be sensitive to the material behavior assumptions
chosen for the stress analysis. The controlling variable in the frequency
modified approach appears to be the least sensitive in this regard. Further
testing and analysis effort to establish the best (or better) material be-
havior assumptions appear necessary at this time.

References
[/] Glenny, E., et al, "A Technique for Thermal Shock and Thermal Fatigue Testing Based
on the Use of Fluidized Solids," Journal of the Institute of Metals, 1958, Vol. 87, pp.
294-302.
[2] Howes, M. A. H. in Fatigue at Elevated Temperatures, ASTM STP 520, American So-
ciety for Testing and Materials, 1973, pp. 242-254.
[3] Howe, P. W. H. in Thermal and High-Strain Fatigue, The Metals and Metallurgy
Trust, 1967, pp. 122-141.
[4] Spera, D. A., "The Calculation of Thermal Fatigue Life Based on Accumulated Creep
Damage," NASA TMX-52558, National Aeronautics and Space Administration, 1969.
[5] Mowbray, D. F. and Woodford, D. A., "Observations and Interpretation of Crack
Propagation Under Conditions of Transient Thermal Strain," International Conference
on Creep and Fatigue, Institute of Mechanical Engineers, 1973, pp. 179.I-179.il.
[6] Foster, A. D. and Sims, C. T., Metal Progress, 1969, pp. 83-95.
[7] Mendelsohn, P/aif/c/y.' Theory and Application, MacMillan, New York, 1968.
[51 Hunsaker, B., et al, "A Comparison of the Capability of Four Hardening Rules to Pre-
dict a Material's Plastic Behavior," Report #75-51. Texas Engineering Experiment
Station, June 1975.
[9] Penney, R. K. and Marriott, D. L., Design for Creep, McGraw-Hill, London, 1971.
110] Manson, S. S., Thermal Stress and Low-Cycle Fatigue, McGraw-Hill, New York, 1966.
[11] Topper, T. H. et al. Journal of Materials, Vol. 4, No. I, 1969, pp. 200-209.

[12] Hildebrand, F. B., Advanced Calculus for Engineers, Prentice-Hall, Englewood Cliffs,
N.J., 1949.
[13] Manson, S. S., Halford, G. R., and Hirschberg, M. H., "Creep-Fatigue Analysis By
Strain Range Partioning," NASA TMX-67838, National Aeronautics and Space Ad-
ministration, May 1971.
[14] Coffin, L. F., Jr., Journal of Materials, 1971, Vol. 6, pp. 388-402.
[15] Halford, G. R., Hirschberg, M. H., and Manson, S. S. in Fatigue at Elevated Tempera-
tures, ASTM STP 520, American Society for Testing and Materials, pp. 658-669.
[16] Coffin, L. F., Jr., Proceedings, Institute for Mechanical Engineers, 1974, Vol. 188, pp.
109-127.

Thermal Stress Concentration

Factors in Large Shafts

REFERENCE: Gonyea, D. C , "Thermal Stress Concentration Factors in Large

Shafts," Thermal Fatigue of Materials and Components, ASTM STP 612, D. A.
Spera and D. F. Mowbray, Eds., American Society for Testing and Materials, 1976,
pp. 30-37.

ABSTRACT: The concentration of thermal stress in a shaft subjected to a ramp

surface temperature change has been investigated. The concentration occurs at a step
change in shaft diameter, and the temperature and stress responses are calculated by
the finite element method. A thermal stress concentration factor (TSCF) is defined,
and an analytical expression is given which defines the concentration factor in terms
of the geometric parameters. Over the range of transient conditions studied, it is
observed that the TSCF is insensitive to the surface temperature ramp rate. A com-
parison illustrates that the TSCF can be significantly different in numerical value than
mechanical stress concentration factors.

KEY WORDS: thermal fatigue, thermal stress, stress analysis, stress concentration

Power generating equipment is frequently subjected to operating con-

ditions that give rise to thermal stresses. In many cases, these thermal
stresses are large and limit the operational flexibility of power plants. In
connection with these limits, the transient thermal stresses developed in
the rotors of large steam turbines have been studied extensively.^" An
important aspect of these studies is the use of a thermal stress concentra-
tion factor to account for variations in the geometry on the outer surface
of the rotor.
The purpose of this paper is to describe the nature of thermal stress
concentrations at step changes in the diameters of large rotors. An ex-
pression is presented that defines an effective (Von Mises) stress con-
' Development engineer. General Electric Company, Schenectady, N.Y. 12305.
^Spencer, R. C. and Timo, D. P., "Starting and Loading of Large Steam Turbines,"
presented at the 36th Annual Meeting of the American Power Conference, April 1974,
Chicago, 111.
'Berry, W. R. and Johnson, L, Journal of Engineering for Power, July 1964, pp. 361-
368.
"Manson, S. S., Thermal Stress and Low Cycle Fatigue, McGraw-Hill, New York, 1966.

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
GONYEA ON LARGE SHAFTS 31

centration factor in terms of geometric parameters for two idealized

rotor shapes, alone and in combination. Thermal stresses are assumed to
be developed due to slow uniform heating of the rotor surface.

Modeling Technique
Figure 1 depicts the three general types of geometries that have been
studied, and Table 1 lists the actual dimensions utilized in this study. A
total of 15 geometries has been investigated which encompasses the range
of parameters typically encountered in actual turbine geometries of the
author's company.
'tufytff

% ROTOR

FIG. 1—Sketch of three types of geometries studied and definition of geometric param-
eters.

TABLE 1—Geometry description and thermal stress concentration factors found using
finite element approach. All dimensions are given in units of inches. See Fig. I for definition
of symbols.

Case D DS D\ R L TSCF

1 20 42 3 3.72
2 20 30 3 2.66
3 20 30 1 3.57
4 20 22 1 1.77
5 20 24 1 2.33
6 20 20.5 0.25 1.74
7 12.3 24 4 2.60
8 22 22 39 0.5625 5 2.03
9 29.25 29.25 39.6 1.25 6 .75 1.80
10 22 22 28.25 0.4375 2 .25 1.74
11 20 30 40 4 4.06
12 20 26 30 4 3.39
13 20 23 30 A 2.56
14 20 23 26 4 2.67
15 20 26 30 6 3.19

A two-dimensional, axisymmetric finite element technique is used to

study the thermal stress response of these geometries. Figure 2 illustrates
the boundary conditions that are applied to the finite element model.

SURFACE
TEMPERATURE r
• THERMALLY INSULATED
• AXIAL DISPL. EQUAL
SAME AS • NO SHEAR LOADS
OPPOSITE • NET AXIAL LOAD ZERO
END

• SOLID, AXISYMMETRIC
• UNIFORM INITIAL TEMPERATURE
• NO EXTERNAL OR BODY LOADS

FIG. 2—Thermal and mechanical boundary conditions utilized in finite element study.

Thermal boundary conditions are applied which approximate those

experienced by a turbine rotor during startup. Specifically, the surface
temperature is assumed to be subjected to a finite linear temperature in-
crease with respect to time. A uniform initial temperature is assumed, and
the entire outer surface is given the same temperature variation with
time. A sohd shaft is used, and the ends of the finite element model are
insulated to simulate a long shaft.
The mechanical boundary conditions at each end of the model are such
that all axial displacements are equal, no radial shear exists, and the net
axial load is identically zero. These conditions, as well as the length of the
finite element model, are selected to approximate a long, unrestrained
shaft. Examination of the results has shown these conditions to be es-
sentially met. Deviations of up to 3 percent between the finite element
results at the end of the shafts and classical solutions were found. In the
case of the "wheel" geometry, advantage is taken of axial symmetry in
geometry and loading about the center of the wheel. The model is as-
sumed to be free of external and body loads in order to investigate solely
the effects of thermal stress.
In the calculations described herein, the material is assumed to be
homogeneous and isotropic and the material properties independent of
temperature. Nominal values of material properties are used which are
typical of turbine rotor alloys. These values are listed in Table 2.
Both the transient thermal and elastic stress analyses utilize the same
finite element grid network. The actual computer codes employed in the
calculations are programs developed by the author's company which in-
corporate the standard equation and solution techniques common to most
two dimensional finite element codes.

TABLE 2—Numerical values of thermal and mechani-

cal material properties utilized in the analysis.

Thermal conductivity = 1.65 Btu/h- in. °F

Specific heat capacity = 0.15 Btu/lb °F
Density = 0.283 lb/in.'
Thermal expansion coefficient = 8 x 10"' in./in. °F
Young's modulus = 30 x 10' in./in. °F
Poisson's ratio = 0.3

Temperatures/Stress Response
Figure 3 illustrates the temperature and stress response that is found to
result from the surface ramp temperature change. The interior tempera-
tures lag the surface temperature on heating and reach a quasi-steady

OUTER / ^
SURFACE / ^

TEMPER- \ / /
ATURE / / \cENTER OF
SHAFT
/ /
L^"

MAX. FILLET
' STRESS, <r'

THERMAL NOMINAL SURFACE

STRESS ^ STRESS, a
L—\V
1 "V.
FIG. 3—Temperature and thermal stress response. Von Mises equivalent stress illustrated.

State condition after approximately 2 h. At this time, the calculated

stresses reach their maximum values and remain constant until the end
of the ramp, at which time the stresses decay as temperature equilibrium
is approached.
During heating, compressive surface stresses develop. On the shaft sur-
face removed from the concentration, an equibiaxial state of stress
occurs. To account for the complex states of stress both nominally (re-
moved from the fillet) and locally (at the maximum point at the fillet),
the Von Mises equivalent' stress is used. The use of Von Mises equivalent
'The Von Mises equivalent stress, o, is defined as

o = - = [(o, - o,y + (oj - o,)^ + (03 - o,yf'

V2
where
"1,02, oj = principal components of stress.

Stress is motivated by design considerations which consider the Von Mises

equivalent stress as the quantity of interest in evaluating the low-cycle
fatigue performance of the rotor due to thermal cycling.^
In addition, a thermal stress concentration factor (TSCF) is defined

Oa
TSCF = (1)

where
TSCF = thermal stress concentration factor,
oa = maximum Von Mises equivalent stress, and
Ooo = Von Mises equivalent stress removed from the fillet.
In Eq 1, the values of stresses are taken after a quasi-steady-state
temperature field develops. During the initial heating portion, the magni-
tude of the thermal stresses are less than those developed during quasi-
steady state. Figure 4 illustrates this initial time dependent nature of the
thermal stresses.

_ K..1 '<, = ^ A T T , M E ,

K, = — AT QUASI-STEADY STATE

a = THERMAL DIFFUSIVITY

t = TIME FROM START OF RAMP

D = LARGER SHAFT DIAMETER

_l_ _i_ _L. _L. J

0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6

FIG. 4—Time dependence of thermal stress concentration during initial portion of heat-
ing. Note that the range illustrated includes wheel, step, and combined geometries.

In addition, the effect of various heating rates on the values of TSCF

has been studied. Model 8 was subjected to surface heating rates of 100
and 333 °F/h, and the value of TSCF was found to be identical in both
situations. Other geometries that have been studied display the same be-
havior. Hence, TSCF is independent of heating rate when a quasi-steady
state temperature field exists.
For other models, a heating rate of 200°F/h is utilized and TSCF de-
termined after 5 h (at the end of the heating). Table 1 lists the values of
TSCF determined in this manner for the 15 geometries studied.
The peak value of thermal stress occurs in the fillet region, a distance

of approximately 0.3 R from the intersection of fillet radius, R, and the

shaft. The surface stress decreases rapidly around the fillet and along
the shaft. Figure 5 illustrates this variation of stress in a normalized
fashion. The scatterband illustrated includes the results from all geom-
etries studied. Note the large decrease in thermal stress within a distance
equal to the radius. This plot is useful for estimating the interaction ef-
fects of adjacent regions of concentration for more complex geometries.

0.3^HJ X <. DISTANCE ALONG SURFACE

•MILIUITX
")l!Alii2>^ ''X " STRESS AT X
c^Q = STRESS ATX = 0

(Too = STRESS AT X - • « >

FIG. 5—Normalized distribution of surf ace stress in region of fillet. All cases investigated
included in scatterband.

Approximate Equation for TSCF

Because it is impractical to study all geometries of interest with the
finite element approach, an approximate analytical expression has been
developed which relates TSCF to the geometric parameters involved. As a
starting point, TSCF values associated with the "step" geometry are used
to establish a fundamental relationship. The more complex geometries,
(the wheel and combined cases) are related to an equivalent step geometry
to provide a secondary relationship. This approach follows that of Peter-
son' in defining the mechanical stress concentration factor for a shaft in
bending.
The approximate relationship that has been developed in this manner,
using the geometric parameters defined in Fig. 1, is

= -v^[^°->f
TSCF = I +\/ - \- ^ - 1r (2)

'Peterson, R. E., Stress Concentration Design Factors, Wiley, New York, 1953.

where
DEQ DS/D for step geometry
D minimum [Dl/D, DS/D + 0.35 L/DS] for wheel and com-
bined geometry
It should be noted that TSCF is defined in terms of the nominal equiv-
alent stress at diameter, D, in all cases.
Figure 6 illustrates that Eq 2 describes the results of the finite element
studies connected with the step geometries in the range of parameters
studied. Figure 7 is a direct comparison of TSCF found by the finite ele-
ment technique and by the proposed analytical relationship, Eq 2 for all
cases studied.
10
R ^
X'P^-^
TSCF-1
1
. ^ '^X
y. ^

0.1 ' J..I 1 1 1 1 11 1 I I I

0.01 0.1 1
SS_i
D

FIG. 6—Comparison of finite element results for step geometry and equation developed
to define thermal stress concentration behavior.

"STEP" TYPE GEOMETRY

"WHEEL" AND "COMBINED"
. GEOMETRIES

1 2 3 4
TSCF FROM EQUATION

FIG. 7—Comparison of TSCF calculated using the finite element technique from Eq 2.

A comparison of TSCF and the mechanical stress concentration in

bending given by Peterson' highlights another aspect of TSCF. Figure 8
illustrates such a comparison. Observe that, for small diameter ratios, the
value of TSCF is roughly equal to or less than that of mechanical bend-
ing. However, for large diameter ratios, the values of TSCF is signifi-

0.0^^^'

• TSCF STUDIES
MECHANICAL BENDING FROM PETERSON
THERMAL STRESS CONCENTRATION EQUATION
0 1 1 LXJ I I l_l_L -L. -L.
0.01 0.1 10

f-'
FIG. 8—Comparison of thermal and mechanical stress concentration factors for shaft
with step change in diameter.

cantly larger than the corresponding mechanical value. This effect is due
principally to a thermal mismatch mechanism that gives rise to additional
stress in the case of the thermal loading, particularly for large diameter
ratios. The difference in average temperatures between the large and small
sections of shaft causes this thermal mismatch. A similar mechanism does
not exist for the case of mechanical loading.

Summary
The finite element technique has been used to investigate the nature
of thermal stress concentrations in large turbine rotors during tempera-
ture transients. On the basis of this analysis, it is tentatively concluded
that the thermal stress concentration factor, TSCF, is independent of time
and heating rate under conditions of slow uniform heating in which a
quasi-steady state temperature field exists. An approximate relationship
is given, Eq 2, which defines TSCF in terms of the geometric parameters
of the rotor. A normalized plot of the distribution of thermal stress in
the vicinity of the fillet region is also given. A comparison of TSCF
with the mechanical stress concentration factor due to bending illustrates
that there may be a significant difference in numerical values of these
two factors. For large diameter ratios, the value of TSCF is larger than
that due to bending, owing to an additional thermal stress component
which is due to a thermal mismatch mechanism.

Low-Cycle Fatigue Analysis of the

Turbine Disk for the National
Aeronautics and Space
Administration High-Temperature
Turbine Rig

REFERENCE: Tepper, S., "Low-Cycle Fatigue Analysis of the Turbine Disli for
the National Aeronautics and Space Administration High-Temperature Turbine
Rig," Thermal Fatigue of Materials and Components, ASTM STP 612, D. A. Spera
and D. F. Mowbray, Eds., American Society for Testing and Materials, 1976, pp.
38-54.

ABSTRACT: The high-pressure, high-temperature turbine test rig was designed for
the National Aeronautics and Space Administration to test turbine components in an
environment that simulates current and future gas turbines operating at 41 atm
(600 psi) and 2204 °C (4000 °F). This paper illustrates the design approach that led
to selection of a typical test rig component: the turbine disk. The final configuration
of the rotor assembly and the turbine disk was determined by analysis of low-cycle
fatigue and creep.

KEY WORDS: thermal fatigue, creep (materials), disks, thermal cycling tests, plastic
analysis, substructures, turbines

During the preliminary design of mechanical components, especially

during definition of component configurations, the analyst must establish
the controlling conditions that affect the design. This paper summarizes
the analytical approaches used in designing the turbine disk for the
National Aeronautics and Space Administration (NASA) Lewis Research
Center's high-pressure, high-temperature, turbine test rig (HTTR).^
The Larson-Miller parameter (LMP) approach was followed in the
creep analysis. Life, as calculated from the plastic analysis, was updated
by a more accurate study of stress relaxation and cyclic operation. In the
low-cycle fatigue (LCF) analysis, the cycles to failure were determined
' Senior engineering specialist. Applied Mechanics Group, AiResearch Manufacturing
Company of California, Torrance, Calif. 90509.
' "High Pressure High Temperature Turbine Test Rig," NAS3-17736, 19 July 1973.

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
I
TEPPER ON LOW-CYCLE FATIGUE 39

by the universal slopes method according to Manson and Halford [7]' and
considering the studies of Morrow, Wetzel, and Topper [2,3]. The inter-
action between creep and LCF, and the strain range partitioning approach
are also discussed [4-6].

The Turbine Disk

The HTTR was designed and analyzed in terms of a number of op-
erating conditions. The objective was to test turbine components in an
environment simulating current and future gas turbines, with maximum
operation at 41 atm (600 psia) and 2204 °C (4000 °F).
The wide range of HTTR operating conditions was imposed by large
variations in operating speed, temperature, and pressure. The rotary as-
sembly has three specified operating speeds: idle, 12 000 rpm; maximum
cyclic, 16 667 rpm; and maximum operational, 17 500 rpm. Its overspeed
design condition is 21 800 rpm. All intermediate speeds between 0 and
17 500 rpm are possible. Cooling is provided by pressurized air, with
temperatures varying between 260 and 621 °C (500 and 1150°F). Inlet
pressures may be as high as 41 atm (600 psia).
Because of its critical nature, the Astroloy turbine disk was subjected
to many combinations of temperature maps, rotating speeds, pressures,
and interface loadings. From the configurations studied, a design with
cover sideplates was selected. These sideplates, which are bolted as units
to the disk, enclose two chambers through which two separate flows of
cooling air are carried to two types of turbine blades.
The two final candidate designs are depicted as finite-element models
in Figs. 1 and 2. Once the second design was adopted (Fig. 2) controlling
conditions became those that affected the interface between the cover
plate, bolt, and disk.

Preliminary Analysis Process

The following initial guidelines are recommended to analysts investi-
gating similar tasks.

Initial Guidelines
Step 1—Establish all specified and tabulated operating conditions. If
each contributory action for a given operating condition could be linearly
superposed, the following expression could be used during the preUminary
design
^ = i i'^Sai
Sa
^<l
' The italic numbers in brackets refer to the list of references appended to this paper.

FIG. 1—Finite-element model of Candidate Disk l.

FIG. 1—Finite-element model of Candidate Disk 2.

where
S = any design condition, for example, stress and hours of life,
a = allowable value,
e = equivalent total, and
/ = corresponding contributory condition.
In this analysis, the only linear contributory load, neglecting differen-
tial stiffening effects, is the pressure. For a given temperature distribution
and rotational speed, only two conditions were analyzed: zero and maxi-
mum pressure. The other two contributory actions were nonlinear, that
is, temperature and rotational speed.
Step 2—Establish intermediate operating conditions if the acting loads
cannot be linearly superposed. In the present study, seven speeds (six
operational and one for overspeed design), four pressure values (two at
each side of the disk), and six temperature maps existed. These variables
produce a total of 192 combinations. However, not all of these combi-
nations are operational.
Step 3—Discard nonoperational and minor operating conditions. Once
the obviously insignificant conditions are eliminated, the discarding of
some intermediate operating conditions can be very challenging. An
example to be considered would be, which temperature maps are the most
significant? Here one must choose between high-temperature conditions
with lower temperature gradients, and low-temperature conditions with
large temperature gradients. The best method is to keep both extreme
conditions and at least one intermediate condition.
Step 4—Retain conditions that define a cyclic operation. Basically, one
tends to study or select only the highly loaded conditions because of their
relation to ultimate or yield strength and to creep life. This approach
is valid when the representative mission profile indicates few cycles.
Frequently, however, that is not the case.
The mission profile analyzed here comprises many long and short tests
in which the corresponding transients affect each of the contributory
parameters—speed, pressure, temperature—and their mutual relationships.
Cyclic life is a fundamental condition to satisfy.
Step 5—If conditions allow, start using an elastic model. Once the con-
tributory loads are defined (in this case, speed, pressure, and temperature
maps), an elastic finite-element model may be used. This model may be
used even if the operating conditions are not Unear. Use of an elastic
finite-element model assumes that: (o) the possible plastic areas are con-
fined to local small spots and (b) these spots will not create a mechanism
between two or more parts of the model.
Elastic models are more economical than plastic models because they
run faster in the computer. Also, there are many large computer pro-
grams able to accept elastic models. In many cases, an elastic-plastic

solution for a given operating condition is valid if previous elastic-plastic

solutions were run and the correct history was kept.
As the conditions just mentioned are met, the analyst may record
the results of the most severe condition, as well as the conditions that
are important enough to deserve a full elastic-plastic ansdysis.
Nonlinear Problem And Finite-Element Model
Figure 3 shows the adopted disk configuration in which loads and
moments represent blades, bolts, and seals. At the given temperatures and

49 088 LB (0.22 MNI

A V( 19 630 IN..
y ^ (2218 Nm|
24 906 IN.-LB
(2814 N m )

960 IN.-LB a
(108 Nm) vj,

FIG. 3—Rotor interface loads and temperatures at 17 500 rpm.

speed, with an internal pressure of 20 atm (300 psia) in the cavities,

the corresponding effective Von Mises stresses and displacements are
shown in Figure 4. Figure 5 depicts the estimated plastic flow based
on the elastic analysis for 17 500 and 21 800 rpm.
A cyclosymmetric finite-element model of the main disk, side plate,
and bolt was used. At 17 500 rpm, the maximum operational speed,
plasticity was found only at the disk hub. However, owing to the
coarseness limitations of the model, a better representation of the side
plate-bolt area was desired. Of particular interest was finding a simple
way to examine all important loading combinations without constructing
large models. One clear method was to separate the main disk from the

+ 0.0022860
+0.0003566 r—1-^

+ 0.0020674
mj
8° ^^^^^^ \ + 0.0018034
60 Tr"^'/^

\
SO 1 *^

(8.701

L
r ^
+ O.OO13970

+0.0004572 1
' li=L,4+0.000«096

100 1===^
no L + 0.0006080
130 (18.861
+0.0003912-
w
b.i
a

(a) Effective stress, ksi (MPa).

(i>) Displacements, m.
FIG. 4—Effective stresses and displacements at 17 500 rpm.

side plate and establish the matching and boundary loads for a sub-
structure surrounding the area under study.
Also, two other actions (bolt loosening and tightening) were simul-
taneously analyzed. By using a bolt made of the same material as the
disk, the thermal expansion was the same for both. However, the cen-
trifugal field acting on the disk produced two principal stresses (radial Op
and hoop Oe) of the same order, while the bolt is mainly under axial
loading. Because the Poisson's ratio effect narrows the disk farther than
the bolt, the bolt could become loose if sufficient preloading is not
provided.
Once preloading is defined, the effect of the following must be investi-
gated: (a) the different working conditions at room temperature and at
idle and maximum operating speeds, {b) the creep associated with the
stress and temperature field, and (c) the maximum manufacturing toler-
ances allowed.
Figure 6 depicts the load distribution at the disk-sideplate interface.
The contact pressure loads are transferred through the bolt. Matching

5027 HR

LEGEND:

AT 17,500 RPM

^ AT21.S00RPM

FIG. 5—Plasticflowand life, hour.

FIG. 6—Distribution loads at interface.

of the resultant loads and the structural displacement was accompUshed

for the two substructures (main disk and sideplate) under seven load

conditions. Five nonlinear equations were solved in terms of displace-

ments by the Newton method. The five equations are

<5p,,.," + Pxip^iP^.n)" + Hidp,(H.„)" = 6p,(P,,H,n)

^PiM + PldPiiPi.r,) ~<ip,lP„H,n)

P, + P2~0
OPtiPi.Pt.H.n) — Op,(^p,^p,,H,n)

where
subscripts = degrees of freedom associated with the
loads,
variables between parenthesis = nonlinear dependencies,
superscripts = loading configurations,
" = free centrifugal field, (17 500 rpm),
R = total resultant displacement along the
radial direction, and
Pi, Pi, and Hi = loading resultants from the contract
pressures PRS, PRD, and PH respec-
tively (see Fig. 6).
The solution of the system of equations gave the shear stresses at the
boh section and the axial preload required to keep the sideplate attached
to the disk. Considering the additional preload required to control the
thermal expansion, the Poisson's ratio effect, the manufacturing toler-
ance, and the creep relaxation, the maximum axial stress in the bolt
is 75 to 80 ksi. The lower axial stress is about 40 ksi for the worst com-
bination of operating conditions. Bolt retightening time is defined by both
creep relaxation and cyclic operation.

First Approach to Creep Study, Substructure Analysis

The region under study was the interface between the main disk and
the sideplate. The loads and boundary conditions were established from
the preliminary runs by using the finite-element model mentioned pre-
viously.
In this approach, an isoparametric elastic-plastic element was imple-
mented in an AiResearch computer program that has been used with
satisfactory results during the last four years. Figure 7 shows the tempera-
ture map. Figure 8 shows the effective stress maps at 17 500 and 21 800
rpm. By using LMP curves, a first solution for creep life was made.

TEMPERATURE, ° F r C I

FIG. 7—Temperature map.

These life zones are depicted in Fig. 9 for both 17 500 rpm and 21 800
rpm and the established creep values.
If the LMP curves reflect minimum values, a safety factor of four is
common practice. However, it should be noted that the expected Ufe
was based on an elastic-plastic solution where no stress relaxation was
considered.

Continued Analysis Process

At this point in the analysis, the following guidelines apply.
Step 6—For economy and simplicity, the finite-element model should
be established in such a way that possible substructures are easily identi-
fiable. A coarse model is an advantage if a critical area can be further
analyzed without reassembling the overedl model. Also, computer time,
which usually is an exponential function of the number of degrees of
freedom (related to the unconstrained nodes), can be substantially re-
duced by proper substructuring.
Step 7—Define the interface loads between two continuous substruc-
tures and then proceed as necessary to increment the number of elements

ioo\ \
11 \ K"
J \ioo
(14.50)y \ (17.40)YV 4ioo
V120 >*
100
/4^
nt
'120

0*0 \^
0 ,
/100
.
yi2o

^
120

100 /

(14.501
100

M /

1 ^
(a) At 17 500 rpm, ksi (MPa).
(b) At 21 800 rpm, ksi (MPa).
FIG. i—Effective stresses.

in the substructure. After the runs, recheck the interface and boundary
agreements.
Step <S—When plasticity is expected, run the full elastic-plastic analysis
for the critical substructure or for the total structure, if required. A
first assessment of possible plasticity should compare the effective stress
with the corresponding temperature and yield stress at a given point.
An effective stress below the yield stress does not guarantee elastic values.
However, the analyst will be able to check the different types of stresses
involved and the particular area under consideration. If a point shows
plasticity, all the substructures (or the structure) should be run in a plastic
solution.
Step P—For the specified creep value (1 percent, 2 percent, rupture,
etc.), define the creep Ufe map. Since no stress relaxation is considered
at this time, the LMP approach should suffice. A typical LMP expression
is

LMP = r(20 + log 0/1000

(a) 17 500 rpm.

(i>) 21 800 rpm.
FIG. 9—Plastic regions and hours of life.

where
T = absolute temperature (temperature °F + 460),
t = expected life in hours, and
log = logarithm to base 10.
Most of the materials have curves relating stresses to LMP

LMP=/{(o)K

where £c is a given creep value. Therefore, the life is easily calculated

from

logr = 1000 LMP/r - 20

Step 10—Clearly mark the plastic areas with special emphasis on the
following three aspects: (a) plastic areas on the surface, (b) if there is
more than one area, mark possible connectivity, and (c) plastic areas
that are large in proportion to their surrounding elastic areas.

Second Approach to Creep Study

Depending on the characteristics of the plastic areas, a more complete
study that includes stress relaxation could be considered. An AiResearch
computer program developed in 1971 allows an incremental time solution
of a plastic solution for afinite-elementmodel.
If only creep were considered, the analysis would be complete at this
point. However, a local plastic area (Fig. 9a) is always a concern in
low-cycle fatigue studies, and a finer analysis is required. In the present
case, plastic propagation was not found. The life values obtained were
slightly lower than those calculated previously.

First Approach to Low-Cycle Fatigue

This method is based on the classical universal slopes solution de-
veloped during the 1960's that provides additional allowance for creep
[1]. Because it uses a safety factor of .ten, it is considered a conser-
vative approach.
If the plastic area is relatively small and is surrounded by a pre-
dominant elastic region, the equivalent (linear) elastic effective stress
obtained from the elastic model can be applied by using the Neuber
hyperbola method.
It has been shown by Topper et al [2] that Neuber's rule [3] may be
applied to inelastic behavior characteristics at a notch root. Neuber
states that the theoretical stress concentration factor is equal to the
geometric mean of the actual stress and strain concentration factors.

K, = {K„ X K^y (1)

Peterson [7] has shown that notch size can affect the fatigue life and
that a fatigue concentration factor, K,, should be used

where

r = root radius and

a = material constant determined empirically. •

In applying Neuber's rule to the notch fatigue problem, Kf is used in

place of K,. By substituting the ratios of stress and strain ranges for K„
and /T,, Eq 1 becomes

(3)
\ ASAE /

Kf {LSbeE)"^ = (AoAeE)'
where E is the Young's modulus.
If strain, e, is elastic, then

KfbS = (La X ^x E)^'^ (4)

(KfLS)'
= AOAE (5)

For a given load and geometry, the left side of Eq 4 is a constant. The
set of solutions of Ao and Ac, therefore, form a hyperbola as shown in
Fig. 10. The intersection at A in Fig. 10 is the solution if the stress and
strain are entirely elastic. Points B and C are the location of true stress

TRUE STRAIN. IN./IN. (M/M|

FIG. \0—Neuter's hyperbola rule.

and strain for two different materials that have the same elastic modulus
but different strengths.
Crews and Hardrath [8\ experimentally demonstrated the validity of
using these control hyperbolas to determine the residual stresses and
strains after a complete cycle. This work was carried farther by Wetzel
[9] in demonstrating the use of control hyperbolas to determine actual
stress and strain throughout the load history.
Figure 11 shows how the actual stresses and strains may be followed
throughout the load history by returning to the control hyperbolas after
each load change. It may also be seen how the hysteresis loop stabilizes
and strain hardening has taken place for the particular material.
Manson and Halford [1\ developed an empirical relationship between

ORIGIN OF 2ND HYPERBOLA

1ST HYPERBOLA

3R0 HYPERBOLA

2ND HYPERBOLA'

lOTH CYCLE
60TH CYCLE
600TH CYCLE

FIG. n—Cyclic loop.

total strain and the number of cycles required to cause fatigue failure
by using conventional tensile properties, as follows

. , ^ UTS.,. + Dr'^-^Nf
(6)
Ac = 3.5 —^Nf

where
AE, = stabilized or final total true strain range,
UTS = ultimate tensile strength,
D, = tensile ductility, and
Nf= cycles to failure.

This formula has been shown to fit fatigue data for many materials. It
is adequate for predicting pure fatigue life where temperature and loading
frequencies are such that cyclic creep damage is not a factor. Many in-
vestigations have been performed using Eq 6 with acceptable correlation
between fatigue life (N/) and total strain (Ac,).
By using this formula, the value AE, can be obtained directly from the
plastic analysis without going through the Neuber's hyperbola approach.
In analyzing the HTTR, both Neuber's hyperbola and the plastic solution
were used. During the selective analysis, however, the analysis was limited
to the nominal or equivalent elastic stresses and the Neuber's hyperbola
approach.

Second Approach to Low-Cycle Fatigue

Creep and LCF interact. A common and somewhat simplified method
was used to determine this interaction. This was the life fractions method
that is based on the Miner's rule

(p = Cpc + CpLCF < 1 (7)

where cp is the life fraction and subscripts C and LCF correspond re-
spectively to creep and low-cycle fatigue
n m
qoc = J. t^(pc, cpLCF = Z ACPLCFM (8)
A = 1 Af = I

A ^* A ^^" /m
Acpc, = — AcpLCF„ = T T - (9)

where
hk = hours spent at condition k,
hi, = total hours possible at condition k,
CY„ = number of cycles used at condition M, and
Cy,„ = total number of cycles possible at condition M.

During the initial phases of the analysis, this method is sufficient for a
comparison of the design alternatives. As can be observed, the results are
the same if most of the cycles are spent at the beginning or at the end of
the creep life. This method does not consider the cycle frequency, nor
does it consider if the mechanical properties of the material related to the
creep can be affected by the cyclic operation.
In Manson and Halford's relationship [7], the number of cycles as
calculated by the Eq 6 was adjusted for creep phenomena. This was done
by using values related to the effective fraction of each cycle for which
the material may be considered to be subjected to the maximum stress,
frequency of stress application, and material coefficients.
In recent years, new techniques have been discussed; the most promising
among them is strainrange partitioning [4-6\. By this technique, any
cyclic behavior can be represented as a combination of four basic types
of inelastic strainranges defined as plastic-plastic, plastic-creep, creep-
creep, and creep-plastic. However, for a given cycle, not more than three
of these types can exist. Each one of these basic types must be tested for
a given material. To take full advantage of this new technique, considera-
ble effort must be expended to compile a library of materials for the
basic cycle types.

Completing the Analysis Process

In completing an analysis of this type, the following guidelines apply.
Step 11—If the projected plastic zone affects a considerable jjortion of
the model or a critical area, run a finite element-finite difference analysis
with a full elastic-plastic model and time increment solution.
Determine if there is any growth of the plastic area and establish the
characteristics of stress relaxation. Stress is important, but the significant
element in this analysis is strain.
Step 12—From the original elastic analysis related to cyclic operation,
calculate the number of cycles to failure by using nominal elastic stresses
and the Neuber hyperbolas.
Step 73—Interact creep and LCF by using Eqs 7-9. Define the best
design under the worst combination of creep and LCF. By using safety
factors Uke 6 to 10 for LCF and 2.5 to 4 for creep, the analytical design
process could stop at this point.
Step 14—If the design requires a more sophisticated approach because
of marginal conditions or large number of production units, strainrange
partitioning with better creep-LCF interaction could be required.

References
[/] Manson, S. S. and Halford, G., "A Method of Estimating High Temperature Low-
Cycle Fatigue Behavior of Materials," NASA Report TMX 52357, Lewis Research
Center, National Aeronautics and Space Administration, Cleveland, Ohio, Jan. 1967.
[2] Morrow, T., Wetzel, R. M., and Topper, T. H. in Effects of Environment and Complex
Load History on Fatigue Life, ASTM STP 462, American Society for Testing and
Materials, 1970, pp. 74-91.
[3] Morrow, T., Wetzel, R. M., and Topper, T. H., Journal of Materials, Vol. 4, No. 1,
March 1969, pp. 200-209.
[4] Spera, D. A. in Fatigue at Elevated Temperatures, ASTM STP 520, American Society
for Testing and Materials, 1973, pp. 648-657.
[5] Manson, S. S. in Fatigue at Elevated Temperatures, ASTM STP 520, American Society
for Testing and Materials, 1973, pp. 744-782.
[6\ Halford, G. R., Hirschberg, M. H., and Manson, S. S. in Fatigue at Elevated Temper-
atures, ASTM STP 520, American Society for Testing and Materials, 1973, pp. 658-669.
[7] Peterson, R. E., Material Research and Standards, Vol. 3, No. 2, Feb. 1963, pp. 122-
139.
[8] Crews, J. H. and Hardrath, H. F., Experimental Mechanics, Vol. 6, No. 6, June 1966,
pp. 313-320.
[9] Wetzel, R. M., Journal of Materials, Vol. 3, No. 3, Sept. 1968, pp. 646-657.

Thermal Fatigue and Its Failure

Prediction for Brittle Ceramics

REFERENCE: Hasselman, D. P. H., Badaliance, R., and Chen, E. P., "Thermal

Fatigue and Its Failure Prediction for Brittle Ceramics," Thermal Fatigue of Mate-
rials and Components, ASTM STP 612, D. A. Spera and D. F. Mowbray, Eds.,
American Society for Testing and Materials, 1976, pp. 55-68.

ABSTRACT: The thermal fatigue behavior of a soda-lime-silica glass subjected to

water quench and silicon-nitride subjected to thermal environment of a turbine
engine was predicted from data of slow (subcritical) crack growth. A numerical inte-
gration technique was developed to calculate the extent of slow crack growth for each
thermal cycle over the total duration of the transient thermal stress and temperature,
as well as the total number of cycles required for catastrophic failure to occur. Good
agreement between the predicted and experimental data was found. The results indi-
cate that, for reliable prediction of thermal fatigue resistance, an estimate of critical
flaw-depth based on a statistical (such as the WeibuU) theory of brittle fracture is
necessary.

KEY WORDS: thermal fatigue, crack propagation, silicon nitrides, silica glass, frac-
ture, fatigue failure.

Structural ceramics and glasses generally exhibit a decrease in load-

bearing ability (that is, strength) under conditions of constant [1,2]^ or
cyclic load [3,4] as well as repeated thermal shock [5.6]. Such behavior is
the result of slow crack growth at stress levels below those required for
rapid fracture. Considerable progress has been made in the failure predic-
tion of ceramic materials undergoing slow crack growth from information
for crack growth behavior. This approach was applied to conditions of
steady [7,8] as well as cyclic load [9] and single-cycle thermal shock [70].
This paper presents the results of a feasibiUty study to extend such tech-
niques to the prediction of thermal fatigue.

' Director, post-doctoral research associate, and assistant professor, respectively.

Ceramics Research Laboratory, Materials Research Center, Lehigh University, Bethlehem,
Pa. 18015.
'The italic numbers in brackets refer to the list of references appended to this paper.

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
56 THERMAL FATIGUE OF MATERIALS AND COMPONENTS

Materials

Soda-Lime-Silica Glass
Soda-lime-silica' glass in the form of rods with radius of 0.236 cm,
identical to the glass studied previously [10], was selected for the present
program. Table 1 lists the appropriate physical properties of this glass.
Experimental data for slow crack growth were obtained from the literature
TABLE l~Datafor physical properties, crack propagation behavior, and heat transfer required
for calculations of thermal fatigue resistance of soda-lime-silica glass rods subjected to a
water quench.

PHYSICAL PROPERTIES

Coefficient of thermal expansion (a) 9.3 x 10"' °C"'

Young's modulus of elasticity (£•) 6.9 x 10'°Nm"^
Poisson's ratio (v) 0.25
Thermal conductivity (A:) 2.5 x 10"'cal- °C"'cm''s''
Specific heat (c) 0.22 cal • g"' °Cr'
Density (e) 2.53 gem"""
WeibuU parameter (a) ~%

CRACK PROPAGATION BEHAVIOR

Fatigue limit (A",) 2.49 x I C N m " " '

Critical stress intensity factor (K\c) 7.49 x 10' N m " ' "
Preexponential factor (in Vo)
LoviK - 1.08 m-kg-s units (m/s)
High AT 10.3 m • kg-s units (m/s)
Velocity constant (b)
Low K 0.1888 m • kg • s units (m/s)
High A" 0.110mkgsunits(m/s)
Activation energy ( g ) 1.088 x 1 0 ' J m o P '
Initial crack depth (a a) 8 and 10 imy

HEAT TRANSFER DATA

Temperature of water bath 33 and 65 °C

Cylinder radius (/?) 0.236 cm
Heat transfer coefficient 0.1 cal • cm'^ "C"' s"'

[11]. For computational purposes, the crack growth behavior was

described by an exponential equation with two sets of constants for the
low and high values of the stress-intensity factor (ATi) expressed by

F = v„ txx>[(bK^ - Q)/RT] (1)

' R-6, Owens-Illinois.

where
V = rate of slow crack growth,
Vo = preexponential factor,
b = a constant,
Q = activation energy,
R = Boltzmann constant, and
T = absolute temperature.
Values for b and Vo corresponding to the high and low Ki are given in
Table 1. The experimental data for crack growth suggests that no crack
growth occurred below the fatigue limit (A"o).
Strength of the glass rods was measured at liquid nitrogen {N2) temper-
ature in four-point bending with a span between the central loading points
of 2.4 cm. This gave an average value of strength of 2.38 X 10* Nm'^
ranging from a minimum value of 1.94 X 10' to a maximum value of
2.76 X 10' Nm"^ for a total of ten specimens. A statistical analysis of the
data gave a value for the Weibull parameter, q ~ 8," also listed in Table 1.

Silicon Nitride
Thermal fatigue data for hot-pressed silicon nitride' were obtained by
Ammann et al [12] using a specimen geometry which approximated the
shape of a vane in a turbine engine. As shown by Evans et al [13], this
rate of slow crack growth in this material could be expressed by

V = AKrexp(-Q/RT) (2)
where
V = crack-velocity in ms"',
A, n = constants, and
Ki, Q. R, and J = as defined in Eq 1.
Table 2 lists the appropriate values for the constants in Eq 2, together
with the values for the physical properties pertinent to the thermal stress
calculations and fatigue predictions. Tensile strength at room temperature
as reported by Nessler [14] ranged from 2.41 to 3.71 x 10' Nm"^ with an
average value of 3.33 x 10' Nm"^

Thermal Environment
Soda-Lime-Silica Glass
The glass rods were subjected to thermal fatigue by repeated quenching
'This symbol corresponds to the letter m in the original Weibull theory.
'HS-130, Norton Company.

TABLE 2—Dataforphysical properties and crack propagation behavior required for

calculations of thermalfatigue resistance of hotpressed silicon nitride.

PHYSICAL PROPERTIES"

Coefficient of thermal expansion (o) 4.6 x 10"'°C"'

Young's modulus of elasticity (£•) 2.9 x 10" Nm"^
Poisson's ratio (v) 0.30
Thermal conductivity (AT) 0.033cal- °Cr' cm"' s '
Specific heat (c) 0.30calg"' °C"'
Density (e) 3.2ggm""'

CRACK PROPAGATION BEHAVIOR

Constant(y4) 5 x 10'''mkgsunits(m/s)
Constant (n) 6
Activation energy (Q) 170 kcal • mol"'
Critical stress intensity factor (A'lc)
Room temperature 4.7 MNm""^
1000°C S.OMNm-'"
" Representative values at 1000 °C, see text.

from an electrically heated laboratory oven into a water bath thermostat-

ically controlled at constant temperatures of 33 or 65 °C. The ends of the
specimens of length 7.5 cm were thermally insulated by wrapping with
glass insulating tape in order to eliminate fracture initiation at the speci-
men ends, leaving a gage section of 5.5 cm subjected to the thermal shock.
A total of nine specimens were tested simultaneously by holding them
lightly at the wrapped ends in a suitable specimen holder, attached to a
pneumatically operated air cylinder, controlled with an electric timer with
a frequency of four cycles/h. The specimens were kept in the oven for 12
min in order to assure thermal equilibrium prior to the quench. They were
then rapidly transferred (1 s) into the water bath where they were held for
approximately 2 min, after which they were slowly transferred back into
the oven in order to minimize thermal shock on heating. Upon emergence
from the water bath, the specimens were examined for fracture after each
cycle. Any broken specimen was removed. The number of cycles (N) re-
quired for fracture to occur was recorded for the first five specimens of
each batch. Due to the extreme scatter in data, TV for the lower value of
AT could not be determined for the sixth to ninth specimen within practi-
cal time limits.
Under the quenching conditions just described, the maximum value of
tensile thermal stresses occur in the surface during the cooling cycle. This
assures the maximum effect of the water environment on fatigue life. On
the heating part of the cycle, the thermal stresses in the surface are com-
pressive and are not expected to contribute to the fatigue.

Silicon Nitride
The thermal fatigue resistance of the silicon nitride, as determined by
Ammann et al [12], consisted of cycling the specimens between an elec-
trically heated laboratory oven and fluidized bed at 40°C. The specimen
was specifically selected such that, for the heat transfer characteristics of
the fluidized bed, the temperature and thermal stress history of the speci-
men was similar to that experienced by a vane in a turbine engine under-
going rapid shutdown. As calculated by finite element analysis, the vane,
originally at 1200°C and suddenly subjected to compressor air at 410°C,
experiences a transient thermal stress at the forward edge as shown in Fig.
la with a peak value of approximately 3.0 x 10' Nm"^, also included in
Fig. la. The transient temperature of the specimen at the forward edge
is shown in Fig. lb.

FATIGUE SPECIMEN

VANE: 1200 - 4 I 0 * C . i ENGINE

SPECIMEN: 1 3 0 0 - 40* d FLUIDIZED BED

TIME (SECONDS)

FIG. la—Transient thermal stress history of silicon nitride fatigue specimen quenched into
fluidized bed and vane in turbine engine subjected to rapid shutdown.

Prediction of Tiiermal Fatigue Life

The failure prediction of the thermal fatigue life followed the general
procedure and assumptions used by Badaliance et al [10] for calculations
of crack growth during a single-cycle quench, based on the numerical in-
tegration of the slow crack growth equation over the period of the ther-

iZ 500-

TIME (SECONDS)

FIG. \b—Transient temperature in forward edge of thermal fatigue specimen.

mal quench. The crack depth was assumed to be much smaller than the
specimen size, such that the crack configuration is closely approximated
by an edge crack in a semi-infinite solid. Also, it was assumed that the
presence of the crack did not affect the transient temperature field. Fi-
nally, in view of the shallow crack depth relative to the specimen size, it
was assumed that the crack "sees" a uniform temperature and stress field
calculated to exist at the surface of the rod. This latter assumption was
shown [15] to be valid for the ratios of flaw depth to specimen size of the
present study. For the glass rods, the temperature and stresses were cal-
culated directly from general equations given in the literature for infinitely
long rods of circular cross section. For the silicon nitride specimens, the
analytical expressions for the temperature and stresses in a cylindrical rod
were used by adjusting the appropriate parameters such that the actual
temperatures and stresses were closely matched to those shown in Fig. 1.
The use of analytical expressions for the transient temperatures and ther-
mal stresses considerably simplifies the calculations of slow crack growth
for the silicon nitride specimens.
The transient surface temperatures, T{R,t) of the rod under condition
of convective heat transfer is given by [16]

2pJ,(pn)
T{R, t) = To + (T, - To) I txp{-p„'xt/R') (3)
n^iim' + /3„Vo(^n)

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27 13:13:08 EST
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement.
HASSELMAN ET AL ON BRITTLE CERAMICS 61

where
R = cylinder radius,
To = temperature of quenching medium,
r, = initial uniform temperature of the cyUnder,
m = Biot's modulus with m = Rh/k,
h = heat transfer coefficient,
k - thermal conductivity,
K = therm2il diffusivity with JC = k/qc,
Q = density,
c = specific heat,
t = time,
Jo and /i = Bessel functions of the zero and first order,
^„ = the root of

mJoifin) - PnJ,(Pn) = 0, « = 1, 2, 3, . . . (3b)

From the distribution of the transient temperatures in the rod, the tran-
sient thermal stresses at the surface of the rod for the state of generalized
plane strain are [17]

OriR, 0 = 0

J.iP.) (,^
(1 - V) (. /.= ! (m^ + P^')Jo'Wn)

[2J,(li„) - p„Jo(Ji„)] exp(-^„^x///J^)|

where
E = Young's modulus of elasticity,
V = Poisson's ratio, and
a = coefficient of thermal expansion.
For the water quench, the heat transfer coefficient, h~0.lcal- cm"^• s"' °C"'
was estabUshed by Evems et al [75]. The corresponding value of the Biot
modulus for this value of h and rod radius of 0.236 cm is m ~ 10. For the
silicon nitride subjected to thermal fatigue in the fluidized bed, a peak
value of thermal stress of 3.0 x 10» Nrn'^ for a value of AT = 1260°C
corresponds to an effective Biot modulus, Weff - 0.65. Similarly, the time
dependence of the transient temperature and thermal stress corresponds to
that of a circular cylinder with effective radii of /?eff = 0.60 and 1.135 cm,
respectively. These values of ntetf and Rett are consistent with the average
specimen dimensions and the heat transfer coefficient of the fluidized bed
with/i = 850Wm-^°C-'.

The thermal stress-intensity factor, Ki, is related to the thermal stress

by [18]

Ki = 1.1215 (/?, O V ^ (5)

In general, the crack depth (a) at any time t can be obtained by integra-
tion of Eq 1 or 2, that is

ait) = ao + QJ Vdt (6)

where at, is the initial crack-depth at / = 0 at the beginning of a thermal

cycle. Substitution of Eq 4 into Eq 5, followed by substitution into Eq 1
or 2, with subsequent substitution into Eq 6 results in a nonlinear integral
expression for crack depth as a function of time within a given thermal
cycle. In order to overcome the mathematical complexity of the resulting
expression, a computer program was developed which numerically calcu-
lates the crack depth and stress-intensity factor as a function of time for a
given number of thermal cycles, as described in detail by Badaliance et al
[70]. Whenever the stress-intensity factor (^i) exceeds K\c, the specimen
is assumed to have failed. If failure did not occur within a given thermal
cycle, the procedure was repeated, and the number of cycles required to
meet the condition K > Kic was recorded as the thermal fatigue life. For
simplicity, the calculations assume that all pertinent properties were inde-
pendent of temperature. This assumption is quite reasonable for the tem-
perature ranges considered in this study for the glass and silicon nitride.
For the glass, the properties listed in Table 1 correspond approximately to
50 to 1(X)°C. For the silicon nitride, the properties listed in Table 2 cor-
respond to approximately 1000 °C, which is the estimated average temper-
ature at the time of maximum thermal stress in the fatigue experiments.
Preliminary results showed that, for good agreement with measured
fatigue life, predicted fatigue life had to be calculated on the basis of a
greater value of flaw depth than the value from the strength test. This is
not surprising since it is well known that the strength of brittle materials is
controlled by statistical variables and depends on specimen volume or sur-
face area as well as the stress distribution. Such effects were formulated
by Weibull [19] on the basis of the concept of the weakest link. For the
glass, a similar approach was followed in the present study using the as-
sumption that failure resulted from a surface flaw distribution. This re-
sulted in the value of the ratio of the average strength in the quench test
(o,) to the average strength in the bend test (oj)

a,/o, = 0.59 (7)

For o, = 2.38 X 10» Nm"^ and a value of Ki, = 8.7 x 10' Nm''''^ at
liquid Ni temperature as reported by Wiederhorn [20], an initial flaw
depth,flo~ 8.6 ^m can be calculated from Eq 5. This value of initial flaw
depth should correspond approximately to the flaw depth of the fifth
specimen out of nine to fail by thermal fatigue.
No such statistical analysis was carried out for the strength data of the
silicon nitride since the number of data points available was insufficient
to obtain a reliable value of the Weibull parameter q. For a critical stress
intensity factor, Kic = 4.7 MN-m"''^ [27] at room temperature, the strength
values reported earlier correspond to flaw depths ranging from about 40
to 95 nm, with a value of flaw depth of approximately 50 /im for the aver-
age value of strength.

Results and Discussion

The experimental and calculated fatigue data for the glass at the bath
temperatures of 33 and 65 °C are shown in Figs. 2a and 2b, respectively.
The fifth specimen to a first approximation represents the average spec-
imen of the nine tested for a given batch. Since the initial flaw depth of
the average specimen, oo = 8.6 p<m, the agreement between predicted
and experimental fatigue behavior must be considered excellent.
In comparing the thermal cycles to failure for the two bath tempera-
tures, it may be noted that the thermal fatigue life at a given AT, or the
AT for a given fatigue life, is less for the bath temperature of 65 than for
33 °C. The reason for this is that, although, for a given AT, the thermal
stresses are independent of bath temperature, the rate and total extent of
crack growth per cycle will be higher for the bath temperature of 65 °C
with its higher specimen temperature than for the bath temperature of
33 °C. It is for this reason also that the slope of the AT versus N curve for
a given bath temperature is controlled by the simultaneous effects of de-
creasing values of AT. It may also be noted that, at the bath temperature
of 33 °C, the scatter in experimental data for fatigue life at a given value
at AT is greater than for the bath temperature of 65 °C. This probably is
the result of the higher rates and extent of crack growth at the higher bath
temperature. Included in Figs. 2a and b are the values of the fatigue limits
(that is, the temperature differences) below which no thermal fatigue fail-
ure will occur. These fatigue limits correspond to those temperature dif-
ferences, and corresponding stress intensity factor (K\) is less than the
value of stress intensity factor (Ko, see Table 1) below which no crack
growth occurs. These fatigue limits are independent of temperature since
ATo was assumed independent of temperature, in view of the lack of fur-
ther information.

EXPERIMENTAL
FIRST SPECIMEN OF 9
FIFTH SPECIMEN OF 9
PREDICTED

FATIGUE QQ • 10/i.m
LIMITS
8/i.m lorc.
Oum 9I»C.

CYCLES

EXPERIMENTAL
BATH TEMP. I FIRST SPECIMEN OF 9
65»C. i FIFTH SPECIMEN OF 9
• PREDICTED

QQ ' 8/im

FIG. 2—Predicted and experimental number of cycles-to-failure of glass rods with initial
flaw-depth = 8.6 jjm subjected to thermal fatigue into water at (a) 33 °C and (b) 65 °C.

Figure 3 shows the predicted and observed thermal fatigue behavior of

the silicon nitride. In general, the range of flaw depth for the predicted
data covers the range of flaw depth of the specimens. However, quantita-
tive agreement with the predicted data occurs for values of initial flaw

10 100 1000
CYCLES TO FAILURE

FIG. 3—Experimental and predicted thermal fatigue behavior of silicon nitride subjected
to thermal cycling from high temperature into afluidized bed at 40°C.

depth much higher than the flaw depth corresponding to the average
strength of the set of specimens discussed earlier. Again, this discrepancy
is likely to be due to the statistical nature of brittle fracture referred to
earlier. In Fig. 3, as expected, the slope of the predicted curves corre-
sponding to the lower values of initial crack depth and higher tempera-
tures are steeper than those for the larger values of initial crack-depth.
Again, this is due to the simultaneous decrease in temperature as well as
level of thermal stress with decreasing temperature difference. For initial
crack depths of 80, 90, and 100 ptm, the temperature levels and thermal
stresses are sufficiently low such that the crack growth per cycle is less
than 10"' [oa. This implies that for these values of a^, the specimen stress
is just below the fracture stress, even at the first cycle. For these flaw
sizes, even a very small decrease in AT can make the specimens last almost
indefinitely. However, the low slopes of thermal fatigue curves imply
that, for accurate thermal fatigue predictions, thermal stresses, tempera-
tures, crack depths, and crack-growth behavior need to be established pre-
cisely.
The results for the glass and silicon nitride as just presented indicate
that reasonably accurate estimates of thermal fatigue resistance can be
made. However, the relative complexity of making such failure predic-
tions is clearly indicated from Table 1, which, including the initial temper-
ature of the specimen, lists a total of 17 values of physical properties,
crack propagation behavior, and thermal environment required for such

predictions. This number can be reduced to some extent by describing

crack growth behavior by Eq 2 rather than Eq 1. On the other hand, if
the temperature dependence of all the physical and environmental vari-
ables were included, the degree of complexity of the fatigue calculations
would increase manyfold. Also, the cumulative error as the result of the
uncertainty in the numerical value for each variable can cause large dis-
crepancies between predicted and observed fatigue life. Especially in view
of its semiempirical nature, estimating the proper flaw depth by use of the
Weibull theory introduces the largest uncertainty in estimating fatigue
Hfe.
In this respect, it should be noted that using the approach outlined in
this paper, differences in predicted life for different thermal environments
should be more reliable than predicted values of fatigue life for any given
simple condition. As a specific example, for the glass rods, for a given
fatigue life at low N, the observed decrease of approximately 7 °C for the
two bath temperatures is correctly predicted. For this reason, the compu-
tational technique developed for this study can be successfully coupled to
experimental fatigue studies and thereby increase the reliability of the
fatigue life predicted for some other condition. Such an approach be-
comes very useful in comparing calculated and observed fatigue life under
controlled conditions in order to predict fatigue life for a thermal envi-
ronment not easily duplicated in the laboratory. In this manner, for in-
stance, the initial slope of the thermal fatigue curve at small values of A'^
may be determined experimentally, from which the thermal fatigue be-
havior for higher values of N may be predicted by using the present tech-
nique. As a further example, for the silicon nitride, one can compute the
critical temperature differences (ATJ required for catastrophic failure in
one cycle for flaw sizes smaller than those considered in the present study.
Such results are shown in Fig. 4, which indicate that near the values of
ao = 10 m, AT,- is virtually independent of crack size. This suggests that
decreasing crack size to a value of 10 /im, which is expected to result in
large increases in strength at room temperature, nevertheless will not per-
mit major increases in engine operating temperatures. For comparison,
AJc in the absence of crack growth is shown in Fig. 4 by the dotted line. The
large differences in ^T^ especially at the lower value of Co, are the direct re-
sult of the slow crack growth. This observation suggests that a reduction
in the rate of crack growth in order to increase operating temperatures is
preferred to reduction in flaw size. At low engine temperatures at which
crack growth is negligible, of course, reductions in flaw size should prove
beneficial in that impact strength, etc., should be significantly increased.
Similar approaches become especially useful for full-scale engineering
applications on which actual experimental testing would be highly imprac-
tical. As a result, computer simulation of thermal fatigue in conjunction
with an experimental program can constitute a powerful technique in

1
«
2000

lOOC

6on
2 N

lb 90 too
^
CRACK DEPTH </im)

FIG. 4—Predicted critical temperature difference required to cause catastrophic failure in

silicon nitride as a function of flaw depth.

making reliable estimates of thermal fatigue resistance of brittle materials.

In summary, a potentially useful technique was developed to calculate
the thermal fatigue behavior of brittle materials and was applied success-
fully to predict the thermal fatigue resistance of soda-lime-silica glass and
siUcon nitride.

Acknowledgment
This study was conducted as part of a long-range research program on
thermal stress fracture of brittle materials supported by the Army Re-
search Office, Durham, under grant DA-ARO-A-31-124-73-G45.

References
[/] Williams, L. S., Transactions of the British Ceramic Society, Vol. 55, 1956, p. 287.
[2] Krohn, D. A. and Hasselman, D. P. H., Journal of the American Ceramic Society,
Vol. 55, 1972, p. 208.
[3] Gurney, C. and Pearson, S., Proceedings, Physics Society, London, Section B, Vol. 62,
1958, p. 537.
[4] Kossowsky, R., Journal of the American Ceramic Society, Vol. 56, 1973, p. 531.
[5] Sarkar, B. K. and Glinn, T. G. J., Transactions of the British Ceramic Society, Vol.
69, 1970, p. 199.
[6] Tumanov, V. I., Goldberg, Z. A., Chernyshev, V. V., and Pavlova, V. L, Porosh-
kovaya Metallurgiya, Vol. 10, 1966, p. 71.
[7] Fracture Mechanics of Ceramics, Vol. 2, Microstructure, Materials and Applications,
Bradt, R. C , Hasselman, D. P. H. and Lange, F. F., Eds., Plenum Press, New York,
1974.
[S] Evans, A. G., Journal of Materials Science, Vol. 7, 1972, p. 1137.
[9] Evans, A. G. and Fuller, E. R., Metallurgical Transactions, Vol. 5, 1974, p. 27.
[10] Badaliance, R., Krohn, D. A., and Hasselman, D. P. H., Journal of the American
Ceramic Society, Vol. 57, 1974, p. 432.
[U] Wiederhorn, S. M. and Bolz, L. H., Journal of the American Ceramic Society, Vol. 53,
1970, p. 543.
[12] Armann, C. L., Doherty, J. E., and Nessler, C. G., Journal of Materials Science, to
be published.
[13] Evans, A. G., Russell, L. R., and Richerson, D. W., Metallurgical Transactions, Vol.
6A, 1975, pp. 707-716.
[14] Nessler, C. G., Transactions of the American Society of Mechanical Engineers, to be
published.

[15] Evans, A. G., Linzar, M., Johnson, H., Hasselman, D. P. H., and Kipp, M. E., Jour-
nal of Materials Science, to be published.
[16] Carslaw, H. S. and Jaeger, J. C., Conduction of Heat in Solids, Clarendon Press, Ox-
ford, England, 1969.
[IT] Boley, B. A. and Weiner, G. H., Theory of Thermal Stresses, Wiley and Sons, New
York, 1960.
[18] Sih, G. C , Handbook of Stress Intensity Factors, Lehigh University, Bethlehem, Pa.,
1973.
[19] WeibuU, W.,Akad. Handl., No. 151, 1939.
[20] Wiederhorn, S. M., Journal of the American Ceramic Society, Vol. 52, 1969, p. 99.
[21] Wiederhorn, S. M. in Fracture Mechanics of Ceramics, Bradt, R. C , Hasselman, D. P.
H., and Lange, F. F., Eds., Plenum Press, New York, 1974.

Description of a Computerized
Method for Predicting Tliermai
Fatigue Life of Metals

REFERENCE: Spera, D. A. and Cox, E. C , "Description of a Computerized Method

for Predicting Thermal Fatigue Life of Metals," Thermal Fatigue of Materials and
Components, ASTM STP 612, D. A. Spera and D. F. Mowbray, Eds., American
Society for Testing and Materials, 1976, pp. 69-85.

ABSTRACT: A computer program called TFLIFE which can be used to predict the
thermal fatigue life of metals and structural components from conventional metal
properties is described. This program is used as a subroutine with a main program sup-
plied by the user. The main program calculates input cycles of temperature and total
strain for TFLIFE which then calculates a stress cycle, creep and plastic strain damage,
and cyclic life. A unique feature of TFLIFE is that it calculates lives according to
several different failure criteria for the same input data. These criteria are surface crack
initiation, interior crack initiation, and complete fracture of both unnotched and
notched fatigue specimens. Sample output tables are shown, together with results for
two typical problems: (a) thermal-mechanical fatigue of bar specimens of the tantalum
alloy T-111 eind (6) thermal-stress fatigue of wedge specimens of the nickel alloy
B-1900. Thermal fatigue lives calculated using TFLIFE have been verified by com-
parison with a variety of laboratory test data on different types of alloys. The com-
puter program is now ready for more extensive evaluation on structural components
as well as additional laboratory specimens.

KEY WORDS: thermal fatigue, fatigue (metals), crack initiation, computer programs,
thermal stress, life prediction, creep

This paper describes a fatigue life calculation system utilizing a computer

program called TFLIFE. TFLIFE calculates the low-cycle fatigue life of
metals subjected to simultaneous thermal and strain cycling. In conjunction
with an elastic structural analysis program, TFLIFE can be used to predict
the fatigue life of complete components in high-temperature equipment
such as engines and heat exchangers. This paper contains a general descrip-
tion of the program, typical output tables, and results for two sample
laboratory test problems.
' Consultant, Fatigue Branch, and mathematician. Computer Services Division, respectively,
Lewis Research Center, Cleveland, Ohio 44135.

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
70 THERMAL FATIGUE OF MATERIALS AND COMPONENTS

The original version of TFLIFE was written more than eight years ago to
aid in the development of a creep damage theory for thermal fatigue [1,2].^
Since that time, it has grown in scope and function, serving as a useful tool
for research on high-temperature metals [3]. It now includes recent develop-
ments in creep-fatigue theory and input/output formats which make it
suitable for general use. To date, TFLIFE results have been verified by ther-
mal fatigue tests on laboratory specimens such as bars, tapered disks,
wedges, and simulated turbine blades [J-4]. In these tests, thermal and
mechanical loads were applied by fatigue machines, fluidized beds, and
burner rigs to specimens of nickel, iron, cobalt, and tantalum alloys.
TFLIFE is now ready for more extensive evaluation by the technical com-
munity on complete components as well as laboratory specimens.
A unique feature of TFLIFE is that it calculates thermal fatigue lives ac-
cording to each of three failure criteria, based on one set of input data.
These criteria are (a) surface crack initiation, (b) interior crack initiation
(applicable to coated metals), and (c) complete fracture of thermal-
mechanical fatigue specimens, both unnotched and notched. Including
these three criteria in a single computer program produces a very general
and complete analysis of thermal fatigue resistance.
A complete set of the equations used in TFLIFE is given in Ref 5,
together with derivations. Computer time for a typical problem is 8 to 17
s on a time-sharing system (IBM 360/67). Required storage capacity is ap-
proximately 12 000 words.

General Description of TFLIFE

Figure 1 is a schematic diagram which describes the general sequence of
operations performed by the life calculation system. TFLIFE is a
subroutine which is called by a user-supplied main program. This main
program, discussed later, contains all calculations of temperature and total
strain specific to the specimen or component being analyzed. Initially, the
main program calls TFLIFE which reads its own input data (Block I) and
divides both the heating and cooling periods into approximately 100 in-
crements. The input data for TFLIFE are of two types: (a) conventional
mechanical properties which describe the monotonic tensile and creep-
rupture behavior of the metal and (b) optional empirical constants based on
cyclic data. The conventional metal properties must be given for a range of
temperatures which includes the thermal cycle being analyzed. The optional
empirical constants pertain to cyclic strain hardening, strain concentration,
and time-independent fatigue behavior. Nominal values for these constants
are contained in TFLIFE for cases in which cyclic data are not available. In-
put data can be in any of the following three systems of units for stress and

' The italic numbers in braclcets refer to the list of references appended to this paper.

TYPICAL
USER-SUPPLIED
MAIN PROGRAM

"TFLIFE" SUBROUTINE

START I. INPUT DATA

TENSILE
I I . TEMP & MECHANICAL CREEP-RUPTURE
pINITE-ELEMENT [-"-|- STRAIN CYCLES
ANALYSIS
SUBROUTINE_j-+- ELASTIC
L UNIAXIAL III. STRESS CYCLE
INELASTIC
STABILIZED

IV. FATIGUE DAMAGE

VARY PARAMETERS
PLASTIC STRAIN
GEOMETRY CREEP STRAIN
TEMP
STRAIN . LIFE PREDICTION
TIME, ETC CRACK INITIATION
(SURFACE, INTERIOR)
FRACTURE
COATING EFFECT
NOTCH EFFECT

V I . OUTPUT TABLES

FIG. I—Schematic diagram of computer program for predicting thermalfatigue life.

temperature: (a) pounds per square inch and Fahrenheit degrees, (d) grams
per square millimeter and Celsius degrees, or (c) newtons per square cen-
timeter and Kelvin degrees. SI (Systeme International) units can be specified
for the bulk of the output, irrespective of the input unit system. TFLIFE
then returns to the main program.

Temperature, Strain, and Stress Cycles

At Block II in the main program, cycles of temperature and mechanical
strain (total strain less free thermal expansion) must be calculated for a
potential failure location in the specimen or structure. The complexity
of these computations can vary greatly, depending on the structure being
analyzed. For a strain-cycled thermal-mechanical fatigue specimen. Block
II may be required only to generate saw-tooth or sinusoidal cycles. For a
thermal-stress fatigue specimen, it may be necessary to calculate transient

temperature distributions and the strains which result from them. In this
case, a linear elastic analysis is usually sufficient, according to the hy-
pothesis of total strain invariance proposed in Ref 6. This hypothesis
states that, under conditions of thermal stress, the displacements in a
body are substantially independent of local inelastic strains. Thus, dis-
placements obtained by a linear elastic analysis can be used to calculate
mechanical strains which can later be divided into elastic, plastic, and
creep components. For a complex component, it may be necessary for the
main program to call a finite-element analysis subroutine. In all cases, a
local temperature and a local uniaxial mechanical strain are calculated
for each time increment. The system then calls TFLIFE a second time.
In Block III, an incremental stress-strain analysis is performed which
divides the mechanical strain cycle into elastic, plastic, and creep com-
ponents. This analysis is repeated until shakedown occurs, when the stress-
strain hysteresis loop becomes stable and repetitive. In this way, a stress, a
plastic strain increment, and a creep strain increment are calculated for each
time increment. Approximately two thirds of TFLIFE's computing time is
used in these calculations. Upon completion of Block III, the mechanical
behavior of the material at the potential failure location has been described
thoroughly in terms of temperature, stress, and strain. Fatigue damage
can now be calculated.

Fatigue Damage and Life Predictions

In Block IV, TFLIFE calculates the fraction of the metal's life con-
sumed in one cycle. This calculation uses a fatigue damage model which
we call the generalized damage fraction theory [5]. The model is so named
because it integrates elements from the work of many investigators into
a unified, practical method of Ufe analysis. This theory uses Palmgren's
linear cumulative damage hypothesis [7] as modified by Taira to include
time as a variable [8]. Fatigue damage is assumed to be of two types:
(a) elastic-plastic strain damage which is independent of time and {b)
creep strain damage which is time-dependent.
Elastic-plastic strain damage is calculated using an empirical fatigue
equation of the exponential type, derived from equations proposed by
Basquin [P], Manson [10,11], and Coffin [12]. Equations for creep strain
damage combine contributions by Robinson [13], Hoffman [14], Taira [8],
Swindeman [75], and Spera [1-5].
In Block V, life predictions are now made for each of the following three
failure criteria: (a) surface crack initiation, (Jb) interior crack initiation, and
(c) complete fracture. The first criterion applies mainly to thermal-stress
fatigue of an uncoated material. The second criterion can be used to predict
the thermal-stress fatigue life of a coated material in which the fatigue life
of the coating exceeds that of the substrate material. The third criterion is

limited to thermal-mechanical fatigue testing in which strains and tem-

peratures are nominally uniform in the test section of the specimen. The
program provides for three strain concentration factors so that the user can
calculate the effect of notch severity on fatigue life. One of these factors is
always unity, representing the unnotched condition. The other two factors
may be specified arbitrarily by the user or nominal values of two and three
will be supplied by the program. For notched specimens, the input tem-
perature and mechanical strain cycles are considered to be nominal rather
than local. Thus, five cyclic lives are calculated in Block V for one pair of
temperature and mechanical strain cycles, as calculated in Block II. This
is a unique feature of the TFLIFE computer program.

Output Tables
In Block VI, the results of the life analysis are printed in the form of six
tables. Samples of some of these tables are shown in Fig. 2 for a thermal-
mechanical fatigue problem which will be discussed later. A table which is
not shown documents the applied cycle in terms of the mechanical
strainrange plus temperature, strain, and time limits. Figure 2a contains the
actual life predictions. These are presented for each of the three failure
criteria just discussed. This figure also contains damage fractions which
show the theoretical proportions of the damage associated with creep strain
and elastic-plastic strain for each predicted Ufe. The sum of the two damage
fractions is always unity. Also, interior crack initiation and complete frac-
ture are assumed to occur simultaneously in a thermal-mechanical fatigue
specimen because of the nominally uniform strain and temperature con-
ditions in the test section.
Figure 2b presents temperature, strain, and stress as functions of time
during heating and cooling. If the user wishes, this figure can be doubled or
tripled in length, resulting in smaller time increments and more detailed
data on the cycle. Figure 2c contains the monotonic tensile data for the
material as a function of temperature. This figure is constructed from the
actual input data using linear interpolation. The number, order, and
spacing of temperatures for which property data are given are arbitrary.
The cycUc hardening ratio in the footnote to Fig. 2c relates cycUc and
monotonic stress-strain data. This ratio is an optional scaling factor for
proportionately increasing or decreasing yield and ultimate tensile stress
values at all temperatures. It is specified by the user according to one of the
following three criteria: (a) If no cyclic stress-strain data are available, the
hardening ratio is unity, which represents neutral hardening. Also, unity is
the nominal value used by the program in the absence of a specified value.
(b) If partial cyclic stress-strain data are available (for example, at only one
temperature) the hardening ratio is adjusted until first-order differences
between calculated and observed stresses are eUminated. (c) If complete

1 I I I 1
1. FAILURE CRITERinN I STRAIN I CYCLES I DAMAGE FRACTIONS 1
1 1 rnNC 1 rn i . . 1
1 1 FACTOR 1 FAILURE I 1 1
1 I I I PLASTIC 1 CREEP 1

1 1 SURFACE 1 1.00 1 111 1 0.01(2 I 0.958 1

1 CRACK 1 1 1 1 1 1
1 INITIATION 1 1 1 1 1 1
1 1 INTERIOR 1 1.00 1 H 1 0.01.2 I 0.958 I

1 1 1.00 1 111 1 0.0lt2 1 0.958 1

1 COMPLETE FRACTURE 1 1 1 1 1

1 (THERMAL-MECHANICAL 1 2.15 1 36 1 0.1f,7 I 0.855 I

1 FATIGUE ONLY) 1 1 1 1 1
1 1 5.00 1 51 1 0.271 1 0.729 I

(a) Fatigue life and damage predictions.

FIG. 2—Sample output tables (in-phase thermal-mechanical fatigue of the tantalum alloy T-
111).

I
1 TIME, TEMP, 1 STRAIN I STRESS,
.1
1 MIN DEG r. 1 MECH. I PLASTIC j CREEP I M/SQ CM
J I 1.
1 o.ono Ii78 1 I I I
1 0.039 506 1 -.009570 I -.000000 I 0.000000 I -56118
1 0.078 f 555 1 -.008622 I -.00000"
1 0.117 565 1 -.007662 I - '

.^o3 I 0 . 0 0 2 2 5 5 I Ii2962
.^ I 0 . 0 0 9 3 6 5 I 0 . 0 0 2 7 9 2 I UOSkC
I u.0087110 I 0 . 0 0 9 3 6 5 I 0.0OJ317 | 3868li
I 1.290 i lli21 I 0.009138 I 0.009565 I 0.005828 I 36739
I I I I I I

(b) Thermal, strain, and stress cycles, heating half cycle.

FIG. 2—Continued.

1 I
TEMP, I YOUNG'S YIELD ULTIMATE TENSILE POISSON'SI THERMAL I
I MODULUS, STRESS TENSILE DUCTILITYI RATIO I EXPANSIONI
(.002), STRESS, I FROM RT,|
DEG K I N/SQ CM N/SQ CM N/SQ CM I PER DEO K|
.1 I
I I
200 I 18.7E 06 56000 65500 0.800 I 6.05E-06 I
269 I 18.3E 06 51000 60100 1 nnn
339 I IS.IE 06 li8075
1(08 I 17.9E 06

i.lbO 0.280 I 7.09E-06 I

iHOSO 1 1.255 0.290 I 7.1IIE-06 I
1 AMiOO
17800 1 1.350 0.300 I 7.19E-06 I
I 1799 I 1 5 . 2 E 06 1 1 12000
15000 1 l.liSO 0.516 I 7.25E-06 I
1 1 1
1 RATIO OF CYCLIC TO MONOTONIC YIELD AND ULTIMATE STRESSES > l.kl

(c) Monotonic stress and strain properties.

FIG. 2—Continued.

cyclic Stress-Strain data are available, they are used to calculate equivalent
monotonic data for a cyclic hardening ratio of unity.
Figure 2d is & summary of the empirical equations which are used in
TFLIFE to calculate creep and rupture times [7]. The coefficients are
calculated from test data by means of regression analysis. Most of the com-
monly used time-temperature parameters are included in these general
equations as special cases, either exactly or to a close approximation. In this
way, the equations in Fig. 2d represent an attempt to select the optimum

2 1
1 LOG(Z) - A • B«LOG(X) • C«X • D»X • E«Y Z - T l » TR I
1 T l • TIME TO 1 PCT CREEP, HR TR - TIME TO RUPTURE, HR |

-1 1
1 CREEP RATE - 0 . 0 1 / T l , HR X - STRESS/1000 N/SQ CM I
BETA
1 Y • (TEMP/1000 K) OR LOG(TEMP/1000 K ) , IF BETA - 0 |

1 1 1 1 1 1 1
1 Z 1 A 1 8 1 C 1 D 1 E 1 BETA 1
1 1 1 1 1 1 1
1 1 t i l l !
1 Tl 1 - 9 . 1 9 8 1 -1I.BB70 1 0.00000 I-.001569 I 23.791 I -1.00 j

1 TR 1 -7.729 1 -11.6670 1 0.00000 |-.00156!> | 23.1i78 I -1.00 |

1 1 1 1 1

(d) Monotonic creep-rupture behavior.

FIG. 2—Continued.

time-temperature parameter for the available data. If only limited creep rate
data are available, simplifying assumptions can be made by the user so that
some of the curve-fit constants in the equation for time to 1 percent creep
are the same as those in the equation for time to rupture.
A final table (not shown) contains data which document the iterations
required to achieve convergence. Each iteration is a new calculation of the
stress and inelastic strain cycles. The stress at the end of the cooling cycle is
carried forward to start the incremental stress analysis for the next iteration.
A residual inelastic strain is recorded which is the sum of the plastic and
creep strains at the end of each iteration. (Both these quantities are set to
zero at the start of each iteration.) A residual inelastic strain of zero
represents a completely stable, repetitive stress-strain hysteresis loop. When
the residual strain becomes less than ten microunits, a curve-fit procedure is
used to accelerate convergence, and iteration is terminated after two ad-
ditional cycles.

Options
After output tables have been printed for one set of input data, there is a
minimum of three options which the user can choose to include in the main

program: (a) terminating the program, (b) starting the solution to a new
problem by reading a new set of data, or (i) varying parameters in the input
data which are already stored in the computer and starting the solution
to a modified problem. The third option enables the user to perform a
parametric study from a single set of input data. As mentioned pre-
viously, the operations in the main program are determined by the needs
of the user. The TFLIFE system, shown schematically in Fig. 1, includes
sufficient flexibility to satsify a wide variety of such needs.

Sample Results
The application of TFLIFE to different types of thermal fatigue con-
ditions will now be illustrated by showing some typical results for the
following two problems: (a) thermal-mechanical fatigue of bar specimens of
the tantalum-base alloy T-111 tested in vacuum [76] and (b) thermal-stress
fatigue of wedge specimens of the nickel-base alloy B-1900 tested in
fluidized beds [17,18]. Figure 3 shows the geometry of these two types of
specimens. The slot notch shown in Fig. 3b has a theoretical elastic stress (or
strain) concentration factor of 3.0 [19]. Predicted lives are given for un-
notched and notched bars of T-111 as a function of imposed strain range

TRW VACUUM FACILITY

GAS REUEF SLOT

a 64 CM

5/8"-18

(a) Thennal-mechanical fatigue bar specimen [16\.

FIG. 3—Geometry of specimens in sample problems.

combined with either isothermal or in-phase thermal cycling. The B-1900

wedge specimen calculations include a study of the effects of changes in

a 02 CM RAD
a 25 CM--

(b) Detail of tangential slot notch.

FIG. 3—Continued.

raiOCM R
r0.07 CM R 1

10 CM

J
C,^

(c) Thermal-stress fatigue wedge specimen [17, IS].

FIG. 3—Continued.

cycle time and edge radius for both the uncoated and aluminide-coated con-
ditions of the alloy.

Thermal-Mechanical Fatigue of T-111 Bars

T-111 was selected as a sample alloy because of the variety of thermal-
mechanical fatigue data reported on it in Ref 16. Also, its stress-strain
behavior is as complex as any which the user is likely to encounter.
Therefore, T-Ill provides an excellent test of the capabilities of TFLIFE
with respect to thermal-mechanical fatigue. The output tables presented in
Fig. 2 are taken from one of the T-111 test cases.
Figure 4 shows the simple type of mechanical strain and temperature
cycles programmed for these tests. When diametral strain is the control
variable, the normal input variable of axial strain is usually unknown prior
rIN-PHASE THERMAL CYCLE
1
1 rlS(^THERMAL
I
t
I / 1500
1—». /\ 1
v
.010 1000 ^,y,p_

/
// 'A'l\ T,
/ If \
.005 / If
tl '
500 ^
If A
MECHANICAL rZ^J
if
STRAIN, 0 il \\\\ 0
Em
il
// \\u
II ^2 )C DIAMETRAL STRAIN (INPUT)
1
-.005 "7
V
\\ ^ A X I A L STRAIN (OUTPUT)

-.010 1 1 ^>
0 1 2 3
ELAPSED TIME, t, MIN

FIG. ^—Typical thermal and strain cycles applied during thermal-mechanicalfatigue testing
of the tantalum-base alloy T-111.

to the stress analysis (Fig. 1, Block III). TFLIFE resolves this difficulty by
using an equivalent axial strain equal to twice the diametral strain plus an
equivalent elastic modulus equal to Young's modulus divided by twice
Poisson's ratio [5].
Figure 5 illustrates the stress-strain behavior calculated for both the
isothermal and in-phase thermal cycles with equal strain ranges. The
significantly different shapes of these two calculated hysteresis loops are
confirmed by test data [16\. Cyclic hardening ratios were selected according
to the criteria described previously for cases with partial cyclic stress-strain
data (see output tables). Two different cyclic hardening ratios were required

60X10^
1422" K7
IN-PHASE THERMAL
CYCLE
ISOTHERMAL CYCLE,
1422° K

TRUE STRESS,

-.010 -.005 0 .005 .010

MECHANICAL STRAIN, t^

FIG. 5—Typical calculated stress-strain behavior during thermal-mechanical fatigue testing

of the alloy T-111.

for these two cycles because the stresses in the in-phase thermal cycle were
found to be significantly larger than those in the isothermal cycle for
equal strain ranges. A hardening ratio of 1.43 was needed to correlate cal-
culated and observed stresses at Point A in the in-phase loop. A much
smaller factor, 1.08, matched the calculated stress amplitude in the iso-
thermal loop (Point B) to the measured value. These hardening ratios
reflect the complex cyclic hardening which has been observed in T-111.
For behavior of this type, an accurate estimate of the hardening ratio
is essential to accuracy in the prediction of creep damage and life.
Figure 6 shows the results of the life calculations for isothermal cycling as
a function of axial mechanical strain range. The failure criterion for these
tests was complete fracture of the bar specimen. Predicted behavior with a
strain concentration factor of unity is in good agreement with the un-
notched specimen data, as shown by the open symbols. This correlation be-
tween theory and experiment gives a measure of confidence in the ex-
trapolated behavior for lives greater than 1000 cycles. For the notched
specimens (closed symbols), predicted behavior is somewhat conservative
when the strain concentration factor is equated to its theoretical elastic
value of 3.00. The available data were best fit using an empirical strain con-
centration factor of 2.15. Life predictions based on this factor are given by
the dashed line in Fig. 6. In the absence of notched specimen data, the
theoretical elastic strain concentration factor would be used as input in
TFLIFE.
In Fig. 7, predicted and observed lives are compared for tests with in-
phase thermal cycling. Unlike the isothermal tests, the in-phase tests of T-
111 showed little notch effect. The TFLIFE predictions are in agreement

1422° K, 0.0067 Hz

o UNNOTCHEDl SHEFFl£R &

• NOTCHED J DOBLE, 72
Z PREDICTED USING TFLIFE

MECHANICAL
STRAIN
RANGE,
STRAIN CONCENTRATION
FACTOR
,>-1.00(UNNOTCHED)
2.15 (EMPIRICAL)
3.00 (THEORETICAL)

.001
10 10^ 10^ 10* 10^
CYCLES TO FRACTURE, N,

FIG. 6—Comparison of predicted and observed fatigue lives for isothermal cycling of the
alloy T-Ul in the unnotchedand notched conditions (1422 K, 0.0067Hz).

IN-PHASE CYCLING, 478° TO 1422° K, 0.0067 Hz

• UNNOTCHEDl SHEFFLER &

• NOTCHED J DOBLE, 72
m z : PREDICTED USING TFLIFE

MECHANICAL
STRAIN
RANGE. ^ ^
t£_ F ^ ^ 1 ^ STRAIN CONCENTRATION
"• FACTOR
LOO (UNNOTCHED)

\ 2.15 (EMPIRICAL)
3.00 (THEORETICAL)
.001
10^ KP 10*
CYCLES TO FRACTURE, Nf

FIG. 1—Comparison ofpredicted and observed fatigue lives for in-phase thermal cycling of
the alloy T-Ul in the unnotched and notched conditions (478 to 1422 K, 0.0067Hz).

with the data, whether the strain concentration factor is assumed to be 1.00,
2.15, or 3.00. This insensitivity to strain concentration is related directly to
an assumption in the creep damage model in TFLIFE. The assumption is
that interior creep damage is not affected by a surface strain concentration
but depends only on the nominal stress. Thus, the strain concentration af-
fects only the elastic-plastic strain component of damage. During in-phase

thermal cycling at relatively large strain ranges (lives less than 1000 cycles)
most of the damage results from creep (Fig. 2a). Therefore, the model
indicates that strain concentration has only a minor effect on the predic-
ted lives at these strain ranges for in-phase cycling. However, as the strain
range decreases below about 0.004, elastic-plastic strain damage becomes
dominant over creep damage and a significant notch effect is predicted
for this alloy. Figure 7 is an example of an extrapolation of data by
TFLIFE that indicates a notch effect at long life which is not present in
the short-life data.

Thermal-Stress Fatigue ofB-1900 Wedges

Rapid heating and cooling of the wedge specimens shown in Fig. 3
produces edge cracks similar to those often found in turbine blades and
vanes. The sequence of calculations required for predicting the initiation
period for such cracks is illustrated by the results shown in Figs. 8-11. In
contrast to the simple cycles shown in Fig. 4, the transient temperature and

1400 . CHANGE DURING

TRANSFER-!

1200

EDGE TEMP,
HEATING IN 1360° K BED
T, ICOO
— COOLING IN 590P K BED

800

600
1 2
J m3
ELAPSED TIME, t, MIN

FIG. 8—Typical temperature transients at the edge of wedge specimens of the alloy B-1900
during thermal-stress fatigue testing.

Strain inputs for thermal-stress fatigue problems are quite complex. Con-
sequently, Block II in the main program (Fig. 1) is much more complex for
thermal-stress fatigue than for thermal-mechanical fatigue problems. For
instance, in this B-1900 problem, thermocouple data taken at several points
in the interior of the wedge specimen are curve-fit in Block II for each of the
approximately 100 time increments during heating and the same number of
time increments during heating and the same number of time increments

COOLING IN
590" K BED

MECHANICAL STRAIN,

-.002

-.004
600 800 1000 1200 1400
TEMPERATURE, T, °K

FIG. 9—Typical calculated strain-temperature behavior during thermal-stress fatigue testing

of the alloy B-1900.

80b(lO^
-950PK

COOUNG IN 59CP K BED

TRUE STRESS, 20 YR^NSFEI
SHAKEDOWN
N/CM^

HEATING IN 1360° K BED

-.004-.002 002 .004

MECHANICAL STRAIN, e,
m

FIG. 10—Typical calculated stress-strain behavior during thermal-stress fatigue testing of

the alloy B-1900.

during cooling, as described in Ref 18. These curve-fits are then used to
calculate the temperature and mechanical strain cycles at the specimen
edges.
Figure 8 shows typical edge temperature transients during heating and
cooling. Note that the length of the heating and cooling periods is not suf-
ficient for the temperatures to reach steady-state levels equal to the bed tem-

590P TO 13MP K
EDGE RADIUS,
r,
MM
0.6
HEATING LO
& PREDICTIONS USING
COOLING TFLIFE
TIMES,
MIN

10 10^ 10?
CYCLES TO CRACK INITIATION

590P TO 1360P K
lOr—
I MM
— EDGE
RADIUS,
HEATING & r, \ OND-
COOUNG TIMES, — MM \ o ^
MIN O a6 CJ [ J t
— D LO
- PREOICTia
TFLIFE
1 1 \ 1
10 10^ lo' 10*
CYCLES TO CRACK INITIATION

(a) Uncoated condition.

{b) Aluminide-coated condition.
FIG. 11—Comparison ofpredicted and observed thermal-stress fatigue lives for the alloy B-
1900 (wedge specimens tested influidized beds at S90 and 1360 K; data from RefU.

peratures. In Fig. 9, transient mechanical strain is plotted versus edge tem-

perature rather than versus time. This is often a convenient way to display
transient data and to ensure that calculated strain and temperature cycles
are compatible. Changes during one of the transfer periods are also noted in
Figs. 8 and 9 when the specimen is moved from the cooling to the heating
bed.
A typical stress-strain hysteresis loop for thermal-stress fatigue is presen-
ted in Fig. 10. A cyclic hardening ratio of unity (that is, neutral hardening)
was used to calculate this loop in the absence of pertinent cyclic stress-strain

data. Comparison of Figs. 5 and 10 shows the wide variety of stress-strain

behavior possible during thermal fatigue. The hysteresis loop in Fig. 10 has
a figure-eight shape which is produced by changes in elastic modulus with
temperature combined with the presence of only small amounts of inelastic
strain. The shakedown process during iteration is shown to produce an in-
crease in this cycle's mean stress.
Figure 11 summarizes the dependence of the number of cycles required to
initiate edge cracks on the parameters of cycle time and edge radius, for un-
coated and aluminide-coated conditions. Comparison of predicted curves
with test data indicate that TFLIFE can correctly model thermal-stress
fatigue life on the basis of thermocouple data and monotonic material
properties. Furthermore, TFLIFE can be used to predict behavior outside
the range of the test parameters.

Conclusions
The TFLIFE computer program has been found to be a useful tool for
analyzing thermal fatigue data in terms of conventional metal properties
and for predicting the thermal fatigue life of metals. It contains models for
the major damaging phenomena which occur during a general thermal-
fatigue cycle, at least to a first-order approximation. Thermal fatigue lives
calculated using TFLIFE have been verified by comparison with a variety of
laboratory test data on different types of alloys. This computer program is
ready for more extensive evaluation by the technical community, not only
on additional laboratory specimens but also on structural components.

References
[1] Spera, D. A., "A Linear Creep Damage Theory for Thermal Fatigue of Materials,"
Ph.D. thesis, University of Wisconsin, Madison, 1968.
[2] Spera, D. A., "Calculation of Thermal-Fatigue Life Based on Accumulated Creep
Damage," NASA TN D-5489, National Aeronautics and Space Administration, Oct.
1969.
[3] Spera, D. A. and Grisaffe, S. J., "Life Prediction of Turbine Components: On-going
Studies at the NASA Lewis Research Center," NASA TM X-2664, National Aeronautics
and Space Administration, Jan. 1973.
[4] Spera, D. A. in Fatigue at Elevated Temperatures, ASTMSTPS20. American Society for
Testing and Materials, 1973, pp. 648-657.
[5] Spera, D. A., "Equations for Predicting Thermal Fatigue Life," NASA TM (to be
published). National Aeronautics and Space Administration.
[(J] Mendelson, A. and Manson, S. S., "Practical Solution of Plastic Deformation Problems
in Elastic-Plastic Range," NACA TN 4088, National Advisory Committee for
Aeronautics, Sept. 1957.
[7] Palmgren, A., Zeitschrift des VereinesDeutscherIngenieure, Vol. 68, No. 14, April 1924,
pp. 339-341. (English translation, NASA TT F-13460, National Aeronautics and Space
Administration, March 1971.)
[8] Taira, S. in Creep in Structures, N. J. Hoff, Ed., Springer Verlag, Berlin, 1962, pp.
96-124.

[P] Basquin, O. H., Proceedings, American Society for Testing and Materials, Vol. 11,1910,
pp. 625,630.
[10] Manson, S. S., "Behavior of Materials under Conditions of Thermal Stress," NACA
TN-2933, National Advisory Committee for Aeronautics, 1953.
[U] Manson, S. S., Experimental Mechanics, Vol. 5, No. 7, July 1965, pp. 193-226.
[12] Coffin, L. F., Jr., Transactions, American Society of Mechanical Engineers, Vol. 76,
Aug. 1954, pp. 931-950.
[13] Robinson, E. L., Transactions, American Society of Mechanical Engineers, Vol. 74,
1952, pp. 777-781.
[14] Hoffman, C. A., Proceedings, American Society for Testing and Materials, Vol. 56,
1956, pp. 1063-1080.
[15] Swindeman, R. V/., Proceedings, Joint International Conference on Creep, Institution of
Mechanical Engineers, London, 1963, pp. 3-71 to 3-76.
[16] Sheffler, K. D. and Doble, G. S., "Influence of Creep Damage on the Low Cycle Ther-
mal-Mechanical Fatigue Behavior of Two Tantalum-Base Alloys," NASA CR-121001,
National Aeronautics and Space Administration, May 1972.
[17] Howes, M. A. H., "Thermal Fatigue Data on 15 Nickel- and Cobalt-Base Alloys,"
NASA CR-72738, National Aeronautics and Space Administration, May 1970.
[18] Spera, D. A., Howes, M. A. H., and Bizon, P. T., "Thermal-Fatigue Resistance of 15
High-Temperature Alloys Determined by the Fluidized-Bed Technique," NASA TM X-
52975, National Aeronautics and Space Administration, March 1971.
[19] Manson, S. S. and Hirschberg, M. H., "Low Cycle Fatigue of Notched Specimens by
Consideration of Crack Initiation and Propagation," NASA TN D-3146, National
Aeronautics and Space Administration, June 1967.

A Study of Thermal
Fatigue Mechanisms

REFERENCE: Howes, M. A. H., "A Study of Thermal Fatigue Mechanisms," Ther-

mal Fatigue of Materials and Components, ASTM STP 612, D. A. Spera and D. F.
Mowbray, Eds., American Society for Testing and Materials, 1976, pp. 86-105.

ABSTRACT: The thermal fatigue resistance of materials has been evaluated in a

series of investigations using the fluidized bed method of thermal cycling. The factors
that influence thermal cracking were studied for many materials and definite patterns
of behavior observed. Particular attention was paid to H13 tool steel and an 18Ni
maraging alloy as these represent materials with distinctly different mechanisms of
heat treatment response.

KEY WORDS: thermal fatigue, fluidized bed processing, thermal cycling tests, nickel
alloys, steel, crack propagation, plastic deformation

At the present time, thermal fatigue testing is not standardized to the

same degree as tension testing, creep testing, or fatigue testing, and it is
difficult to compare thermal fatigue results by different investigations.
Part of the difficulty results from the fact that a temperature cycle may
not produce the same stress cycle because of specimen geometric differ-
ences and heat input variations due to the testing method. It would be
convenient, therefore, to predict thermal fatigue behavior from the results
obtained by more standardized tests. One problem is that thermal fatigue
involves mechanisms that do not appear to the same extent as in other
testing methods. For example, cyclic oxidation often occurs during ther-
mal fatigue testing, but continuous oxidation occurs in creep tests. It is
believed that a better understanding of thermal fatigue failure mecha-
nisms would improve the accuracy of predictive equations and make them
more applicable to a wide range of materials.
This investigation used the fluidized bed method of thermal cycling.
The objective was to improve the understanding of the thermal fatigue
mechanism by control of the test environment and analysis of the test re-
sults. The experimental work has been carried out on H13 and 18Ni mar-
' Director, Metals Division, Illinois Institute of Technology Research, Chicago, 111. 60616.

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
HOWES ON THERMAL FATIGUE MECHANISMS 87

aging alloy, but examples of microstructures obtained in other investiga-

tions are also referred to.

Test Procedure
The equipment used for thermal fatigue evaluation consists of two flu-
idized beds, one heated and one cooled. The specimens are cycled between
the two beds at preset intervals. A full description of the equipment is
contained in an earUer paper [1].^
The specimens are disk-type configurations 16.35 cm diameter and 1.27
cm thick. The fin section is approximately 3 mm thick. These specimens
are used either unnotched or notched [1]. A notched specimen has two V-
notches of the same dimensions as the notch in a Type A Charpy impact
bar at 180 deg. The stress intensification factor of this type of notch is ap-
proximately 2.8.
Thermal fatigue testing often requires attempting the duplication of
both the stress and thermal cycles that occur in an actual component situ-
ation. In this investigation, it was desired to find the range of stress vari-
ation that could be produced with the same thermal excursion limits. This
would be done by varying the fluid bed test conditions, making accurate
temperature measurements throughout the thermal fatigue specimen test
section, and using a computer program to calculate the stress gradients.
The stress levels would then be compared with the thermal fatigue results
to determine if the stress level to which the specimens are subjected causes
cracking. Test results from other programs would also be examined where
mechanistic effects had occurred.
The specimens used for the stress det3rminations were made from two
materials chosen to be representative of steels requiring a hardening and
tempering treatment and materials requiring solution treatment and aging.
The compositions are shown in Table 1.
The materials were processed in the following way: the starting point
for each material was a 15-cm-square billet. From this billet, transverse
samples were removed at or near the midradius position for tension test-
ing. A portion of the billet was press forged to a 7-cm-round bar and used
for thermal fatigue specimens. The remainder of the 15-cm billet was re-
duced by forging and rolling to a 2.5-cm-diameter round bar for use in
creep, strain tempering, and static tempering tests. In this way, all mate-
rial of one composition came from the same billet, ingot, and heat.
The mechanical properties of both materials were measured up to 700 °C
and are shown in Figs. 1 and 2. Creep data are given in Fig. 3 and 4. Tem-
pering and aging curves are shown in Figs. 5 and 6.
The specimens were all thermally cycled in a manner that caused the

^The italic numbers in brackets refer to the list of references appended to this paper.

TABLE 1—Materials used for stress determinations.

Analysis, %

Description H13 18Ni

C 0.399 0.018
Mn 0.33 <0.I0
P 0.018 <0.01
S 0.003 0.007
Si 0.93 <0.01
Cr 5.27 ND
Mo 1.13 5.10
V 1.00 ND
Ni 0.17 18.2
W <0.01 ND
Al ND° 0.10
Ti ND 0.65
Co ND 8.8
Zr ND <0.01
B ND 0.003
Melting conditions air melted induction vacuum melted
and vacuum arc melted

" ND = not determined.

outer periphery of the fin to fluctuate between 593 and 204 °C. However,
the temperature excursion was produced in two different ways.
1. A severe heating cycle was produced by immersing the specimens in
a hot bed at 1038 °C and removing them when the outer edge of the fin
had reached 593 °C.
2. A mild heating cycle was produced by immersing the specimen in the
hot bed at 649 °C and removing them when the outer edge of the fin
reached 593 °C. In addition, a lower degree of bed fluidization was used.
The thermal fatigue test conditions are given in Table 2. It is seen that
there is a considerable difference in total cycle time between the two cy-
cles.

Temperature Calibration and Calculation of Stresses

Temperature profiles were established using fin specimens equipped
with ten thermocouples. Calibration was carried out on specimens that
had been preoxidized since the oxide film would affect heat transfer
characteristics. The sheathed chromel-alumel thermocouples used for the
curves shown in Figs. 7 and 8 and H13 were approximately 0.025 mm
diameter to ensure a rapid response to temperature change. Figure 7
shows the severe cycle where very large gradients are produced. Figure 8

100

200
<
50 S.
IS

150 -S
25 O

2000
100

1500

£
8
£ 1000

500

aoSED POINTS INDICATE RESULTS ^ ^

OBTAINED AFTER HOLDING 2 0 0 HOURS ^i
AT TEMPERATURE BEFORE TESTING.

- j _ - i -
100 200 300 400 500 600 700
Test Temperature , ' C

FIG. 1—Mechanical properties ofHlS tool steel.

100 -1250

75 I- /
(9
S* 50 -o"-2?
<
0.«lt O / 1
150 S
25

RA.

2000 100
-'^%

1500

1000

500

_1 I L.
0 100 200 300 400 500 600 700
Test Temperature , 'C

FIG. 2—Mechanical properties of 18Ni maraging alloy.

1000
TIME, HOURS

FIG. 3—H13 creep rupture data.

10 100 1000
TIME, HOURS

FIG. 4—18Ni maraging alloy creep rupture data.

H 13
45

o
-J 4 0
UJ
538*C
u
o
"^35

§30 \
< 566 °C
^ 593° C

25 -

20 1 1 1 1 1 - 1
3 4 5

TIME, HOURS X 100

FIG. 5—Tempering of HIS.

I8NI
55

o
^ ^
ll 50
UJ
^ - - ^ ^ ~ - _ sse'c
0= 4 5 ^ ^ - - ~ - 566°C
tf)
^ V.^
UJ ^*-~ 593° C

§40
<
I

20
3 4 5
TIME, HOURS X 100

FIG. 6—Aging of 18Ni maraging alloy.

TABLE 2—Thermal fatigue test conditions.

Air Flow Measured

Temperature, at Bed Temperature,
Cycle Material Bed Time, s mVmVh

Mild 18Ni heating 649 156 600

cooling 43 85 1200
H13 heating 649 178 600
cooling 43 86 1200
Severe 18Ni heating 1038 VA 750
cooling 43 22 1500
H13 heating 1038 4 750
cooling 43 24 1500

LOCATION OF
THERMOCOUPLES
IN TEST SPECIMEN

TIME. SECONDS

FIG. 7—Heating and cooling of HI 3 thermal fatigue specimen fin during the severe cycle.

shows the mild cycle where temperature changes are much less severe. In
many cases, the temperature changes at the top edge of the fin are identi-
cal to changes at the bottom edge.
The temperature data are not of prime interest in themselves but are
obtained in order to predict stress levels in the specimen. This can be done
using finite element technique to calculate both thermal and stress gradi-
ents throughout the specimen.

N
CM

o
*
<M

•si
ti
S'
o §
CM
s
«j

•s:
o«
O •^
CO 3
t3
— .C
-^
« S
u o

o
UJ
?

i
i?
Si
en 4
c:
O lU •^
CM •s •Si
H

8
e

f
o

3 930 '3aniVy3dW31

Figure 9 shows the mesh configuration representing the specimen. It

consists of 157 nodes and 123 elements. Six thermocouples (for location,
see Fig. 9) recorded temperature changes during the test. The method fol-
lowed was to determine the temperature distribution with the first nine
temperature couple readings as input source and the tenth temperature
reading as a check against the computed temperature at the same loca-
tion. With a known temperature field, a thermal stress field can thus be
calculated.
The temperature and thermal stress computations involved modified
versions of existing finite element analyses for axisymmetric solids. The
same finite element mesh was utilized for both the temperature and stress
calculations.
Temperature distributions were calculated according to the procedures
described in Ref 2. Steady-state temperature field was determined first
using all ten thermocouple temperature readings as input source. Tran-
sient temperature distributions were then obtained employing previous
steady-state temperatures as initial temperatures and the first nine thermo-
couple readings as input temperatures. Results were stored in digital form
for subsequent thermal stress calculations.
The thermal stress analysis was based on the finite element formula-
tions of Farhoomad and Wilson [3] with modifications to the solution
procedures and materials formulation.
Results show that the fin is under high tensile stresses during the heat-
ing cycle, but the sign of stresses is reversed during the cooling cycle.
The stress cycles are shown for HI3 for both the severe and mild cycle
in Figs. 10 and 11. The maximum stress levels experienced in both mate-
rials are summarized in Table 3.

Thermal Fatigue Results for H13 and 18Ni Maraging Alloy

The stress results show that, at 593 °C, even the severe thermal cycle
will not cause plastic strain to occur at the start of testing. However, as
discussed later, the properties of H13 degrade and, after about 8000 to
16 000 cycles, the strength has decreased to a point where plastic defor-
mation becomes possible in an unnotched specimen using the severe ther-
mal cycle.
Thermal fatigue results are shown in Figs. 12 and 13. The approximate
cycles to initiate cracks are shown in Table 4. Feiilure did not occur within
the test cycles under conditions not involving plastic working. A possible
exception is the notched 18Ni specimen tested under the mild cycle. In this
specimen, the calculated stress at the notch is 538 MPa, while the yield
strength at 593 °C is 758 MPa. The strength degraded during testing but
was still 600 MPa even after 200 h (28 000 cycles). Thus, at crack initi-
ation (3000 cycles), the crack has apparently started in a zone subject to

^I^
^
3
•Sf

is
a
-s:

-4^

500
f c
i sV co
1
Z 1^
400 o
1 ^
'
300 -

200

100

r
9 ^ ^ 2 or 6
0 7^
s. > _-,-^"^ • ^

K
\
-100 ^

1
^
-200 3or7
^
-300 - \
2ar6
1 ^/
-400 -
\ /
-500 - \ lor 5

— ttcin 1 1 1 1 1 1 1
10 15 20 25 30 35
Time, Seconds

FIG. 10—Stress cycle of HI3 specimen fin during the severe cycle.

elastic strain but at a stress level near the point of plastic working. A
possible explanation is that the stress concentration at the notch was
higher than expected.
During testing, additional specimens were cycled and removed at inter-
vals until sufficient data had been obtained to observe the hardness changes
across the fin section. Specimens were removed at 250, 500, 1000, 2000,
4000, 8000, and 16 000 cycles. The specimens were sectioned and micro-
hardness surveys performed on the fin. These hardnesses were converted
to Rockwell C values, and isohardness curves were drawn across diagrams
representing the fins. The information is summarized in Fig. 14. This
figure also gives an estimate of the equivalent time that the outer edge of
the fin has spent at 593 °C. The estimates were made from tempering
curves by summing the effects of small temperature increments from 204
to593°C.
If the change in hardness (Fig. 14) is compared with the tempering

-Tronsftr H»ating Cooling

lor S
. 1

2 or 6

TV®
-..JorT ^ ^ 5 v
(/'
l/fiorS

VL 4ore

h) '°

1 3 or 7
1——-^
2 or 6

-
\^rl^ , 1.- '
100 ISO

Tim* Sftconds

FIG. 11—Stress cycle of H13 specimen during the mild cycle.

TABLE 3—Maximum stress levels at the surface of the thermal fatigue specimens.

Stress, MPa

Ultimate
Severe Cycle Mild Cycle Tensile 0.2%
Strength Yield
Alloy Plain Notched Plain Notched at593°C at593°C

H13 537 1504 241 674 737 606

(400)« (372)
18Ni 400 1120 192 538 930 758
(716) (600)

" Values in parentheses are for specimens tested after a 200 h exposure at 593 °C. 200 h is
equivalent to 25 000 mild cycles or 3.2 x 10' severe cycles for H13 and 28 500 mild cycles or
3.9 X 10' severe cycles for 18Ni maraging alloy.

curves in Figs. 5 and 6, it will be seen that stress acts to accelerate the
normal tempering changes in the structure. In the case of the 18Ni marag-
ing alloy cycled under the severe thermal cycle, there is evidence that
strain hardening occurs in the body of the fin due to the high stresses im-
posed in this cycle. The effect observed in the HI3 tool steel and in 18Ni

(o) H 13 UNNOTCHED <b) NOTCHED i ' 18 NI

- DISK SPECIMEN , H.I3 AND/
MARAGMiy
18 Ni SPECIMEN DID NOT //. SPECI- f
CRACK WITHIN 30,000 4// MEN /
_ CYCLES /f/

11 /
17*
If
/ /
//' /;
fi
1 / / / , 1 // 1
15,000 20,000 2S,000 0 SpOO
NUMBER OF THERMAL CYCLES

FIG. 12—(a) Crack growth of successive cracks in one unnotched HIS specimen, (b) Aver-
age crack growth of four cracks in HIS and ISNi notched specimens. Specimens cycled be-
tween 59S and 204 °C under the severe conditions.

0 5,000 10,000
Number Of Thermal Cycles

FIG. 13—Average crack growth of successive cycles in four notched HIS and JSNi marag-
ing alloy specimens. Specimens cycled between 59S and 204 °C under the mild conditions.

maraging alloy exposed to the mild cycle is one of failure produced by the
onset of cracking in plastically deformed material.

Effect of Microstructural Changes

Two sets of six unnotched disk specimens designated A and B were pre-
pared from a steel containing 0.48 percent carbon, 0.62 percent manganese
0.34 percent nickel, and 0.53 percent chromium. Both sets were heat
treated by oil quenching from 850°C and tempering at 565 °C. However,
Set A was heated to the austenitizing temperature and immediately
quenched, whereas Set B was held at temperature 30 min before quench-
ing. The structures produced are shown in Figs. 15 and 16. Set A shows

TABLE 4—Approximate thermal cycles to initiate cracking.

Severe Cycle Mild Cycle

Alloy Plain Notched Plain Notched

H13 16 500 0 >20 000 1500

18Ni >50 000 1000 >20 000 3000

NOTE—Cracking observations were made with a x 30 microscope.

THERMAL CYCLES

250 500 1000 2000 4000 8000 I6(

•>.A>vC*-^>.>-AA>-IU kAAAAX>>-J>»-«-| i->J>JV.*.JmjL*X>.*J |JI_IUt.AjCA.A>^I.A^ 1_*.X«.AJ1..I.A>.AA>4 I.^.A.'UUt.AJUkA^JU LXAAAJ

H {3 {Z^^^^ L_«=^yJ [ _ « = ^ ^ [i;;R;5r;_i [;;;;R?ir;j [;;;;;R;?r;:] \ ^

SEVERE
THERMAL
CYCLE i± n
Equivalent Exposure 47 46 44 43 42 39
At 593-C (Hours) (0016) (0.0321 (0125) (025) (0.5) (1.0)
I Rc45 I I nc42 I I Rc40^ I
H 13
MILD
THERMAL
CYCLE
Equivalent Exposure 42
(Hours) (2)
jU>.A>-t LA^Jt.AAAAJCitJU LAAAAAA.AJ<_IU(^ iJUUUUI_*JkXAA^ lJLAAA»>Jt_^A^ I.A»^JULAX>I.A>^

18 Ni
SEVERE Q r \ r Rc53 1 Llnc53_] L_R||3_J |1_RC53_| |_Rc53_|
54
THERMAL
CYCLE
I H.
Equivalent Exposure 53 53 53 53
(Hours) (0.013)
).0i3) (0.025) (0.05 (Ol)
Ujl.A>.A>-l>AA.AJk,J LJ„K>.JUlJiJl,J'.*~AJ^ |JKJkAJULAA.A.AAAJ

18 Ni Rc50 \
MILD
THERMAL
CYCLE
Equivalent Exposure 49 47
(Hours) (1.75) (3.5)

FIG. 14—Hardness changed in thermal fatigue specimen fins after cycling 250 to 16 000
times.

an incompletely hardened structure, but Set B is almost completely tem-

pered martensite. The specimens were then tested by thermally cycling the
fin between 550 and 260 °C using a bed temperature of 940 °C; the heating
time was 7.5 s, and the cooling time was 27 s. The thermal fatigue results
are shown in Fig. 17 where it is seen that all the Set A specimens cracked
almost immediately when testing commenced, with the cracks propagating
in the weak material at the grain boundaries. Set B specimens initiated
cracks in the range of 6000 to 24 000 cycles.
There are many examples of the importance of structure on thermal
fatigue results. A paper [4] in this publication discusses the effect of struc-
ture on thermal fatigue results obtained with MAR M-509. An example of
the effect of structure is given in NASA CR-121212 [5], where a pro-

FIG. 15—Steel with slack quenched structure (Set A), xJOO.

*^Tj:^ t - ^ - 4^-*^'

M^ ^ ^ > . . . ^ ^

FIG. 16—Steel with correctly quenched structure (Set B), xlOO.

nounced dendritic solidification pattern is shown to be detrimental to the

thermal fatigue resistance of TAZ-8A.
In general, it is important to give as much attention to processing as to
the composition of the alloy. There will be an optimum structure for
avoiding points of weakness such as dendrites, porosity, and weak micro-
structural constituents concentrated at grain boundaries. Structure can be

10000 15000 ZOOOO

Numbar Of Thtrmal Cycles

FIG. M—Effect of structure due to different quench rates in steel specimens on crack
initiation during thermal fatigue testing. Specimens cycled between 550 and 260°C.

made to work to improve thermal fatigue resistance, and the prime ex-
ample of this is directional solidification and single crystal structures.
Great improvements have been reported when directional solidification
can be used in a cast part [5-7].
Even if it is assumed that a particular composition is processed to pro-
duce the structure with the highest thermal fatigue resistance, there is
another effect to be considered. During thermal exposure there is often a
degradation in properties and potential crack sites can also develop.
Figure 18 shows a section through a double-edged wedge IN-100 speci-
men that has been subjected to thermal cycling between 1088 and 316°C
[5]. It can be seen that the alloy is subject to oxidation and a surface layer
of material with a lower content of carbon, chromium, aluminum, and
titanium can be produced which is thus of lower strength than the base
material. It is less able to resist the stresses produced by the thermal cycUng.
Furthermore, at the points where the grain boundaries meet the surface,
the effect is intensified, and the oxidized or weakened layer tends to pene-
trate further at these points. This has the effect of creating notches or
stress intensification points at the surface, and, when oxidation has pro-
ceeded far enough, a crack will initiate and follow the line of least resis-
tance in the structure. The success of coatings in improving thermal fa-
tigue resistance is largely due to their ability to retard oxidation.
This is one reason why there is an apparent incubation period before
thermal cracking commences, but another reason is the degradation of the
material properties themselves. Most materials are strengthened by a mar-
tensitic or an age hardening mechanism. Thus, during thermal cycling,
there is a tendency to temper the structure or to cause overaging and
thereby lower the strength of the material. The decrease in surface hard-
ness of twelve superalloys during thermal cycling has been reported [6],

and this almost certainly indicates that the strength of the material at the
surface has been reduced. A further effect is that tempering and over-
aging are accelerated by stress [8], and thus the degradation of properties
occurs faster while the material is being thermal cycled.

Summary
To summarize the events leading to a thermal fatigue failure, it has
been shown that failures occur because the materials have been stressed in
the plastic range. If the stresses are purely elastic, then failure would only
normally occur after alignment of dislocations during an incubation period
causes a void which would exceed the critical crack length and initiate a
self-sustaining crack.
Under high stress levels or in situations where stress concentration ef-
fects cause a crack to propagate, crack initiation occurs at the start of
thermd cycling or very soon thereafter. However, the common experience
during thermal fatigue testing is that a definite incubation period is re-
quired before cracking commences. It is believed that a number of rate-
dependent processes can be in progress during this period, either singly or
concurrently, and it is this factor that makes the development of predic-
tive equations very difficult.
One process that occurs is plastic deformation, and it is fairly common
to see considerable local plastic deformation on the surface of a specimen
before cracking commences. When cracking does occur, it is almost in-
variably in the areas where plastic deformation was the highest, and this
may be regarded as an exhaustion of ductility effect. A further factor is
that the strength of the material will be slowly degraded due to overaging
or tempering effects, and this will be stress accelerated. At points of max-
imum stress, it is likely that plastic deformation will occur to a greater ex-
tent than in the bulk of the section, and the mechanical properties will
therefore be degraded faster in these areas. As the mechanical properties
decrease, plastic deformation becomes easier, and thus the weakening
process is self-accelerating in individual small areas which will ultimately
provide the sites for crack initiation.
A second mechanism that occurs is oxidation depleting the surface of
materials at the higher exposure temperature. A nickel-base superalloy
can expect to be depleted of carbon, chromium, aluminum, and titanium,
at least. Depletion will often be preferential at the grain boundaries, and
wedges of low strength material will be formed which act like a notch.
After some finite time, the notch will assume the dimensions of the criti-
cal crack size, and a thermal crack will initiate and propagate.

Acknowledgments
The author is grateful to Kin Yeung, Edward Welch, and Anne Hum-
phreys who developed the computer program for predicting stresses under
thermal cycling conditions.

References
[1] Howes, M. A. H. in Fatigue at Elevated Temperatures, ASTM STP 520, American So-
ciety for Testing and Materials, 1976, pp. 242-254.
[2] Farhoomad, I. and Wilson, E. L., "Nonlinear Heat Transfer Analysis of Axisymmetric
Solids," Report PB 199169, University of California, Berkeley, 1970.
[3] Farhoomad, I. and Wilson, E. L., "A Nonlinear Finite Element Code for Analyzing the
Blast Response of Underground Structures," Report No. N-70-1, University of Califor-
nia, Berkeley, 1970.
[4] Beck, C. G. and Santhanam, A. T., this publication, pp. 123-140.
[5] Howes, M. A. H., "Additional Thermal Fatigue Data on Nickel- and Cobah-Base
Superalloys," NASA CR-121212, National Aeronautics and Space Administration,
Washington, D.C., 15 March 1973.
[6] Howes, M. A. H., "Thermal Fatigue Data on 15 Nickel- and Cobalt-Base Alloys,"
NASA CR-72738, National Aeronautics and Space Administration, Washington, D.C.,
15 May 1970.
[7] Howes, M. A. H., "Thermal Fatigue and Oxidation Data on TAZ-8A, MAR-M200, and
Udimet 700 Superalloys," NASA Cr-134775, Washington, D.C., 15 Jan. 1975.
[8] Howes, M. A. H., "Investigation of Stress-Accelerated Tempering and Aging of Fer-
rous Alloys," presented at XVI International Heat Treatment Conference, Stratford,
England, May 1976.

Thermal-Stress Fatigue Behavior of

Twenty-Six Superalloys

REFERENCE: Bizon, P. T. and Spera, D. A., "Thermal-Stress Fatigue Behavior of

Twenty-Six Superalloys," Thermal Fatigue of Materials and Components, ASTM STP
612, D. A. Spera and D. F. Mowbray, Eds., American Society for Testing and
Materials, 1976, pp. 106-122.

ABSTRACT: The comparative thermal-stress fatigue resistances of 26 nickel- and

cobalt-base alloys were determined by fluidized bed tests. Cycles to cracking differed by
almost three orders of magnitude for these materials, with directional solidification and
surface protection showing definite benefit. The alloy-coating combination with the
highest thermal-stress fatigue resistance was directionally solidified NASA TAZ-8A
with an RT-XP coating. Its oxidation resistance was also excellent, showing ap-
proximately a Vi percent weight loss after 14 000fluidizedbed cycles.

KEY WORDS: thermal fatigue, thermal stress, oxidation, coatings, nickel alloys,
cobalt alloys,fluidizedbed, heat resistant alloys, mechanical properties

Thermal-Stress fatigue is a potential mode of failure in any component

that is exposed to fluctuating temperatures. Resistance to thermal-stress
fatigue is one of the major criteria which should be considered when se-
lecting an alloy for a fluctuating temperature application. However, ther-
mal-stress fatigue resistance is not a basic material property such as tensile
strength, density, etc., but a structural response. The number of tem-
perature cycles required to cause thermal-stress fatigue cracking depends
not only on the alloy but also on the specimen geometry and test conditions.
In this investigation, the term "thermal-stress fatigue resistance" is used in
a comparative, rather than absolute, sense. Alloys are compared to one
another, all having been tested under the same test conditions using the
same specimen geometry.
This investigation is part of a program of studies of failure mechanisms
and life prediction of turbine components being performed by the Lewis
Research Center of the National Aeronautics and Space Administration
(LRC/NASA). An overview of this program is given in Ref 7,^ together
' Materials engineer and fatigue branch consultant, respectively. National Aeronautics and
Space Administration Lewis Research Center, Cleveland, Ohio 44135.
^ The italic numbers in brackets refer to the list of references appended to this paper.

106

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
BIZON AND SPERA ON SUPERALLOYS 107

with a bibliography of NASA-LRC publications on fatigue, oxidation and

coatings, and turbine engine alloys. References 2-7 contain incremental
portions of the data obtained in the thermal-stress fatigue tests described
herein. All significant data from these reports are compiled and described
in detail in Ref 8.
The primary objective of this investigation was to determine the com-
parative thermal-stress fatigue resistance of a large number of nickel- and
cobalt-base alloys. A secondary objective was to determine the effects of
directional solidification and several coatings on both the thermal-stress
fatigue and oxidation resistance of some of these alloys. The alloys selected
include those used or proposed for use in the the hottest sections of ad-
vanced aircraft gas turbines. Therefore, the test conditions for this
evaluation may be more severe than present use conditions for many of the
alloys. As an example, the maximum metal temperature during a typical test
cycle was about 1074°C (1965 °F).
The thermal-stress fatigue tests were carried out in a fluidized bed facility
that was designed, built, and operated by the Illinois Institute of
Technology Research Institute (IITRI) under contract to NASA-LRC.
Fluidized beds were first used for rapidly heating and cooling thermal-stress
fatigue specimens by Glenny and his co-workers at the National Gas Tur-
bine EstabUshment in England in 1958 [9]. Since that time, fluidized bed
cycling has become widely used for evaluating the thermal-stress fatigue
behavior of both alloys and components [70-/5].
In addition to thermal-stress fatigue resistance, some mechanical proper-
ties of the alloys were also determined at selected temperatures. The con-
ventional mechanical property tests for the materials included in this in-
vestigation were conducted at the LRC.

Materials and Test Specimens

Alloys
The 35 combinations of composition, grain structure, and surface protec-
tion which were studied in this program are listed in Table 1. Table 2 lists
the compositions of the dloys (chemical analyses provided by vendor or in-
dependent laboratory of heats used for specimens) along with the various
heat treatments applied to them. Mechanical properties are given in Table 3
for the same heat of alloy that was used for the thermal-stress fatigue
specimens.
A new alloy included in this investigation is directionally solidified NASA
TAZ-8A. In an earlier investigation [4], the random polycrystalline form of
this alloy was found to have the highest resistance to thermal-stress fa-
tigue among 14 uncoated alloys. This indicated that directional soUdifica-
tion or coating or both of TAZ-8A might produce a material with even

TABLE I—Alloys-coatings investigated (coated and uncoated

specimens purchased from vendors except where noted).

NASA TAZ-8A DS + RT-XP coat B 1900 + Hf + Jocoat'

NASA TAZ-8A DS + NiCrAlY overlay B 1900 + Jocoat
NASA TAZ-8A DS" B 1900
NASA TAZ-8A
U 700 cast
U 700 wrought
NX 188 DS + RT-IA coat'
NX 188 DS' X40
NX 188 + RT-lAcoat' IN 162
NX 188' TD NiCr
Mar-M200 DS + NiCrAlY overlay IN713C
Mar-M200 DS Mar-M509
Mar-M200 + Jocoat NASA VI-A
Mar-M200 Rene SO'
IN 738
IN 100 DS + Jocoat RBH''
INIOODS Mar-M302
IN 100 + Jocoat WI52
IN 100 M22
NASAWAZ-20DS + Jocoat"
NASAWAZ-20 + Jocoaf

NOTE—DS itidicates that the alloy was cast with a directionally solidified
grain structure.
"Specimens cast at NASA Lewis Research Center.
' Specimens with and without coatings supplied by Pratt & Whitney Air-
craft Corp.
'Specimens supplied by General Electric Corp.
•*Specimens supplied by Cabot Corp.

better thermal-stress fatigue resistance. Specimens with directional poly-

crystalline structure were then cast from the same alloy heat, both at the
LRC and commercially, for evaluation in this investigation.

Surface Protection
Brief descriptions of the four protective coatings used in this study are as
follows:
Jocoat—This is a commercial silicon-modified nickel aluminide coating
(Pratt and Whitney Aircraft (PWA) proprietary process specified as PWA
47).
RT-IA—RT-IA is a commercial chromium-aluminum duplex coating
(Chromalloy American Corporation Research and Technology Division
proprietary process similar to specification PWA 32 but with a lower
process temperature).
RT-XP—RT-XP is a coating containing an aluminide with a case depth
of about 70 Mm (2.7 mils) (Chromalloy American Corporation Research
and Technology Division proprietary process).
NiCrAlY—This is a commercial Ni-15.2Cr-12Al-0.33Y electron-beam

vapor-deposited overlay coating about 135 /im (5.3 mils) thick (PWA
proprietary process specified as PWA 267).

Test Specimens
In this study, two types of specimen configurations (Fig. 1) were used:
bars with double-wedge cross sections and round cross-sectional tensile
bars. No significant differences were observed in the results of duplicate
tests of the double-wedge shape specimens whether end grooves or holes
were used for fastening them into the supporting fixtures.
Figure 1 also shows the designs of specimen used to measure conventional
tensile and stress-rupture properties. Designs A and B are similar except
that Design B has flat grips so that it can be fabricated from 0.635-cm (0.25-
in.) plate. The only difference between Designs C and A is the longer gage
length in Design C.
All cast alloy specimens were cast to size. For the random polycrystalhne
specimens, inoculated molds were used to produce fine grain structures.
Typical surface grain size at the test sections was about 1.6 mm (0.06 in.)
diameter. The directionally solidified polycrystalhne specimens were made
using a controlled solidification process similar to that detailed in Ref 16.
Only directionally solidified specimens with no grain boundaries in-
tercepting the leading edge were tested. Specimens of the two wrought
alloys (U 700 and TD NiCr) were machined with the specimen axis parallel
to the rolling or extrusion direction. All specimens were given radiographic,
fluorescent penetrant, and visual inspections before testing.

Fluidized Bed Facility and Test Procedure

Fluidized Bed Facility

Figure 2 shows groups of specimens in their respective supporting fix-
tures. The outer two specimens were always dummy test specimens. The
supporting fixture using holes is preferred for tests at higher temperatures
because of the lower mass of the fixture and because the supporting rods are
only partially exposed to the high temperatures. The specimens were loosely
supported in both types of fixtures to allow free thermal expansion.
Figure 3 is a drawing showing a cutaway view of the fluidized bed test
facility. Both heating and cooling beds consist of retorts filled with 300 to
540-fim (0.012 to 0.021-in.)-diameter alimiina particles through which air is
pumped. Adjustment of the airflow allows the particles to develop a
churning, circulating action, hence, the name "fluidized." The large num-
ber of particles in the beds and their fluid action promote uniform, high
heat transfer rates. Each group of specimens in the supporting fixture
was cycled between the beds by means of an automatically controlled

M J= -C J= i' i-
«N •» '9- NO ri
t. 1- fc
0 5 ^ 0 -c 0 -c
t^ t— 4— 0

FP? uT F- |_^ —

' ^
8 1s r-
ON
•B 8Q °«^ s§ S 0
X-1
K S 0 £
tn (N b u. ^ <o SS2 SF a £•
^-- 0

yd 8 0
NO
3—
? ?—
| Uh U
r — 1^1
I
^O 0 rr r- C- r. 3 o ^ O
00 (8 —• '-^ nj — ^ S3 "
flj r- 00 «

>
I 0
0
V

d 0
X

V 5 z £ £
5IN dS 0S -0
Si
£!?

s:
o r^ d d

Ia
a o .0

— VO •^
(N O
-: d 0 0 0 0

— . f^ •.s.

ID <*> I d « : S . — (N
00 «X

NO r~ O f^

^ p3 —
00 ~" I '

i
fo d «

00 X) X>

S Xi ^ JS Xt XI O -C -C X>
Ills
•3
X
8 ?
NO d

I -O "O 00 O ~-

§
-s: 0 0 0 0
•xs
• m
d d d
s d
'^ d d
~
d d d
Ic V V 9 V V

.0 0 0 r^ 0
• 0 0 ™ 0 0 0
d d d d d d d d d
V
V 0 V
I — 00 0
0 •t
0
0 i - 3 0
- r- •0 r~ ^ 00
NO 00
0 0 0 0 d d d d d d d d d d d d d d d
<3
I
W
_! ss 3 3::^
00
VO
io

no •V Q X U > f^ \ o «8 m
u-
X sa
< i- iii al
(N t
—• t o
^
*
00
— :s H u
£

< N <
i S +
u
S i
ss <
Is s S 8 | s i < 8 3 z 2
r-
00

s^ Xz ^<z II
(/5 O 2R —
7:
2 Z OS < * - z Q 2
Z X M S H

li
^^^

-S3 2^u S
8 s^S" 5-* «o g •-
o ^ C- •«

^1
m iU

s •

2 a!

o o

- s
s RS

8
3

U 8 •&. (S w^ vo O I

S o s sOTf\oa>»^r-)Ooo
S «" —
= 'vo (STfrf^O^ — &<rof0^r-«*>w^O
r- Z H

Si = i . •/% iC «^ 00 2 ^ ^ 00 —• IN -00

is 3
9 • S3 ? « Ss g - °

ih I <S <N —• — I S IN I — — M
• ^ o v o—
o —
<N
O w ^ ^ o—
<N
o f—
^tO
(S
'CsO
^ s

r ^ ^ O 0 \ O ( N ^ P ^ O ^ « N C T \ r ~ r - —

oo rr . ^O to (N <N VO <S m r- r-
\o — <= 0\ O lO «o
oo
. »i->

• 00 (S
—
—.—
Ov00 00 <s oo «-.» <Nr- ~^
fS 00 00
s2
• (*i
«
if!! m
• ^
:c QO S OS OS
fM
o • (S 00 f*^
M-1 00
w-1 QO 00
<o - U-1

OdOv'*lf~* — — O

J8
li.
I 00 — ^ > 0 " o O » * 1 0 \ ' ^ ^
— (NO-^-^r- — o o ^ > 0 ' « Q o o

I
mr^xn^OOsooMfS
S O i O O O t f s C S i n O — rrt O » o S - - r - - O O ' O Q 0 O t - - * O O '

BQ
3|
iili (N>*
o o
.OO^CTitOTf
-r-oo-^-* — -•I
— 1 roro
Oro
•-TOOf^oo'.O'O-g-v-ifNQ^
Or-OsOsr^t— OisOOOOi

• r- • 'O 00 . vo 1^ I

M«il
O u ^ s p v ^ O Q - O - O O M f S l N
• r- • * 00 • — o 1 S O O O O ^ O — O -00 • S 0 0 \ 0 0 v

OOr'^tri(NfnO'Oio — QO'*v-i'«tO
- H T f O m O — — • ^ O O Q O — <N —

O O 00 m I s p *0 OS r o '
Os TT t ^ OS I
• oooor-r^Of^osoocs

II! u • u < u u < < < < m ^ u u u u u u <<<<<<

•^ i? S
*^ 8 <• X N s <<
<
CO = i s yz
? i 81 ^< +8 H<o or. ^ o f^
• < - < * - _ Q §l8
I z m zX m S H S2zzz«So< S D S S s D

i
•111
11 a

{|ii!L
iM^
f
pIlHIi

Onn m1> DESIGN B

^-^'^^—rlO 25)
EACH rRADIUS,a067ia026) DESIGNS A AND C
EDGE ra635
y (a 250) 0 47 --L27
RADIUS, 3Q?.
(019)- (O500I
al02(a(H0) A / 118
asM
ia352)
|-a.25)--v (a 499)

rl.OO
a 395)
L440
( 0 567).
© ^-RADIUS.

CLEARANCE
DIAMETER
7.«
(iOO)
r -o 9.53
a 75)

iai6
L —
0 6 4 ( 0 25)

DESIGN D L
A 0643 Z03
0.«3a375^.^ a 00)
(0253) (0801
RADIUS, B 0500 a 03
a318(ai25l (0197) (0 8 0
L440
C 0643 386
j Lj'-vP
o ( 0 567),

047
(0 253) (152)

( 0 19)-

THERMAL FATIGUE AND 0X1DAIION TINSILE AND RUPTURE

FIG. 1—Geometry of test specimens (.all dimensions in cm (in.)).

transfer mechanism operated by pneumatic cylinders. Additional descrip-

tion of the bed and its operation can be found in Ref 17.

Thermal-Stress Fatigue Test Procedure

Comparative thermal-stress fatigue resistance was determined ex-
perimentally by simultaneously testing different materials of the same
geometry and comparing the number of cycles required to initiate the first
crack. A group of up to 18 double-wedge shaped specimens was alternately
heated for 3 min and then cooled for 3 min. All materials were tested with
the heating bed temperature held at 1088 °C (1990°F) and the cooUng bed
temperature at 316°C (600 °F). A typical transient temperature distribu-
tion [8] shows that maximum metal temperatures were about 14°C (25 °F)
less than the heating bed temperature, and minimum metal temperatures
were about 33 °C (60 °F) higher than the cooling bed temperature. The ef-
fect on some of the materials of raising both bed temperatures 42 °C
(75 °F) was also evaluated.
Prior to testing and periodically thereafter, both edges of each specimen
were examined visually using a microscope with a magnification of x 30.
Only the surfaces within ±3.8 cm (± 1.5 in.) of midlength were examined
for cracks. Inspections of most tested specimens were made according to a
schedule of 25, 50, 100, 200, 300, 500, 700, and 1000 cycles. Inspection in-
tervals for any series never exceeded KXX) cycles, including the series which
was tested the longest (15 (XX) cycles). For some materials (mostly those with

i. ..Sw J

Em GROOVE SUPPORTS THROUGH HOLE SUPPORTS

FIG. 2—Specimen supporting fixture.

POSITIONED BY
PNEUMATIC
ACTUATORS-.

ALUMINA RETAINER

HEATING BED
123 CM DIA)

ALUMNA PARTICLES

SIUCON CARBIDE - COOLING-BED

HEATING ELEMENTS (12) r' \ (36 CM DIA)

REFRACTORY ^ -SPECIMENS
INSULATION- •'
r INSULATION
INSPECTION PORT-
RETORT
THERMOCOUPLE COOUNG AIR
AIR
l€AT EXCHANGER
INSPECTION PORT

CALROD HEATING ELEMENTS (3)

A!R»

FIG. 3—Fluidized bed test facility at Illinois Institute of Technology Research Institute.

longer lives), duplicate or triplicate tests were run. The number of cycles to
crack initiation was taken as the average of the number of cycles at the last
inspection without cracks and the number of cycles at the first inspection
with a crack.

Results and Discussion

Thermal-Stress Fatigue
The numbers of cycles required to initiate cracks in the 35 combinations
of alloys and coatings are shown in Fig. 4. All results are for the 0.635-cm
(0.25-in.)-radius edge only. For the same specimen geometry and test con-
ditions, the lives varied from less than 25 to 12 500 cycles. Reproducibility
of test data was generally good.
The class of materials having the longest thermal-stress fatigue lives was
found to be cast alloys with directionally soUdified polycrystalline grain

NASA TAZ-8A DS + RT-XP coat J

Mar-M 200 DS t NiCrAlY overlay
NA$ATAZ-8A DS tNiCrAlY overlay 1
1
NX188DS+RT-lAcoat 1
NASA TAZ-8A DS 1
NX ISS DS 1
Mar-M200 DS
IN 100 DS + Jocoat
INIOODS
NASA WAZ-20 DS +Jocoat
B IWO-fHf-f Jocoat
,1
B 1900 4 Jocoat
1
NASA TAZ-8A
NX188 + RT-lAcoat
1
X«)
B1%0
IN 162
INIOO + Jocoat
TO NiCr
IN713C fff
Mar-M 509
NX 188
, 1
NASA Vl-A
NASA WAZ-20+Jxoat j
1
Rene'80
IN 733
RBH
Mar-M 302
U 700 cast
Wl 52
IN 100 1 Bed temperatures
Mar-M 2 0 0 * Jocoat 1 1088 and 316° C (1990 and 600° Fl
Mar-M 200 3 min. immersion in each bed
U 700 wrougtit
M22
1
100 1000 lOOOO 100000
Cycles to first cracii

FIG. 4—Comparative thermal-stress fatigue resistances of nickel- and cobalt-base alloys.

Structures. The alloy-coating combination with the highest thermal-stress

fatigue resistance was directionally solidified NASA TAZ-8A with an RT-
XP coating. Its thermal-stress fatigue life was almost double that of the
next best alloy-coating combination, directionally soUdified Mar-M200
with the overlay coating NiCrAlY. The application of a coating in all
cases improved the thermal-stress fatigue resistance of the substrate alloy.
Of all the random polycrystalline alloys tested, hafnium-modified
B 1900 with Jocoat had the highest thermal-stress fatigue resistance (1480
cycles). However, of the 19 uncoated random polycrystalline alloys tested,
NASA TAZ-8A again had the highest thermal-stress fatigue resistance
(700 cycles).
Some of the materials were also tested when both the maximum and
minimum temperatures of the fluidized beds were raised by 42 °C (75°F).
A comparison of the thermal-stress fatigue resistances for materials tested
at both bed temperature conditions is presented in Fig. 5. This figure
indicates that raising the bed temperatures resulted in only minor changes
in relative ordering of the thermal-stress fatigue resistances.

10000

cycles to firsi crack

FIG. S—Comparison of thermal-stress fatigue resistances for alloys tested at both bed tem-
perature conditions.

An interesting observation was made regarding the two alloys NASA

TAZ-8A and M 22. Reference to Table 2 shows the similarity of the
compositions (considering the refractory elements as a group) of these
two alloys. It is apparent from Fig. 4, however, that their thermal-stress
fatigue resistances differ widely. It is suggested that further study of these
two alloys might serve to identify the microstructural aspects which af-
fed thermal-stress fatigue resistance in nickel-base alloys and perhaps lead
to a better understanding of the thermal-stress fatigue process in such
materials.
The superior resistance to thermal-stress fatigue of the five direc-
tionally solidified polycrystalline alloys in this study may be attributed to
two factors: lower modulus of elasticity and an absence of transverse
grain boundaries. The modulus of elasticity of a directionally solidified
alloy (along the growth direction) can be 30 to 50 percent lower than that
of the same alloy in the random polycrystalline form [8], whereas the
thermal properties (thermal expansion, conductivity, etc.) are independent
of the structure, being dependent only on the composition. Therefore, in
specimens of the same composition and geometry under identical test con-
ditions, the directionally solidified specimen will have significantly lower
thermal stresses for the same induced strain than a specimen with a
random polycrystalline structure. Also, the directional casting process
produces long columnar grains aligned in the longitudinal direction of the
specimen, and no grain boundaries intercept the specimen leading edge.
This removes locations which might serve as crack nuclei. The beneficial

effect of the elimination of transverse grain boundaries can be seen by

noting crack propagation in directional alloys. When a crack did initiate,
the directional grains tended to be barriers to propagation [8]. Figure 6a
shows a macrophotograph of directionally solidified Mar-M200 before
and after testing for 2400 cycles at bed temperatures of 1130 and 357 °C
(2065 and 675 °F). Note how the crack turned to grow in the longitudinal
direction.

Oxidation
The materials exhibited a wide variation in oxidation resistance. In
general, a coating or vapor-deposited overlay on an alloy gave a greatly
reduced weight loss. Without surface protection, the directionally solidified
alloys oxidized much faster than the same compositions in the random
polycrystalline form [8], The severe oxidation of an uncoated directionally
solidified alloy can be seen by referring to Fig. 6b which shows direc-
tionally solidified IN 100 before testing and after testing for 2400 and
2900 cycles at bed temperatures of 1130 and 357 °C (2065 and 675 °F).
Figure 7 shows a comparison of the weight changes for directionally
solidified TAZ-8A and Mar-M200, the alloy with the next highest thermal-
stress fatigue resistance. All data are for double-wedge specimens tested at
bed temperatures of 1088 and 316°C (1990 and 600°F). Results are pre-
sented for both alloys without a coating and with the NiCrAlY overlay
coating. In addition, data for TAZ-8A with the RT-XP coat are also pre-
sented. The much higher weight loss of Mar-M200 compared to that of
TAZ-8A in the uncoated condition is apparent. The effect of applying
overlay coatings to both alloys is to greatly improve their oxidation re-
sistance. Note that TAZ-8A with the RT-XP coating exhibited only a Vi
percent weight loss after 14 OCX) cycles of testing, indicating excellent re-
sistance to oxidation.

Summary of Results
The comparative thermal-stress fatigue resistances of 26 nickel- and
cobalt-base alloys were determined using the fluidized bed technique. A
total of 35 combinations of compositions, grain structure, and coatings
were studied. All materials were evaluated by thermally cycling double-
wedge specimens between two fluidized bed furnaces maintained at 1088
and 316°C (1990 and 6(X)°F). Some materials were also exposed to bed tem-
peratures of 1130 and 357 °C (2065 and 675 °F). Immersion times were 3 min
in both the high and low temperature beds. Thermal-stress fatigue resist-
ance was based on the number of cycles required to initiate a crack. The
major results obtained are as follows.
1. Thermal-stress fatigue lives under identical test conditions ranged

OtSCSIPTIOH, imCOATtB
cvcusroFiBsrMACK. ms
C V O i ^ TO PHOroSRAPH,. 0

iaiMA(t-Mmtis*aoy.

25 nm

iWINtOOCSftlUW.

FIG. 6—Directionally solidified thermal-stress fatigue specimens before and after testing at
bed temperatures of 1130 and357°C (2065 and 675 °F).

NASA TAZ-IA DS * flT-XP COAT

M S A TAZ-lA DS * NICrAIV W E R I A Y
I NASA TAZ-SA DS IMCOAKD
< MAR-M 200 DS 4 NKrAlY OVEHUY
' MAR-M ZOO DS UNC0A1ED

1 AVC CRACK Wn\lADON

tEO TEMPS.
•---. C F

31^ lUft
1 THREE-MINUTE
IMMERSION IN
EACHIED

«m «oa tm am lOODO 12000 li

FIG. 7—Weight change for two directionally solidified alloys from fluidized bed tests.

from less than 25 to 12 500 cycles among the 35 combinations of com-

positions, grain structure, and coatings.
2. The class of alloys having the longest thermal-stress fatigue lives were
cast alloys with directionally solidified polycrystalline grain structures.
3. The four coatings employed always increased the thermal-stress
fatigue resistance of the substrate alloy. The application of the diffusion
coatings or the vapor-deposited overlay coating on the directionally
solidified alloys also greatly reduced weight loss.
4. The alloy-coating combination with the highest thermal-stress fatigue
resistance was directionally solidified NASA TAZ-8A with an RT-XP
coating (proprietary coating containing an aluminide from Chromalloy
American Corporation). Its thermal-stress fatigue life was almost twice that
of the next best alloy-coating combination. Its oxidation resistance was also
excellent. Of the 19 uncoated random polycrystalline alloys tested, NASA
TAZ-8A had the highest thermal-stress fatigue resistance.

References
[/] Spera, D. A. and Grisaffe, S. J., "Life Prediction of Turbine Components: On-Going
Studies at the NASA Lewis Research Center," Technical Memorandum X-2664,
National Aeronautics and Space Administration, 1973.
[2\ Bizon, P. T.'and Oldrieve, R. E., "Thermal Fatigue Resistance of NASA WAZ-20 Alloy
With Three Commercial Coatings," Technical Memorandum X-3168, National
Aeronautics and Space Administration, 197S.
[J] Spera, D. A., Howes, M. A., and Bizon, P. T., "Thermal Fatigue Resistance of 15 High-
Temperature Alloys Determined by the Fluidized-Bed Technique," Technical Memoran-
dum X-52975, National Aeronautics and Space Administration, 1971.
[4\ Howes, M. A., "Thermal Fatigue Data on 15 Nickel- and Cobalt-Base Alloys," IIT
Research Institute, Report IITRI-B6078-38, Report CR-72738, National Aeronautics and
Space Administration, 1970.
[J] Howes, M. A., "Additional Thermal Fatigue Data on Nickel- and Cobalt-Base
Superalloys," IIT Research Institute, Report IITRI-B6107-34 (Part 1), Report CR-
121211, National Aeronautics and Space Administration, 1973.
[6\ Howes, M. A., "Additional Thermal Fatigue Data on Nickel- and Cobalt-Base
Superalloys," IIT Research Institute Report IITRI-B6107-34 (Part 2), Report CR-
121212, National Aeronautics and Space Administration, 1973.
[7] Howes, M. A., "Thermal Fatigue and Oxidation Data on TAZ-8A, Mar-M 200, and
Udimet 700 Superalloys," IIT Research Institute, Report IITRI-B6124-21, Report CR-
134775, National Aeronautics and Space Administration, 1975.

[8] Bizon, P. T. and Spera, D. A., "Comparative Thermal Fatigue Resistances of 26 Nickel-
and Cobalt-Base Alloys," Technical Note D-8071, National Aeronautics and Space Ad-
ministration, 1975.
[P] Glenny, E., Northwood, J. E., Shaw, S. W., and Taylor, T. A., Journal of the Institute
of Metals, Vol. 87, 1958-9, pp. 294-302.
[10] Glenny, E. and Taylor, T. A., Journal of the Institute of Metals, Vol. 88, 1959-60, pp.
449-461.
[11] Franklin, A. W., Heslop, J., and Smith, R. A., Journal of the Institute of Metals, Vol.
92, 1963-4, pp. 313-321.
[72] Howes, M. A., Die Casting Engineering, Vol. 13, 1969, p. 12.
[13] Howes, M. A. and Saperstein, Z. P., Welding, Vol. 48, 1969, pp. 543-545.
[14] Rostoker, W., Journal of Materials, Vol. 4, No. 1, 1969, pp. 117-144.
[75] Mowbray, D. F., Woodford, D. A., and Brandt, D. E. in Fatigue at Elevated Tem-
peratures, ASTM STP 520, American Society for Testing and Materials, 1973, pp.
416-426.
[75] Freche, J. C , Waters, W. J., and Ashbrook, R. L., "Application of Directional
Solidification to a NASA Nickel-Base Alloy (TAZ-8B)," Technical Note D-4390,
National Aeronautics and Space Administration, 1968.
[77] Howes, M. A. in Fatigue at Elevated Temperatures, ASTM STP 520. American Society
for Testing and Materials, 1973, pp. 242-254.

Effect of Microstructure on the

Thermal Fatigue Resistance of a
Cast Cobalt-Base Alloy, Mar-M509

REFERENCE: Beck, C. G. and Santhanam, A. T., "Effect of Microstructure on the

Thermal Fatigue Resistance of a Cast Cobalt-Base Alloy, Mar-M509," Thermal Fatigue
of Materials and Components, ASTMSTP 612, D. A. Spera and D. F. Mowbray, Eds.,
American Society for Testing and Materials, 1976, pp. 123-140.

ABSTRACT: A study was undertaken to determine whether the thermal fatigue re-
sponse of the cast cobalt-base alloy, Mar-M509, could be modified by changing the
casting parameters. Single-edge wedge-type specimens were tested in a fluidized bed
using a thermal cycle of 1915/52S''F with a cycle time of 6 min divided equally be-
tween heating and cooling. The results revealed that the thermal fatigue life of Mar-
M509, from the point of view of both crack initiation and propagation, could be sig-
nificantly improved by changing the casting variables. Cracking was found to be pre-
dominantly transgranular and often initiated at interdendritic carbides. A good cor-
relation was obtained between the secondary dendrite arm spacing and the number of
cycles for crack initiation, as well as the crack propagation rates. Small dendrite
spacings reduced the thermal fatigue life by accelerating both the crack initiation and
propagation. Precipitation of MC and MuCt carbide occurred during the thermal
cycling and affected the crack propagation rate. Based on these results, a model for
thermal fatigue cracking in Mar-M509 is presented.

KEY WORDS: thermal fatigue, fatigue (materials), cobalt alloys, microstructure

In gas turbines, critical components such as blades and guide vanes are
subjected to rapid changes in temperature during startups and shutdowns.
The outer layers and thinner sections of the components respond more
quickly to changes in temperature than the interior. The resulting thermal
gradients give rise to thermal strains and stresses. Repetition of the stress
cycles during service can sometimes cause cracking of the component. This
type of cracking under fluctuating temperature conditions is termed ther-
mal fatigue. The susceptibility of blade and vane materials to thermal
fatigue is therefore a matter of vital concern to the designer.

'Engineer and senior engineer, respectively, Westinghouse Research Laboratories, Pitts-

burgh, Pa. 15235.

123

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyi-ight 1976 b y Aby
S I M InteiTiational www.astm.org
University of Washington (University of Washington) pursuant to Lic
124 THERMAL FATIGUE OF MATERIALS AND COMPONENTS

The evaluation of the thermal fatigue resistance of a material is a for-

midable task since it depends on a number of variables such as the
maximum and minimum temperature, heating and cooling rate, specimen
geometry, environment, etc. One of the methods developed in recent years
to simulate the thermal cycles experienced in actual applications is fluidized
bed testing [7]^ in which a specimen is rapidly heated and cooled in beds of
fluidized particles, and the number of thermal cycles required for crack
initiation is measured. While a good number of high temperature materials
have been ranked on this basis for thermal fatigue resistance [2], very Uttle
attention has been given to controlling and optimizing the microstructure
for the minimization of thermal fatigue cracking. A study was undertaken
to determine whether the thermal fatigue response of the cobalt-base
alloy, Mar-M509, could be modified by changing its microstructure.

Experimental
The cobalt-base alloy, Mar-M509, is presently used as one of the guide
vane materials for land and air gas turbines. The chemical composition of
the heat used is given in Table 1. Specimens were cast using different mold
and superheat temperatures which produced variations in grain size, den-
drite spacings, and amount of carbides. Single-edge wedge-type specimens
with geometry as shown in Fig. 1 were prepared from these castings.
The thermal fatigue testing was carried out in a fluidized bed test facility
at the Illinois Institute of Technology Research Institute (IITRI). It consists
of a high-temperature bed heated externally with silicon carbide elements
and a low-temperature bed equipped with an air- or water-cooled heat ex-
changer and an internal heat source of low power. The specimens are cycled
between the high- and low-temperature beds by means of automatically
controlled pneumatic cylinders. The bed consists of alumina particles in the
28 to 48 mesh size range. Air enters from the bottom and keeps the bed
fluidized. As a result, the rate of heat transfer to the specimen is high. The
heat content of the media is also high, so that a leuge number of specimens
can be heated and cooled rapidly without lowering the bed temperature
significantly.
A thermal cycle of 1915/525 °F (1046/274 °C) was used in the present
study. The specimens were immersed in each bed for 3 min. Thermocouples
were placed at five locations across the specimen section, and the tem-
peratures were recorded during heating and cooling. The time history of the
tip (thin trailing edge) temperature and that of the center of the specimen is
displayed in Fig. 2. Note that the tip gets heated and cooled more rapidly
than the center (Thermocouple Position 3) and gives rise to a transient tem-
perature difference which maximizes at short times. It is this transient tem-

^ The italic numbers in brackets refer to the list of references appended to this paper.

o 13
U 03
m

<
ao
1
o
V
g
a. 1
d
V
3
<6
o V

a §
d
o
d

I H
00

d
Si
I d
^
m
n Z O
<
l-i en
o
1/1 8
d
K
i
d
V

u d

<zifn

.241
.231 K-.98-H

3.94

.141,
.135'
(?)
.025"-| ..325'U».350",
00 I--040"
.030'|M n

270
K63< 282

All thermocouples are 0.010" below sur-

face and at mid-chord of specimen

FIG. 1 —Fluidized bed thermal fatigue specimen.

- ••' 1 1 '- "T 1 ' •• 1 ' 1

1800 Ti£,
f 1600 Center y^Center -
-
1s. 1400

1200
1
1
1 1000
800
(a) -
1 1 1 1 1
MO 0 60 120 180 240 300 360
' 1 1 1 I 1
+400
•»200 -
? 0
5 -200
(b)
1 1 1 1 1 1
-400
C 60 120 180 240 300 360
Elapsed Time, (sec)

FIG. 2—(a) Measured temperatures during rapid heating and cooling of wedge type
specimen of Mar-M509 and (b) transient temperature between the tip and the center of the
specimen.

perature difference that gives rise to thermal strains and stresses and con-
sequent cracking of a component.
The specimens were removed from the test rig at regular intervals, and the
edges were examined for crack initiation using a x30 microscope. In-
spections were made at 25, 50, and 100 cycles, after every 100 cycles up to
2000 cycles, and at 2500 and 3000 cycles. When a crack was observed, its
length from crack tip to specimen edge was measured on both surfaces, and
the average value was taken as the crack length.

Crack Initiation and Propagation

The number of cycles required to initiate the first crack and the rate of
propagation of the crack were determined for all the specimens representing
the different casting conditions. There are several methods by which the
number of cycles for crack initiation can be determined. One of the
methods suggested by Glenny [3] involves averaging the number of cycles
between one inspection cycle when no crack was observed and the next in-
spection to show a crack. This was determined for all the specimens. Figure
3 illustrates another method in which the average crack length is plotted as a
function of the number of cycles, and the curve is back-extrapolated to
"zero crack length." (In the present study, the extrapolation was carried to
a crack length of 0.01 in.) The thermal fatigue lives calculated by this
method were much shorter than those obtained by the averaging method. It
was felt that a measurement of crack area would be more appropriate. This
was computed from the crack length measurements on both surfaces.
Figure 4 shows the crack area as a function of the number of thermal cycles

Superheat, "F
200 300 400
.10 I- ^2000 o 2.5
11900 ^
E
Sl800 o
.08 i0 3
i
I .06 1.5 2
c

I
.04 LO

.02 0.5

0 0
1000 2000 3000
Number of Cycles

FIG. 3—Crack length versus number of cycles for various casting parameters ofMar-M509.

.006
Superheat, "F
200 300 400
.005 I - **". 2000 a
t 1900 A
- 3

T .004
1 1800O
a
I
•s
.003

1 .002
.001
Numbers

0
1000 2000 3000
Number of Cycles

FIG. A—Area of crack versus number of cycles for various casting parameters ofMar-M509.

for the specimens tested. The thermal cycles required for crack initiation
were determined by extrapolating the curves to zero crack area (actually
0.001 in.^). Figure 4 demonstrates that the specimens representing the dif-
ferent casting variables differ in their response to thermal fatigue. There are
differences in both the number of cycles for crack initiation emd in the crack
propagation rate. Furthermore, crack propagation occurs in two stages.
(This is true for all the specimens except Specimen 2 in which only Stage 1
was observed within 3000 thermal cycles.) The first stage crack propagation
is faster than the second stage. Table 2 summarizes the crack initiation and
propagation data for each of the specimens. As can be seen in this table,
based on the measured cracked area, the thermal fatigue lives, from the
point of view of crack initiation, varied from 12(X) to 2100 cycles. The first
stage crack propagation rates differed by a factor of=1.5.

Microstructural Aspects
A microstructural study was performed on the specimens before and after
the thermal fatigue test with a view to identifying the structural features
associated with crack initiation and propagation. The specimens were
ground to the midplane, polished, etched, and examined by optical
microscopy.
In general, the as-cast microstructure of Mar-M509 consists of five
phases:
1. A face-centered-cubic (fee) cobalt-base matrix.
2. An interdendritic script-like MC carbide.

p O 00 O "^
<N -^" O ~ ' —"
CO a E
22222
>n 00 VO t^ d
'^ o d o '^
OS

4> 'G ?^^??

o o o o o
•a t« 3.
<n • ^ 00 - ^ 00
f*^ - ^ «n fn fn
Q "^
<

XX : X X
«S TF ( S 00

I& f*^ ><-^

— -;
O 00
—• o

f o ooo o
X X X X X

1 f*^ f*l 00 rn vo
• ^ fn <S "*' ^*^

o o oo o
v~i trt *ri u^ v^
m Tt <s <^ -^
t-. -« (N rt -•

I R ^ ;8S
<s - ^ O
^ « (S ^ -^
fn >ri

a
a
<
Z < 5

iJi !S88i
I (S <N m '

ill! 800 S
^ ^
888
Ch O 00 00
(S ^ ^

*o Tt <N r- 00

3. Lamellar eutectic consisting of M7C3 carbide and matrix.

4. Another lamellar eutectic made up of M23C6 carbide and matrix.
5. A fine precipitate of M23C6/MC carbides around large MC carbide
particles (see Fig. 5).
A typical microstructure of the specimen after the thermal fatigue test is
shown in Fig. 6. Note that the cracks often initiated at interdendritic car-

•'V'- -'

I %
E

1. A fee cobalt-base matrix.

2. An interdendritic script-like MC carbide.
3. Lamellar eutectic consisting of M7C3 carbide and matrix.
4. Another lamellar eutectic made up of MzjCe carbide and matrix.
5. A fine precipitate of MjsCs/MC carbides around large MC carbide particles.
FIG. 5—As-cast microstructure ofMar-MS09.

bides. Very few cracks initiated at grain boundaries. The crack propagation
was predominantly transgranular and occurred in a direction normal to the
longitudinal axis of the specimen. This is illustrated in Fig. 7. Varying the
grain size from 0.5 to 2 mm did not affect either the crack initiation or
propagation as can be seen in Fig. 8.
Figure 9 shows the crack growth path after 3000 thermal cycles for
various soUdification conditions with associated variation in the secondary
dendrite arm spacings.' Two significant features emerge from this figure:
'Hereafter, the term "dendrite arm spacing" will be used to denote "secondary dendrite
arm spacing."

P^^'V/\':..

FIG. 6—Crack initiation sites in Mar-MS09.

(o) cracks grow both interdendritically and across the dendrites and (b)
overall crack length decreases with increasing dendrite arm spacing.
In an attempt to monitor the structural changes that occur during thermal
cycling, the various carbide phases were carefully examined. It was ob-
served that the large script-like MC carbides were stable during the thermal
cycling test. However, fine M^jC. and MC carbides precipitated within the
dendrite areas (Fig. 10). To determine the total amount of carbide
precipitated during the test, the matrix was dissolved electrolytically [4],
and the insoluble residue was collected, weighed, and identified by X-ray
diffraction. Table 3 summarizes the variations in the type and amount of
carbides before and after the thermal fatigue test. The as-cast microstruc-
ture consists of two MC carbide phases which differ in their lattice pa-
rameter (MC #1 > MC #2) and one M7C, carbide phase. During thermal
cycling, the metastable M7C3 phase decomposes into MiaC* and MC car-
bides. Both of these carbides also precipitate from the matrix. As a result,
for a given dendrite arm spacing, the total amount of carbide increases
during thermal cycling. Table 3 shows that there is a greater increase in total
carbide in Specimen 6 (smallest dendrite arm spacing) than Specimen 2
(coarsest dendrite arm spacing). Also note that there is a greater increase in
the amount of MC #2 carbide in Specimen 6 than in Specimen 2. It was ob-

FIG. 7—Transgranutar nature of the crack growth in Mar-MS09. The longitudinal axis of
the specimen is vertical (x50).

served that the MC #2 carbide precipitates preferentially within dendrite

areas. The significance of this phase with regard to crack propagation will
be discussed later.

Correlating the Microstructure with the Thermal Fatigue Properties

The various solidification conditions employed in this study produced
variations in grain size and dendrite spacings. No direct correlation was ob-
served between the grain size and the thermal fatigue resistance. However,
the dendrite arm spacings had a significant influence on crack initiation and
propagation behavior. This is illustrated in Figs. 11 and 12. Note that smedl
dendrite arm spacings reduce the thermal fatigue life by accelerating both
the crack initiation and the initial crack propagation rate.
A previous study of thermal fatigue behavior of Mar-M509 by Howes,
using the same thermal cycle as employed in this investigation, indicated
that the number of cycles to initiate a crack in this material was between 125
and 350 [5]. The dendrite arm spacing was estimated to be ~10 /^m from a
micrograph given in that reference. The data point is included in Fig. 11. It

I
I
fS

?1
y-s 00 a

11:
CO o ^

^
o

I
d

v..

J* .« 4.1
•n

kf..
V
*^

t
!>: i
7
(a)

-**sJL
« * ;,
^ jii. ' * * ^"

:^r:
' ? « ? •
s - "•••
» .-. /..
^-
^ .y <i

'"'* -.
V^v •><•
f «

(b)
FIG. 10—£Xfec/ of thermal cycling on the microstructure of cobalt superatloy Mar-MS09.

q *ri
en m

Ills
I c
s "!
I o
e^
<S

^ 00

t3 f*i •o
I vd >X

4> OU
>r) < N
o -S * o\ cK

I o
s
\o
d d
QO

H
<^
iL> bO
•S 'C

I 5^
T3 #
2
VO <N
d o

1 rt
J
^^
5t

u
r n r-;
•X •<t
:2

CO
«> iri r^
VO
•/^

(0
< M VO

3000

i 2000

Present Study

1000

40 60 80 100
Dendrite Arm Spacing (|;m)

FIG. 11—Correlation between the number of cycles to initiate a crack and the dendrite arm
spacing for Mar-M509.

40 60
Dendrite Arm Spacing (pun)

FIG. 12—Correlation between the initial crack propagation rate and the dendrite arm
spacing for Mar-M509.

is interesting to note that Howes' data point lies on the line extrapolated
from the present data.
The first stage crack propagation rate could also be correlated with the
amount of MC #2 carbide phase that precipitates during thermal cycling.
This correlation suggests that the effect of dendrite arm spacings on crack
growth rate could be an indirect one. No correlation was obtained between
the second stage crack propagation rate and the microstructure.

Discussion
The present study demonstrates that the response of a material to thermal
fatigue can be changed by changing its microstructure. In particular, it has
been shown that in cast cobalt-base alloy, Mar-M509, coarse dendrite arm
spacings improve the thermal fatigue resistance by delaying crack initiation
and also by reducing the crack propagation rate. It has also been shown that
thermal fatigue cracks initiate at the specimen periphery in the in-
terdendritic areas but propagate both interdendritically as well as across the
dendrites. Based on these observations, a model for thermal fatigue
cracking can be proposed as follows.
Let us consider Fig. 13. In this schematic drawing, two microstructural
conditions are depicted. Figure 13a shows a specimen with fine dendrite arm
spacing, whereas Fig. I3b refers to a specimen with coarse dendrite arm
spacing. In both cases, the cracks are shown to initiate at interdendritic
areas but to propagate both inter- as well as transdendritically.
The interdendritic regions contain a large number of script-like MC-type

Primary Dendrite

Surface

Cracks

Primary
Dendrite Arm
H' V Surface g

a. Fine Dendrite Arm Spacing b. Coarse Dendrite Arm Spacing

FIG. \3—Schematic of two microstructural conditions depicting a model for thermal fatigue
cracking in Mar-M509. Markers indicate secondary dendrite arm spacings.

carbides, (Ta, Zr) carbide, and islands of eutectic MTCs-type carbide

(CFTCS). The carbides can act as stress concentrators and provide crack
initiation sites. In addition, preferential oxidation of script-like (Zr, Ta)C
carbide [6] can promote notch formation at the periphery of the specimens.
The likelihood that microporosity will occur is also greater in the in-
ter dendritic areas than within the dendrites. The micropores could act as
stress raisers and favor crack nucleation. All these arguments suggest that
crack initiation is easier in the interdendritic areas. The probability of crack
initiation would also be higher in the specimens with fine dendrite arm
spacings (larger amount of interdendritic areas) than in the specimens with
coarse dendrite arm spacings.
Once a crack initiates at an interdendritic site, it would tend to propagate
along the interdendritic area since it contains favorable conditions for crack
growth such as the presence of brittle carbide phases. It was shown that the
cracks propagate faster in specimens with fine dendrite arm spacings than in
specimens with coarse dendrite arm spacings. The schematic drawing shown
in Fig. 13 again provides a clue to the observed behavior. If the dendrite
arms are inclined to the surface, as shown in Fig. 13, the crack propaga-
tion along the carbide network cannot continue indefinitely since the crack
is constrained to propagate in a direction approximately normal to the
longitudinal axis of the specimen along which the stresses operate. The
crack would therefore have to change direction and travel across the den-
drite arms. This would imply that, in the specimen with large dendrite
arm spacings (Fig. 13), the crack has to travel longer distances across
the dendrites before meeting the adjacent carbide network. The rate at
which the crack grows across the dendrites would obviously depend on the
microstructural features within the dendrite areas. Table 3 shows that
precipitation of MC and M23C6 carbides occurs during thermal cycUng. Of
these, the amount of M^jCs and MC #1 carbides that precipitate during ther-
mal cycling is nearly independent of dendrite spacings; however, a smaller
amount of MC #2 carbide precipitates in the specimen with coarse dendrite
arm spacing (Specimen 2) than in the fine dendrite arm spacing specimen
(Specimen 6). The MC carbides generally precipitate along crystallographic
directions within the dendrite areas [7,8]. When aligned favorably, these
carbides could provide easy paths for crack propagation. The smaller
amount of MC #2 carbide observed in the coarse dendrite arm spacing
specimen would therefore provide a less continuous path for crack
propagation than the carbides in the fine dendrite arm spacing specimen.
This observation, coupled with the fact that the crack has to travel longer
distances across the coarse dendrites, would imply that the crack
propagation would be slower in the coarse dendrite arm spacing specimen.
The experimental evidence for this behavior C£in be obtained in Fig. 9 which
demonstrates that, for the same number of thermal cycles (3000 cycles), the
crack is shorter in the specimen with coarse dendrite arm spacing than in the
specimen with fine dendrite arm spacing.

Conclusions
1. The thermal fatigue resistance for a given heat of cast Mar-M509 can
be improved by employing solidification conditions that result in coarse
dendrite arm spacings.
2. This improvement occurs in both the number of cycles to initiate a
crack as well as the rate of crack propagation.
3. The carbides at the interdendritic areas act as preferred crack
nucleation sites and also provide easy paths for crack propagation.
4. The smaller number of interdendritic sites as well as a lower amount
of MC #2 carbide precipitated within the dendrite areas during thermal
cycling can explain the superior thermal fatigue resistance of the speci-
mens with coarse dendrite arm spacings.

Acknowledgment
The assistance of P. M. Yuzawich in preparing metallographic specimens
is gratefully acknowledged.

References
[1] Howes, M. A. in Fatigue at Elevated Temperature, ASTMSTP520, American Society for
Testing and Materials, 1973, pp. 242-254.
[2] Bizon, P. T. and Spera, D. A., "Comparative Thermeil Fatigue Resistances of Twenty-Six
Nickel- and Cobalt-Base Alloys," NASA TN D-8071, National Aeronautics and Space
Administration, 1975.
[3] Glenny, E. et al. Journal of the Institute of Metals, Vol. 84,1958-1959, p. 294.
14] Donachie, M. J., Jr., and Kriege, O. H., Journal of Materials, Vol. 7, No. 3, Sept. 1972,
pp. 269-278.
[5] Howes, M. A. H., NASA Report CR-121212, National Aeronautics and Space Ad-
ministration, 1973.
[6] Woulds, M. J. and Cass, T. R., Cobalt 42, March 1969, pp. 3-13.
[7] Sims, C. T., "The Superalloys—Chapter 5," C. T. Sims and W. C. Hagel, Eds., Wiley,
New York, 1972.
[8] Drapier, J. M., Leroy, V., Dupont, C , Coutsouradis, D., and Habraken, L., "Structural
Stability of Mar M-509—A Cobalt-Base Superalloy," International Symposium on Struc-
tural Stability in Superalloys, Seven Springs, Pa., 4-6 Sept. 1968, pp. 436-459.

Some Thermal Fatigue

Characteristics of Mild
Steel for Heat Exchangers

REFERENCE: Laub, J. S., "Some Thermal Fatigue Characteristtcs of Mild Steel for
Heat Exchangers," Thermal Fatigue of Materials and Components, ASTM STP 612,
D. A. Spera and D. F. Mowbray, Eds., American Society for Testing and Materials,
1976, pp. 141-156.

ABSTRACT: Thermal fatigue of mild steel has received little attention until recently
even though this material is used extensively in heat exchangers for gas-fired appUances.
A survey of the thermal fatigue problem of thin shell mild steel structures subjected to
arbitrary temperature fields is given. Data from end-constrained thermally cycled test
specimens is also presented. These data have been validated through use in developing
production heat exchangers.

KEY WORDS: thermal fatigue, heat exchangers, steels, fatigue (materials)

Nomenclature
A£, Total strain range, elastic and plastic
AT Temperature range
T„ Mean cyclic temp, T^^ + T^Jl
R Strain ratio £n,in/£max
a WeibuU location parameter
c Weibull shape parameter
M£ Microstrain, 10"* unit length/unit length

Ordinary low-carbon mild steels are not normally considered to be high-

temperature structural materials; however, large quantities of 16 to 20-gage
(1.6 to 0.9-mm-thick) sheet is used every year to make gas-fired appliance
heat exchangers. These devices usually take the form of an irregular thin
shell, often with a plane of symmetry coincident with flanges used to weld
two drawn halves together. Figure 1 shows a typical heat exchanger cell
' Senior engineer. Research Division, Carrier Corp., Syracuse, N. Y. 13201.

141

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
142 THERMAL FATIGUE OF MATERIALS AND COMPONENTS

FIG. 1—/4 typical gas-fired heat exchanger cell.

fired by a gas burner inserted into the lower opening to a thermal rating
of 33 000 Btu/h. The upper opening is attached to a flue collector box
which transfers the flue gas of this and an arbitrary number of other cells
to the stack. Residential and commercial gas furnaces have multiple num-
bers of these units determined by their capacity.
Heated internally by a gas flame, the cells are cooled by room air cir-
culated in most cases by a fan over their outside surfaces. Internal area
variations between adjacent cell surfaces or an outer cell surface and the
furnace casing produce variable and arbitrary air flow patterns which com-
bine with the flame thermal pattern to give a complex and time-varying tem-
perature pattern in the material. During typical furnace operation, any
given point in the cell will be subjected to a temperature cycle determined by
the operational functions plus the timing and duration of the flame and cir-
culating fan. The net result is that the thin shell of an appliance heat ex-
changer presents one of the most difficult and challenging thermal strain
and thermal fatigue problems to be found in engineering practice.
Economic considerations dictate that any new design must be well proven

against thermal fatigue, but, until recently, the only accepted method for
doing this was the laborious, time consuming, imprecise, and inefficient life
test progrsim. Even life testing is subject to a wide divergence of opinion
since any given exchanger may be used in a multiplicity of appliance types
(upflow, downflow, crossflow, space heaters, and combination heating
and cooling units); and each installation has its own demand pecuHarities
which determine, to some extent, the cycle and temperatures encountered.
Further, by the time a life test program is complete, changing conditions
have often dictated new and previously unforeseen uses necessitating a
new, long-term test program. Although a component may cost only a few
dollars to manufacture, it will be made by the tens or hundreds of thou-
sands per year which necessitates a quicker and more efficient method of
proving structural integrity.
Experimental techniques for thermal strain analysis on thin shell struc-
tures, while difficult, have been developed in the author's laboratory to the
point of providing part of the thermal fatigue problem answer. Over a
period of several years, conventional experimental stress analysis methods
have been combined, refined, and adapted so that acceptably accurate
strain measurements can be made with readily available equipment.
Ceramic brittle coatings give initial external surface tensile strain patterns
above a threshold strain varying with the local temperature but ranging nor-
mally between 600 and 1200 fit. Medium temperature foil strain gages with
self-temperature compensation can be used in locations of the shell which
do not exceed 650 °F (343 °C). The manufacturer's apparent strain versus
temperature correction curve can be incorporated in the data reduction for
improved accuracy. For areas of the shell which exceed this temperature,
strain measurements require high temperature, wire strain gages applied
with flame-sprayed molten ceramic along with special temperature com-
pensation techniques and data reduction programs.
Acceptable strain measurement is one necessary part of the thermal
fatigue-reliability problem; the second is an accurate knowledge of the ther-
mal fatigue behavior of the exchanger material. Until recently, very little in-
formation was available for ordinary low-carbon or mild steel under con-
ditions of both varying strain £md temperature. Some work was published
on strain cycUng at constant elevated temperatures [1-3],^ for ex£unple, but
this did not really come to grips with heat exchanger operation where strain
changes in concert with temperature. As a consequence, thermal fatigue
tests were set up on specially constructed equipment following Coffin's
earUer work [4-7]. This paper presents the results of these tests. The ac-
curacy of the data has been estabUshed by successfully using it in designing
and qualifying rehable heat exchangers.
Recently, some new work has been pubUshed by Jaske [8] which confirms

^ The italic numbers in brackets refer to the list of references appended to this paper.

the general accuracy of these data. His data were generated on equipment
allowing constant strain range and ratios and, as such, is more amenable to
conventional analysis. The data presented here were generated on much
simpler equipment and, consequently, required the proof of time and use.
Fortunately, the many uncontrollable and variable factors involved seem to
be similar for specimen testing, furnace life testing, and furnace field use.
Heat exchangers with reliability proven by service have been produced using
these data and the related experimental techniques, while other designs with
excessive strain levels were identified early in the development period and
corrected or discarded.

Procedure
Completely end-constrained axially loaded plain and notched thin-wall
tubes comprised the thermal fatigue specimens in this test. Figure 2 shows
the details of the test section. Large brass adapters were silver soldered into

THRU BOTH WALLS

6.4" COLO DRAWN SEAMLESS

HYDRAULIC TUBE AlSI 1008
FULLY ANNEALED

CHEMISTRY
CARBON 0.05-0.15
MANGANESE 0.30-0.60
PHOSPHORUS 0.040 MAX.
SULPHUR 0.050 MAX.

25.4 DIA.-»

1.2
STANDARD MODIFIED
AREA 94.20inm^ HIGH STRAIN
LOW TEMR
AREA 64.52mm^

FIG 2—Specimen geometry and material composition.

each end, providing alignment and gripping. These adapters were water
cooled to give a constant temperature heat sink at each end of the specimen.
They were also drilled axially to allow cooling air to be blown through the
tube interior during the cooling portion of the thermal cycle.

A thermal fatigue cycle consists of several more or less separate evems

which may interact in different ways. We define the initial portion as
heating, the next as hot hold time, then cooling, and finally, cold hold time.
Each portion has a set time associated with it. In the present data, both hold
times were 5 s for all tests. Heating and cooling times were functions of the
temperature range, electric heating power, cooling air supply pressure, and
cooling air passage area. The latter three were held constant, allowing the
heating and cooling times to vary somewhat through each mean tem-
perature series.
The heating portion of the cycle raises the specimen gage length tem-
perature from the minimum to its maximum value. Tensile and compressive
strains and, consequently, stresses accompany the temperature change in
actual heat exchangers. With axial, end-constraint test equipment, the
heating portion necessarily causes compressive strain in the specimen.
Hot hold time affected the specimen fatigue Hfe by allowing compressive
stress relaxation. To stabilize this effect in the present data, it was limited to
5 s. This time allowed some redistribution of temperature along the
specimen, broadening the high temperature length at the center and raising
the temperature profile of the rest of the specimen. The maximum com-
pressive strain was reached at the end of the hot hold time as the maximum
tensile strain was similarly reached at the end of the cold hold time. A
typical temperature profile at these extremes is shown in Fig. 3. The total
thermal strain range, Atr, may be readily determined from the area between
these curves using some simplifying assumptions about the material
behavior.

1000
1
Tm»600*F END OF HOT HOLt
aT=60< 3*F
800

• \ ^
600

<
400
END OF C}LD
< HOLD

200

^
BOTTOM 0 1 TOP
-X.in.—4 -I 0 I 4-•X.in.
SPECIMEN LENGTH

FIG. 3—Specimen temperature profile.

Cooling followed the hot hold time. In these data, the cold hold time
power was switched on at the same time as the cooling air. Room tem-
perature air was blown over the outside and through the inside of the
specimen until the low cycle temperature value was reached. At this point,
the air was switched off and only the cold hold time power applied to the
specimen. Redistribution of the low temperature followed for 5 s, allowing
the tensile thermal strain and stress to reach their maximum values. This
completed one thermal fatigue cycle.
Major restrictions with a test of this type are that compressive strain oc-
curs only by heating and tension by cooling; in order to vary the amount of
thermal strain available from a given temperature cycle, the specimen
geometry must be changed; and also, the strain ratio, R, is not controlled.
For these reasons, the data are presented on the basis of total cyclic strain
range, At,. R values generally fall between 0 and - 0.5.
On the other hand, the completely constrained test equipment is simple
and rugged. The concept was successful enough that other materials were
tested on a second similar machine. The total cost of the equipment at the
time it was built was approximately $5000.
Resistive heating was provided by a large transformer, the secondary coil
of which consisted of 1 Yi turns of heavy copper bar. Each end of the bar ex-
tended out to provide the connection to the ends of the specimen. A typical
electric power level during the heating portion of the cycle was 2500 A at
1.5 V. This power setting resulted in a heating rate at the center of the
plain tube specimen of 35 °F (20°C)/s. A wide range of control was pro-
vided for the transformer so that heating rates could be varied according
to the heating time required for a given test series and also to provide for
other specimen sections and sizes.
The end-constraint fixture consisted of an upper and lower platen which
gripped the specimen end adapters. Massive threaded tie bars and nuts with
faces ground parallel to the platen faces positioned the platens. Electrical
isolation between the holding fixture ends was accomplished by a set of
bakelite shoulder washers between the top platen positioning nuts and the
platen.
Specimens were loaded into the fixture platens but not immediately con-
strained. Several check and adjustment cycles could be run before actually
straining the specimen. These were used to obtain the total displacement of
the specimen and to make power adjustments when the thermal cycle was
changed. The movable platen was always constrained at the end of the hot
hold time so that the initial part of the strain cycle was tensile.
After constraint was imposed, no further attention was required until
machine shutdown indicated specimen failure or, on rare occasions, failure
of some component of the test equipment. Continuous temperature record-
ing at the center of the gage length provided a record for each specimen
which could be referred to at the end of the test. Any unusual temperature

excursion shut the equipment down. Excessive temperature in the gage

length indicated a crack concentrating the current flow. This provided a
shutdown signal, and each fatigue failure recorded here consisted of a crack
from 12 to 18 mm in total length. The number of cycles between tem-
perature increase (a noticeable crack) and shutdown varied by no more than
50 cycles and was not considered significant.
Cyclic thermal strain was measured by a noncontacting electro-optical
tracker. This instrument used a photomultiplier tube to electronically servo
on an optical discontinuity. The optical discontinuity in this case was a light
field between dark boundaries produced by two back-lit metal flags at-
tached to pins spot welded to the specimen sides at the gage length extremes.
A differential ampUfier processed the output signal from the photo-
multiplier tube giving a voltage proportional to the distance moved by
the pins. The total thermal cycUc strain was then the unconstrained
displacement of the flags minus the constrained displacement. Single
specimens were set up and run through each temperature range at each of
the mean cycle temperatures. The total strain range values given with the
data are average values obtained after stable cycling had been underway for
some time. The measured values were checked against values calculated
from specimens which had a series of thermocouples providing the LT
and profile data similar to the example in Fig. 3. The results of each
means of determining thermal strain were comparable within about
±200 fit, a result considered adequate for the available equipment. The
absolute accuracy of the electro-optical device was about ± 125 jit. Align-
ment changes of the tracker targets during cycling accounted for some ad-
ditional inaccuracy of the measured values.

Results
Each data point on Fig. 4 consists of the average value of from four to six
specimens tested under identical mean temperature and temperature range
conditions. All of these data have 5-s hot and cold hold times, while the
heating and cooling times vary. On the average, cychc rate is about 45
cycles/h. At long lives, data are accumulated slowly. This accounts for the
scarcity of data beyond 10 000 cycles. Since a major effort was required to
obtain the data, a minimum statistical sample at each point was considered
necessary.
The data of Fig. 4 covers the area of greatest interest for furnace heat ex-
changers. While by no means complete, enough of the strain-life field is
covered to show general trends and conditions to be avoided if medium-
to long-term thermal fatigue reliability is to be achieved. Heat exchangers
are warranted for 10 to 15 years of life. From experience, 100 000 cycles
is considered an adequate design goal in most cases, so part of the de-
signer's task is to preclude low-cycle fatigue. For cooler areas of the ex-

u 0.51
^"-^
I 0.4, b.r^?^i=^^i=5[q
°= 0.3
O 0 4O0'F(2O5'C) "^^ B
K 0.2 "~-Tfc-
Q D 500rF(260'C)
• • 550'F(288'C)
^ — - A 600'F(3I6"0)
^ 0.1
1,000 apOO 5,000 10,000 20POO 50,000 100,000
N, .AVERAGE CYCLES TO FRACTURE

FIG. 4—Thermal fatigue data for low-carbon steel.

changer where the mean cycle temperature is 400 °F (260 °C) or below, an
upper limit of 0.2 percent strain is acceptable. In hotter areas, this must
be reduced to 0.1 percent or less.
As would be expected, thermal fatigue strength is greater at 400°F
(205 °C) than at any higher mean temperature. This corresponds to the
highest tensile strength for mild steels. It was necessary to use the second
(reduced cross section) type of specimen shown in Fig. 2 for these and the
450°F (232 °C) tests since, for plain tube specimens, fatigue cracks oc-
curred at the cool ends under these conditions.
All curves tend to converge in the high strain-short life area where the
strain range dominates the other peuameters. Here, the cyclic temperature
has less influence than at lower strains where temperature and time become
stronger factors. No fatigue limit indications are found in this data
although, under certain conditions, they may exist. All data shown here are
for material which goes well into the plastic range at the cycle extremes. The
change in slope between 500 °F (260 °C) and 550 "F (288 °C) may be due to
material characteristics, or it may be fortuitous. Further testing is needed
to clarify this situation.
Approximately 60 specimen tests make up the curves of Fig. 4. The
mean life value is presented without probability bounds since the sample
size at any given condition is too small to allow reasonably accurate ex-
treme values to be projected. In any case, there is so much uncertainty in
other parts of the fatigue analysis that more precision in these data
would be of doubtful value. As noted in the discussion section, use of
the data against full-scale furnace testing has shown the accuracy to be
completely adequate.
Figure 5 is an example of the way the data were treated. It was found in
some cases that the data fit the logarithmic normal cumulative distribution
function somewhat better than the WeibuU distribution. In other test
sequences, the two- or three-parameter (including the position parameter)
WeibuU distribution provided a better fit by the criterion of minimum
variance about the regression line. The ordered data for the 600 °F (316°C)
tests are shown by small dots. Log-normal population parameter estimator

1 J. '

99.5
1 i^,\ ^ '^ 9 0 % CONFIDENCE
1 . <& ! BAND
99 f 1
1
1
/ / /
f
i 1 /(
/ / /» T
70 / / / / _4
/ ' :? V y ' ^
/I +'
U) 50 _^^ -« —4
i 40 ' P"
i 30 ,/
u- 20 ;
it 15.9
10
/ 1
n
)
5 / «
2
/ /1
T„=600 r
/ /
/ / '7
0.2 / / 7j
/ 3 4 5 6 7 81000 2 3 4 5 6 7SI0p00 3 4 5 6 7 OOPOO
CYCLES TO FAILURE

FIG. 5—Log-normal probability offailure with various temperature ranges at 600°F mean
temperature. Nonparametric lower 90 percent confidence band shown.

lines found by simple linear regression (least squares) are drawn through
these points. The median and one standard deviation below the median
value are noted by larger circled points. Population mean estimate points
are shown by the crosses denoted X. Because of the small sample sizes, it
was not considered appropriate to estimate the population extreme value
statistics by using an assumed distribution. Lower 90 percent confidence
bands (5 percent) obtained by the Kolmogorov-Smirnov nonparametric
procedure are shown. Statistical procedures just noted are all standard prac-
tice. The details of their use may be found in many texts; Refs 9 and 10
are examples.
For the mean life value, suitable confidence bands are available directly
from the log-normal Unes. Figure 6 is typical and shows the data of Fig. 5 as
they appear in the composite data, Fig. 4, but with the addition of a lower
90 percent confidence band on the mean. A good idea of the mean fatigue
life variability of the material can be obtained from this plot. Data from
other mean temperatures is similar, showing generally greater variability as
life increases at lower strain levels. The host of statistical time-dependent
processes increases the data spread as strain severity decreases and testing
time increases.

Discussion
The data presented here do not claim to give a complete picture of the
behavior of mild steel under thermal fatigue conditions. Thermal fatigue is

tf IJ3

0.5 * ~ ^ _ ^~^
ST
0.4
~> /MEAN
0.3 r*--
0.2 ^ LOWER 90%
~ CONFIDENCE
BAND 600 ' F T „
2
o
0.1
500 1,000 2,000 5,000 10,000 20,000 50,000
Nf,AVERAGE CYCLES TO FRACTURE

FIG. 6—A verage life as a function of strain range at 600°F (315 °C) mean temperature.

the result of such a complex set of phenomena that considerable reservation

was felt about the usefulness of the data until they were proved in practice.
The data in conjunction with experimental strain measurement techniques
were tested by use in prediction of failures and failure-free designs. While
every effort was made in designing the fatigue test to duplicate as nearly as
possible the actual material environment, compromises necessarily placed
restrictions on the acceptability of the results. Some of the crucial areas
ignored by these tests are now being investigated; among these are the ef-
fects of the strain ratio, R. Other materials and coatings are being fatigue
tested on more advanced equipment which provides variable strainrange in
concert with temperature changes while controUing the strain ratio. The
results of these tests may be available in the near future.
Since, until recently, there was little to compare these data to, empirical
experience was required to establish their validity. Initially, comparison of
test fractures to those found in heat exchangers confirmed that the fracture
morphologies were identical. Figure 7 details typical test failures. Figure la
shows some typical cracks developed on these tests. Once a macrocrack had
formed, the area for current flow was reduced, concentrating heating in the
material in the same plane as the crack. Depending on the location of the
thermocouple with respect to the initial crack, there was some variation in
the total length of the crack which could be considered as strictly due to
thermal fatigue under the test conditions. This variation in cycle number
was small in terms of the test Ufe and has been disregarded in the results.
Figure lb shows a low power microscopic view of the fracture surface
morphology. This appearance is identical with thermal fatigue failures in
actual heat exchangers. In areas near the initiation site, the appearance is
generally planar, but with variously rounded (oxidized) steps. Macrocracks
are formed by coalescence of the small "heat checks" joining together as
they reach a critical size by longitudinal shear steps.
A typical family of heat checks is seen in the metallographic section of

mm
JffiW

(a) Typical failures showing buckle in high temperature area at crack tip.
{b) Planar area of flat thermal fatigue (x 6).
FIG. 1—Detail of thermalfatigue test failures.

Fig. Ic. The through crack is at the left, and a series of more or less uni-
formly spaced surface cracks or checks occur along the top edge, the
outer tube surface in this case. These cracks have formed during the cool-

(c) Metallographic cross section adjacent to the through crack ('^ x 100).
(d) Detail of intermediate length crack of Fig. 7c showing oxidation (~ x 400).
FIG. 1—Continued.

ing (tension) part of the cycle. Bulging in the area of the cracks occurs
only after the crack has developed to the point of concentrating current

flow and heat at its ends. Oxidation of the fracture surfaces is a uni-
versal feature, as is transgranular cracking. Experience indicates that, if
intergranular cracking is found in mild steel exchangers, some other fail-
ure mechanism such as corrosion or creep rupture is at work.
Figure Id shows the "oxide wedge" of the intermediate length crack of
Fig. 7c. Because the oxidation product has a greater volume than the
original material, there is a wedging action on the crack tip. In these tests,
there is a strong compressive load during heating which attempts to close
the cracks. After the oxide wedges have formed, the cracks cannot close,
and the effect is to shift R toward pulsating tension at the crack tip. It
follows that crack growth is not a simple function of the imposed strain
cycle. Further, the numerous cracks formed in the relatively uniformly
strained gage area interact until the time one crack is clearly dominant. At
this point in the tests, failure is only a few cycles away, so there is a gage
length size effect determined by the temperature profile which defines the
critically strained area. All of these factors influence the fatigue life of the
specimen, but they are also involved in actual exchangers, so there are com-
mon qualitative effects which are more or less equal in both situations.
Estimation of the fatigue performance of numerous heat exchangers over
the past five years has shown that the fatigue data hold up very well. Con-
sidering that the accuracy band of experimental strain measurements on
operating heat exchangers is fairly broad, great refinement of the thermal
fatigue data is not necessary. For strain measurements to about 650 °F
(343 °C), medium temperature foil strain gages give results with an error
band about ± 100 ME wide. High temperature wire gages with special tem-
perature compensation techniques are used from 650 to 1000 °F (343 to
530 °C) with a considerably wider error band estimated to be about ±300
[ii. on the average. Combination of experimental measurements with the
fatigue data to predict the response of several new heat exchangers, as well
as existing exchangers used in new applications, has estabUshed confidence
in both technical areas. Where strain and temperature measurements fall
within the limited hfe region of the data, failures Eire encountered, generally
at lives well within the statistical scatter expected. Usually in furnace Ufe
testing, failures £ire scattered over a much broader life range than for most
machine components because of the unavoidable vagaries of burner flame
and cooUng air flow patterns. Even for very severe strain conditions, the
failure rate in life test exchangers rarely exceeds 50 percent within
reasonable test times. These statistics are described by positively skew
Weibull probability of failure density functions (shape parameter 1.0 < c
< 3.5).
Several very reliable heat exchanger designs have been developed with the
help of the data. Furnaces have been manufactured for more than five years
which were given a clean bill of health by linear extrapolation as noted in
the Results section. Lack of field failures in combination with complete test

series survivals to 100 000 cycles has built confidence and established a
sound basis for using these data. Note that field results are a necessary part
of the qualification of the thermal fatigue analysis because direct
correlation of life testing to operational use is not available.
The most limiting features of completely constrained testing are the lack
of a constant compression-tension strain ratio and the difficulty of
changing the total strain range without changing the temperature range.
New, more sophisticated equipment has overcome these problems to a great
extent by providing strain control in a closed-loop feedback circuit. Tests
now in progress use a combination of strain and temperature control
providing more tractable, conventional low-cycle data on other materials of
interest. Because of previously noted shortcomings, no attempt has been
made to generalize the data or extract universal trends. There is a danger in
this which applies to even the most closely controlled laboratory data
because mild steel behaves in a nonhomogeneous, nonlinear manner under
high strain and temperature cycling. Even when a short gage length ex-
tensometer is used, the average strain measured is only an indication of the
section gross behavior once the plastic strain range is entered.
An example of nonhomogeneous behavior is shown in Fig. 8, a specimen

'1 * ( - -•ai^^'isi^jj*

FIG. »—Maximum thermal strain specimen, 800°F (445°C) AT at dOO'F (316°C) T„

A£,„„, 0.63 percent.

subjected to an above normal temperature range, 800°F (445 °C) AT at a

mean temperature of 600 °F (316°C). The oxide scale formed above about
900 °F (482 °C) provides a visible indication of the plastic behavior not
ordinarily available since furnace heat exchangers and hence the test data
are restricted to operation below the oxide formation temperature [7/]. A
slip band was formed during each cycle at arbitrary locations within the
maximum temperature zone. Strain hardening along the slip band pre-

vented further yielding of this material for a number of cycles. Under

these cyclic conditions, strain aging is nearly complete within the time of
one cycle, so the upper yield point returned for the yielded zone by the
time the next cycle's yield load was reached, and, as can be seen in the
figure, the damage tends to accumulate in the central 25 mm of the speci-
men. Note that the macrocrack progressed in the direction transverse to
the principal tensile stress, directionally unaffected by the slip. The area
of strain concentration at the crack tip has shed its oxide scale as a result
of the greatly increased strain in this zone.

Conclusions
The data presented in this paper provide one of the engineering elements
needed to eliminate thermal fatigue in gas appliance heat exchangers.
Another necessary ingredient is a reasonably accurate experimental strain
measurement technique which may be applied to operating exchanger
shells. Development of a strain measurement procedure has necessarily been
carried on concurrently with generation of fatigue data. Although mea-
suring quasistatic strains on thin shell structures at up to 1000°F (538 °C)
presents great difficulties, techniques have been evolved which give ex-
cellent predictive results in combination with the data. Prediction of failure
or survival under any normal operating conditions has been assured with
confidence through actual operating experience.
Application of these fatigue design techniques still requires a very high
level of technical and manual skill. High temperature measurements are
time consuming and relatively expensive when compared to the cost of the
heat exchanger element and normal experimental strain analysis. Further,
actual exchangers, preferably from production tooling, are necessary to do
the fatigue analysis. This means that the design is well along, and options
for change are limited and expensive by the time fatigue life prediction can
be made.
Finite element methods (FEM) have given hope of accelerating the design
of reliable heat exchangers by providing an analytical strain field solution
before tooling has been set up. This would be ideal because contours could
be changed in the computer model with minimal time and expense. To date,
however, FEM results obtained by the author's group during heat ex-
changer development have been disappointing, if not downright misleading.
The cause of this is that, to date, only a linear-elastic FEM code has been
used. Except for very preliminary indications of problems (where the stress
values exceed the local yield strength) this approach is not valid since critical
areas of the shells deform plastically and, in most cases, are entering
displacement controlled buckling at about the same time. Nonlineeu
material and geometry properties along with large deflection capabilities are
required in the finite element analysis. It remains to be seen if the nonlinear.

large displacement programs which are available will provide an acceptable

addition to the experimental procedures.

References
[/] Mackenzie, C. T. and Benham, P. P., "Push-Pull Low Endurance Fatigue of En 25
and En 32B Steels at 20 °C and 450 °C," Proceedings }965-J966, The Institution of
Mechanical Engineers, Vol. 180, Part I, No. 30, London.
[2] Mackenzie, C. T., Burns, and Benham, P. P., "A Comparison of Uniaxial and Biaxial
Low-Endurance Fatigue Behavior of Two Steels," Proceedings 1965-1966, The In-
stitution of Mechanical Engineers, Vol. 180, Part 31, London.
[i] Benham, P. P., "High-Temperature Low-Cycle Fatigue: Survey of British Work," Paper
presented at 1967 Society for Experimental Stress Analysis, Spring Meeting, Ottawa,
Ont., Canada.
[4\ Coffin, L. F., Jr., and Wesley, R. P., "Apparatus for Study of Effects of CycUc Ther-
mal Stresses on Ductile Metals," Transactions, American Society of Mechanical En-
gineers, Aug. 1954.
[5] Coffin, L. F., Jr., "A Study of the Effects of Cyclic Thermal Stresses on a Ductile
Metal," Transactions, American Society of Mechanical Engineers, Aug. 1954.
[6\ Coffin, L. F., Jr., "An Investigation of Thermal-Stress Fatigue as Related to High-
Temperature Piping Flexibility," Transactions, American Society of Mechanical
Engineers, Oct. 1957.
[7] Coffin, L. F., Jr., "Design Aspects of High-Temperature Fatigue With Particular
Reference to Thermal Stresses," Transactions, American Society of Mechanical
Engineers, April 1956.
[<S] Jaske, C. E. and Porfilio, T. L., "Thermal-Mechanical Low Cycle Fatigue of AISI 1010
Steel, "Catalog Number H54375, American Gas Association, Cleveland, Ohio, 1975.
[9] Hald, A., Statistical Theory with Engineering Applications, Wiley, New York, 1952.
[10] Johnson, N. L. and Leone, F. C , Statistics and Experimental Design in Engineering
and the Physical Sciences, Vol. I, Wiley, New York, 1964.
[77] USA Standard for Gas Fired Gravity and Forced Air Central Furnaces, ANSI Z21.47-
1968, American Gas Association, Cleveland, Ohio, 1968.

Low-Cycle Thermal Mechanical

Fatigue Testing

REFERENCE: Hopkins, S. W., "Low-Cycle Thermal Mechanical Fatigue Testing,"

Thermal Fatigue of Materials and Components, ASTM STP 612, D. A. Spera and
D. F. Mowbray, Eds., American Society for Testing and Materials, 1976, pp. 157-169.

ABSTRACT: A test method is described for performing uniaxial low-cycle thermal

mechanical fatigue tests at elevated temperatures on nickel-base superalloys. Both
smooth specimen crack itiitiation and notched specimen crack propagation testing is
discussed. Standard closed loop servohydraulic testing machines with closed loop elec-
tric induction heaters are utilized. The temperature and mechanical strain are in-
dependently controlled, blowing for a wide range of relationships between temperature
and mechanical strain.
Tubular specimens are directly inductively heated in air and cooled with compressed
air. The temperature is measured with a radiation pyrometer which eliminates ther-
mocouples and premature failures caused by welded thermocouples. Axial strain in the
specimen is measured from machined internal ridges on the specimen utilizing a linear
variable differential transformer and a quartz internal extensometer.
An electronic system compensates for the thermal expansion component of strain for
both the control loop and the recorders. This system permits direct recording of the
load-induced mechanical strain. Load-strain hysteresis loops and strain-temperature
plots can simply be recorded during the test.

KEY WORDS: thermal fatigue, mechanical tests, nickel alloys, crack initiation, crack
propagation

Thermal fatigue has been defined by Garden [ly as that process by which
cracks originate and are propagated in materials that have both cyclic tem-
perature and cyclic mechanicsd strain at a critical location. There are essen-
tially two experimental techniques used to obtain such material data. The
conventional thermal fatigue (TF) technique entails cycling the test
specimen between a hot source and a cold source. With this method, it is the
specimen geometry, the heating rate, and the temperature difference which
produce the cyclic mechanical strains. The resultant mechanic2d strains and
stresses have to be calculated analytically. For anisotropic materials, such as
'Presently, mechanical engineer. Failure Analysis Associates, Palo Alto, Calif. 94304;
formerly, senior materials engineer. Materials Engineering and Research Laboratory, Pratt
& Whitney Aircraft, Middletown, Conn. 06457.
' The italic numbers in brackets refer to the list of references appended to this paper.
157

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
158 THERMAL FATIGUE OF MATERIALS AND COMPONENTS

gas turbine blade materials, this calculation is extremely complex. The other
experimental technique, known as thermal mechanical fatigue (TMF),
cycles both the specimen's mechanical strain and its temperature in-
dependently. It is this second technique and equipment that will be presen-
ted in this paper, with emphasis on those items or procedures which differ
from isothermal low-cycle fatigue (LCF) testing.
Before proceeding, definitions of the different types of strains are in or-
der. Thermal strain is defined as a change of specimen dimension caused by
thermal expansion at zero load (free thermal expansion) divided by its
original dimension. Mechanical strain is defined as that strain which is
caused by or which generates stress. Total strain, which is the displacement
measured from the specimen divided by the original gage length, is defined
as the algebraic sum of both the thermal strain and the mechanical strain.

Experimental Equipment and Procedures

Testing Machine
Presently, there are four TMF machines operating around the clock in the
Mechanical Behavior Laboratory of Pratt & Whitney Aircraft. All of these
machines are standard closed-loop, servohydraulic systems and are equip-
ped with load, strain, and temperature controllers, along with strain and
temperature programmers. One of these systems is shown in Fig. 1. For the
most part, this equipment can be purchased from any of the closed-loop
fatigue machine manufacturers. These machines were chosen over the com-
mercial screw driven machines because of their ability to control total strain
rate and because of their response time. For fast temperature response and
good control, a completely electronic closed-loop temperature controller is
used which has much better dynamic control than the old-fashioned elec-
trical-mechanical temperature controller that nulls by a slide wire system.
The details of how the strains and temperatures are controlled, measured,
and programmed will be discussed later.

Loading System and Specimen

The specimen ends and grips used for this testing were selected because
they satisfy two major concerns: the grips are stiff enough to maintain
alignment in both tension and compression; and the specimen is easily
removable after prolonged exposure to high temperatures. This design was
originally used by Wells [2], and the deteiils are shown in Fig. 2 with a solid
isothermal LCF specimen assembled in one grip.
The specimen design currently in use is shown in Fig. 3. It has the same
end design that Wells [2] used for his solid specimens. This tubular design
with internal ridges to measure strains was selected for the following

FIG. 1—Typical closed-loop servohydraulic fatigue machine with a 10 kilohertz induction

heater and a pyrometerfor temperature measurement usedfor TMF testing.

reasons. The thin wall allows for fast heating and cooUng cycles with a
minimum temperature gradient across the wall thickness. The wall thick-
nesses can be varied to model a component wall thickness. The strains are
measured on the centerline of the specimen which mechanically averages the
strains as Tishler and Wells [3] have shown, and the specimen is suitable for
both crack initiation and crack propagation testing. The alignment of the
loading system is checked and maintained to keep the bending strains within

-PULL ROD

-I!<-I8 IIEF-21IID.

iSc^:

-COLLAR
~
at -SPLIT RIHOS

I^ r .027

-SPECIMEN
2.4S« 3000ia.
1.000 y -050 R

I. ] N 0<0 R
>y i \,,
(-.901 'Oio.-j

FIG. 2—Solid isothermal LCF specimen with one end assembled in the standard tension-
compression gripping system.

FIG. 3—Strain control tubular fatigue specimen showing the internal ridges and the quartz
rod and tube extensometry system.

the specimen to less than 3 percent of the appHed axial strain. At small
strains where percent bending is meaningless for a stiff gripping system,
the bending strains on the specimen's surface are maintained below 10 '*.

Extensometry
To minimize the imposed strain errors and measure only displacements in

the gage section of the specimen, fused quartz was chosen for the ex-
tensometer material. Fused quartz has a low coefficient of thermal ex-
pansion and a high melting temperature, making it ideal for this applica-
tion. Fig. 4 is a sectional view of the specimen and extensometry. The
displacements are measured between the two internal ridges by the quartz
rod and tube which are extensions to the Unear variable differential trans-
former (LVDT). There are two high temperature springs: one on the bot-
tom to hold the quartz rod firmly against the lower ridge and one on the top
to hold the quartz tube against the upper ridge. Each spring applies less than
5 N force to the ridges. With this technique, the displacements measured
reflect only the relative movements of the two ridges.

MYBirAULIC 1*11

FIG. 4—Sectional view of the loading train with the quartz extensometry installed and the
collars and split rings removed.

Load Cell
The loads are measured from a standard load cell in series with the
specimen. The cell can be mounted in one of two locations: either above the
specimen, attached to the frame, or below the specimen, attached to the
hydraulic ram. There are two advantages to mounting the load cell to the
hydrauUc ram.
1. The temperature of the load cell is easier to maintain. The heat due to

convection is eliminated and the conduction heat is minimized by water

cooling the pull rod.
2. It is easier to electrically isolate the load cell from the testing frame.
When using high frequency induction heating, multiple connections to
ground can cause ground-loop problems. The hydraulic elastromeric seal in
the ram isolate the lower pull rod, and Micarta washers and spacers on the
top isolate the top pull rod from the testing frame. With this technique, the
load cell ground cable remains as the only possible electrical connection to
ground.

Heating and Cooling

The specimen is heated with a standard 10 kHz induction heater. In-
duction heating was chosen because of its fast response. Indirect resistance
heating is too slow, and direct resistance heating is not acceptable for crack
propagation testing due to the local heating at crack tips. The 10 kHz heater
produces a lower wall thickness gradient than do the higher (450 kHz)
frequency heaters. Also, ground-loop problems are lessened with 10 kHz
heater than with 450 kHz heaters.
The temperature controller is a soHd state controller with mean level, am-
plitude, rate, and reset adjustments. For the feedback loop, an infrared
radiation pyrometer sighted on the specimen outer surface is used. This
pyrometer, which averages the temperature over 3-mm spot on the
specimen, has an output voltage linear with temperature. Welded ther-
mocouples and premature specimen failure due to thermocouples are
eUminated.
Compressed air is used to cool the specimen during the cool-down por-
tion of the temperature cycle. The cooling air manifold Eilong with the
heating coil are shown in Fig. 5 surrounding the specimen. The amount of
air is controlled by a motor-driven cam system and a needle valve. The cam
drives the spring-loaded needle valve and is shaped to gradually turn on and
off the air. The time for one rotation of the cam system is set to be just less
than the time for one complete TMF cycle. After one complete revolution,
the cam system stops and waits for an electrical signal from the counter
panel. This electrical signal commands the cam system to make an ad-
ditional revolution. With this technique the air is always synchronized to the
temperature cycle and is repeatable. The cooling rate is controlled by the
heater ballasting the air blast cooling. This air system is not part of the tem-
perature control loop which consists only of induction heater, specimen,
pyrometer, and controller. Therefore, timing problems between two con-
trollers in the same control loop are avoided.

Programmers
The temperature and strain programmers can be a number of different

FIG. 5—Typical TMF setup with the cooling air manifold and heating coil surrounding
the specimen.

devices depending upon the test frequency and the complexity of the
mechanical strain versus temperature curve that is to be programmed. Fig. 6
shows three standard cycles used to obtain material and design data. Cycle I
is a linear relationship between strain and temperature with the tensile strain
occurring at the lowest temperature and the compressive strain occurring at
the highest temperature. Cycle II is also a linear relationship between strain
and temperature and is the exact opposite of Cycle I, the tensile strain oc-

^TE«P»nU*E

•«SEI«LL CYCLE

TEMrEMTIWE

FIG. 6—The three standard TMF cycles conditions used to obtain material and design
data. Cycle I, Cycle II, and Baseball cycle.

curring at the maximum temperature. The third standard cycle, the Baseball
cycle, is in between a Cycle I and a Cycle II with both the tensile and com-
pressive strains occurring at the same middle temperature.
The original technique uses a dual channel Data Trak' follower for
programming total strain and temperature and is as follows: The tem-
perature versus mechanic^ strain and temperature versus time relationships
are chosen. Following this, the temperature command input curve is hand-
drawn on the Data Trak chart and then, the temperature is cycled main-
taining zero load on the specimen. During this cycling, the thermal strains
are recorded. Having acquired this thermal streiin data, the mechanical
strain and thermal strain are added algebraically, and the resulting total
strain would then be plotted onto the other channel of the Data Trak syn-
chronous with the appropriate temperature. With this procedure, there is no
way to ensure that the total strain curve is entirely plotted correctly until the
TMF test is started and the loads observed. This entire procedure has to be
repeated for each TMF test.

Thermal Expansion Compensator

A new method of strain programming has been developed that minimizes
the drawing of different Data Trak program curves. The method utilizes an
electronic thermal expansion compensating device which produces an elec-
trical signal analogous to the thermal strain measured from the specimen.
This device consists of a nonlinear ampUfier as its heart, and two linear am-
plifiers for scaling as shown in Fig. 7. The linear amplifiers were added to
ensure accurate setup and to aid in data reduction. The compensator is
'T.M. Research Incorporated.

AMPLITUDE ADJ.

PYROMETER NON-LINEAR STRAIN

AMR SPAN POT
FEEDBACK 0 0 9 0 0
0 0 0 0 0
SHAPE ADJ. (|0)LJ

RECORDER
Ly.D.T.
'*'{ MECHANICAL STRAIN)
FEEDBACK

FIG. 7—Thermal expansion compensator used to generate a signal analogous to the

thermal strain for closed-loop strain control programming and a signal analogous to me-
chanical strain for test setup and data reduction.

divided into two parts. The first part uses the temperature feedback signal
from the pyrometer as the input and outputs a voltage that is adjusted to
match the thermal strain feedback voltage. This adjustable voltage is fed in-
to the closed loop electronics for control through one of the two strain span
pots, usually Span Pot B. The mechanical strain signal from the Data Trak
is fed into the other span pot, usually Span Pot A. The second part of the
compensator has as inputs the adjustable voltage signal and the strain feed-
back signal from the LVDT. It is wired to subtract the adjustable voltage
from the strain feedback signal and produces the resultant mechanical
strain signal. Once the adjustable voltage is correctly adjusted, this me-
chanical strain signal will be zero during thermal cycling at zero load and
will be the applied mechanical strain during a TMF test.
In setting up a test utilizing the thermal expansion compensator, the
specimen is cycled through the temperature range at the desired test rate.
During this time, the loads are maintained at zero, and the thermal strains
are measured. These strains are also plotted on an x-y recorder against tem-
perature feedback. This is done to check the strain extensometry system and
the temperature gradient. Any hysteresis loops on the x-^ recorder indicates
one of these items has to be corrected. The extensometry system could either
rub against itself or not seat properly against the internal ridges of the
specimen, producing a hysteresis loop. Also, a temperature gradient along
the gage length of the specimen which is different for the heating and
cooling portion of the cycle will produce a hysteresis loop. This is produced
because the pyrometer only measures temperature at a point, and the ex-
tensometer measures the strain over the gage section. Adjustment of the in-
duction coil corrects this aspect.
Having made the corrections to eliminate the hysteresis loop, the ad-
justable voltage from the nonlinear amplifier is adjusted to match the ther-

mal expansion feedback signal. This has been achieved when it is observed
that the mechanical strain signal will not vary during thermal cycling at zero
load. The curve drawn on the strain channel of the Data Trak must now
represent the mechanical strain rather than the total strain to be controlled.
This minimizes the precision to which the curve needs to be drawn and in-
creases the overall test accuracy.

Digital Computer
It is difficult to reduce the hysteresis loop of thermal strain versus tem-
perature to exactly zero. With a digital computer as the programmer, this is
not necessary. The computer can acquire and store the thermal strains as a
function of both temperature and direction, whereas the thermal expansion
compensator's output is only a function of temperature and not direction.
The computer also allows more complex TMF testing to be conducted with
no additional effort from the operator. Such tests may include mode
transfers, strain holds for a given time or to a given stress relaxation, and
stress holds for a given time or to a given creep strain.

Crack Growth Testing

Since the fatigue Ufe of any component can be divided into a crack
initiation phase and a crack propagation phase, data for both phases is
necessary to make an accurate fatigue life prediction. Different testing
techniques are usually required for each phase; however, the technique and
specimen described in this paper are used for both [4,5\. The only difference
for a TMF crack propagation test is that a through-the-thickness notch is
required. The cracks in these notched tubular specimens were found to
propagate in a fan direction, as shown in Fig. 8. Because of this, the surface
crack lengths, measured either from a traveling microscope (2a,) or from a
cellulose acetate replica (2a,), must be converted to (2o) at the mean radius r
of the specimen. This change is simply a matter of geometry. The stress in-
tensity factor is given in Fig. 8, where G (a/Rt) is the curvature correction
factor from Erdogan and Ratwani [6\. For the strain controlled TMF tests,
a strain intensity factor rather than a stress intensity factor is used to reduce
the data [4]. This is done by using the mechanical strain range (A£„) rather
than the stress range (Ao).

Discussion
Three different techniques for programming the strain-temperature-time
relationship were mentioned in the previous section: Data Trak only. Data
Trak and thermal expansion compensator, and computer. The Data Trak
only technique requires a great deal of setup time for each test. The ac-

THKKNESS.t

-RADIUS, r

AK-AoViw G(a/rt)

FIG. 8—The Erdogan and Ratwani stress-intensity factor used in reducing crack growth
data from tubular specimens.

curacy to which these Data Trak curves has to be drawn is quite great,
especially for small mechanical strain ranges. The Data Trak and thermal
expansion compensator technique reduces the accuracy to which the Data
Trak curves must be drawn, reduces the setup time for each specimen, in-
creases the accuracy of the test, and aids in data reduction during the test.
However, the computer is still the optimum device to program this type of
test.
The accuracy to which the programmed strain curve has to be drawn to
has been mentioned a number of times. The following simple calculation
will prove this point and show why good temperature control is just as im-
portant as good strain control. A typical range in free thermal strain and
mechanical strain for a TMF test is 0.012 and 0.003, respectively. If the test
needs to be conducted within 10 percent accuracy on mechanical strain, this
means that the temperature control must be such that the free thermal strain
does not vary more than 0.0003 out of 0.012. This also requires that the
temperature and temperature gradient be controlled within 2.5 percent over
the entire temperature cycle. If total strain is being programmed from a
Data Trak, it must be drawn within 2.5 percent; however, if only
mechEuiical strain is being programmed from a Data Trak, it must be drawn
within 10 percent. Obviously running a TMF test to only 10 percent ac-
curacy is not acceptable; however, to run the test within 1 percent accuracy
can not be achieved with only Data Trak control. To hand draw the total
strain curve on the Data Trak or control the temperature dynamically to
within 0.25 percent is virtually impossible. The real accuracy lies between
1 and 10 percent with only Data Trak control depending upon all the
control parameters. With the thermal expansion compensator, accuracy is
only limited by how closely it is adjusted to zero mechanical strain during
the thermal cycling. Obviously, for a given thermal strain error, the test ac-
curacy increases as the mechanical strain range increases.
To conduct these types of tests with a computer is difficult and expensive.

If the job is done correctly, the payoffs are worth the expense. The dynamic
accuracy to which load and total strain can be controlled to, with closed-
loop equipment, is better than 1 percent. The problem in TMF testing is
matching temperature control equipment to heaters for dynamic closed-
loop control. As already mentioned, small changes in temperature can trans-
late into large changes in mechanical strain if the total strain is left un-
changed. However, a small change in temperature during a cycle without
changing the mechanical strain usually does not change the fatigue response
of the material. Therefore, a computer should be progrjunmed to take this
into account. The computer should almost continuously read the in-
stantaneous temperature and adjust its total strain output to compensate
for any temperature error. The frequency at which the computer does this
should depend upon the maximum heating rate of the temperature equip-
ment and the accuracy desired from the test. For the materials and strain
ranges associated with gas turbine apphcation, the computer has to update
its total strain signal for every 2 °C of temperature error to maintain a 1 per-
cent accuracy on the mechanical strain. Presently, this type of programmer
is being installed to program this equipment.
These control problems never surface for isothermal LCF tests because
static temperature control is much better than dynamic temperature con-
trol, and any change in temperature that does occur only reflects in a
mean strain shift rather than a strain range shift. Fatigue life, measured
as either cycles to crack initiation or the crack propagation rate, is af-
fected more by a small change in strain range than it is by the same
change in mean strain.

Summary
A TMF test method has been described which has the capability of con-
trolling any mechanical strain-temperature-time cycle. A tubular specimen
design was presented that can be used for either crack initiation or crack
propagation testing. The axial total strains were measured with a unique ex-
tensometry system and separated into two components (thermal strains and
mechanical strains) through the use of a thermal expansion compensator.
The necessity for precise temperature control was demonstrated in terms of
overall testing accuracy, along with how digital computer control can im-
prove this type of testing.

Acknowledgments
The test method described in this paper was developed over a period of
years due to the dedicated efforts of numerous people who are or have been
associated with the Materials Engineering and Research Laboratory of
Pratt & Whitney Aircraft, a division of United Technologies.

References

[}] Garden, A. E. in Manual on Low Cycle Fatigue Testing, ASTM STP 465, American
Society for Testing and Materials, 1970, pp. 163-188.
[2] Wells, C, H. in Manual on Low Cycle Fatigue Testing, ASTM STP 465, American
Society for Testing and Materials, 1970, pp. 87-99.
[3] Tishler, D. N. and Wells, C. H., Materials Research and Standards, American Society
for Testing and Materials, Vol. 6, Jan. 1966, p. 20.
[4\ Rau, C. A., Jr., Gemma, A. E., and Leverant, G. R. in Fatigue at Elevated Tempera-
tures, ASTM STP 520, American Society for Testing and Materials, 1973, pp. 166-178.
[5] Gemma, A, E., Langer, B. S., Leverant, G. R., this publication, pp. 199-213.
[6] Erdogan, F. and Ratwani, M., International Journal of Fracture Mechanics, Vol. 6,
No. 4, Dec. 1970, pp. 379-392.

Thermal-Mechanical, Low-Cycle
Fatigue of AISI 1010 Steel

REFERENCE: Jaske, C. E., "Thermal-Mechanical, Low-Cycle Fatigue of AISI

1010 Steel," Thermal Fatigue of Materials and Components, ASTM STP 612, D. A.
Spera and D. F. Mowbray, Eds., American Society for Testing and Materials, 1976,
pp. 170-198.

ABSTRACT: This study was undertaken to develop an understanding of the fatigue

resistance of AISI 1010 steel under conditions of combined thermal and mechanical
strain cycling in air. Comparative evaluations were made with existing thermal-
mechanical fatigue data on carbon steels and with results of a comprehensive, com-
panion study of the fatigue behavior of this same steel under isothermal conditions.
Thermal-mechanical fatigue behavior was investigated for constant-amplitude,
fully reversed, strain cycling of uniaxially loaded specimens at three ranges of
temperature: (a) 93 to 316°C (200 to 600°F), (b) 93 to 427°C (200 to 800°F), and
(c) 93 to 538 °C (200 to 1000°F). Experiments were conducted both with maximum
strain in phase with maximum temperature and out of phase with maximum tempera-
ture. Considering differences in experimental techniques and the difficulties as-
sociated with conducting these types of experiments, the present data agreed with
similar data from the literature in hmited instances where comparisons could be
made. Thus, these results were considered to be representative of this type of steel,
and they provided a significant extension of existing knowledge on the thermal-
mechanical fatigue resistance of low-carbon steel.
Dynamic strain aging was observed to cause more cyclic hardening in these experi-
ments than in isothermal fatigue experiments. In terms of total or plastic strainrange,
out-of-phase cycling was more deleterious than in-phase cycling, and thermal-mechan-
ical fatigue life was much less than isothermal fatigue life. However, on the basis of
stable stress amplitude, there was little difference in fatigue life between in-phase
and out-of-phase cycling, and the fatigue life was reasonably well correlated with
isothermal fatigue results.

KEY WORDS: thermal fatigue, thermal cycling fatigue, alloy steels, tests, precipita-
tion hardening

Nomenclature
AT Cross-sectional area of specimen at temperature
E Elastic modulus
' Senior researcher, Structural Materials and Tribology Section, Battelle's Columbus
Laboratories, Columbus, Ohio 43201.

170

Copyright by ASTM Int'l (all rights reserved); Sun Dec 27

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
JASKE ON AISI 1010 STEEL 171

eat Total diametral change in specimen (both thermal and mechanical)

K Specimen compliance
K' Cyclic strength coefficient
No Number of cycles where tensile load started to drop just prior to
complete fracture
Ni Number of cycles where tensile load dropped 5 percent just before
complete fracture
Nf Number of cycles to complete fracture
n' Cyclic strain hardening exponent
AP Load range
T Temperature
Ar Range of temperature
Te Equivalent temperature
/max M a x i m u m temperature
imin M i n i m u m temperature
t Time
a Linear coefficient of thermal expansion
£ Mechanical strain
£o Proportional limit strain value in cyclic stress-strain function
Lide Range of mechanical diametral elastic strain
Ltdp Range of mechanical diametral inelastic strain
Ed, Total diametral mechanical strain
£max Maximum mechanical strain
£min Minimum mechanical strain
A^ Inelastic mechanical strain range
Af, Total mechanical strain range (A£» = Cmm - £min)
V Frequency in c p m
Ve Elastic Poisson's ratio
0 Stress
La Stress range (Ao = omax - omin)
Maximum stress
Minimum stress
oo Proportional limit stress value in cyclic stress-strain function

Mechanical equipment and structural components are often subjected

to cyclic straining while operating at elevated temperature. In some in-
stances, the temperature may be relatively constant, while, in other cases,
the cychc straining may be accompanied by thermal cycling. The problem
becomes even more complex when the time relationship between tempera-
ture and strain must be taken into account.
Considering the simple case of a bar rigidly restrained at each end,
the mechanical strain becomes compressive when the bar is heated and
reaches a minimum (maximum negative) value at the maximum tempera-
ture. The thermal strain has an equal magnitude and opposite sign to the

mechanical strain because the total strain is zero. The maximum me-
chanical strain (t = 0) occurs when the temperature returns to the original
value, and the corresponding stress is tensile (positive) if yielding occurs.
This situation is usually referred to as out-of-phase, thermal-mechanical
strain cycling. At the other extreme is in-phase cycling where maximum
(positive) strain is at the maximum temperature and minimum (maximum
negative) strain is at the minimum temperature. In a complex structure,
the strain-temperature relationship may be somewhere between these two
extremes.
This paper describes the results of a study where the low-cycle fatigue
behavior of AISI 1010 steel was investigated for conditions of combined
thermal and mechanical strain cycling. Experiments were limited to the
two extreme conditions of in-phase and out-of-phase cycling. Detailed
tabulations of data developed during this program and results of a com-
panion study on the isothermal, low-cycle fatigue resistance of this same
steel are reported elsewhere [1,2]}
The objective of this study was to obtain information on thermal-
mechanical fatigue behavior of a commonly used low-carbon steel. Ex-
periments combining both thermal- and mechanical-strain cycling are
more representative of many service conditions than strain cycling at con-
stant temperature. However, since thermal-mechanical fatigue testing is
more time consuming, costly, and complicated than isothermal fatigue
testing, the approach taken in this program was to conduct a limited
study of thermal-mechanical fatigue behavior over the temperature range
of 93 to 538 °C (200 to 1000 °F) for correlation with previous isothermal
work [2] on the same material. An upper temperature limit of 538 °C
(1000°F) was considered to be reasonable for most practical uses of AISI
1010 steel. The lower temperature limit of 93 °C (200°F) was chosen as a
reasonable approximation of cooling to room temperature. This lower
temperature level was well below the blue-brittle temperature range of 149
to 316°C (300 to 600 °F) previously observed for this material [2], yet it
was high enough to permit thermal cycling at a reasonably fast rate.
Most laboratory studies of thermal-fatigue behavior, as described by
King and Smith [3], have been conducted either by thermal cycling of
end-restrained specimens or by strain cycling of axially loaded specimens
at constant temperature. It was also noted that two other techniques of
evaluating thermal-fatigue resistance have been used: (a) testing of com-
ponents under actual service conditions and (b) thermal cycling of wedges,
disks, thick-walled cylinders, and locally heated sheet specimens [3].
Although the testing of components provides very useful and practical
data, it is usually not economical at early design stages and does not
provide a knowledge of thermal fatigue behavior under well-defined

^ The italic numbers in braclcets refer to the list of references appended to this paper.

Stress-Strain conditions. The second type of thermal cycling test is some-

times useful in sorting materials, but stresses and strains developed in the
specimens are not well defined and are difficult to determine by analysis.
Garden [4\ developed a testing technique for evaluating the behavior
of materials under a combination of simple axial loading and temperature
cycling. The objective of his approach was to make it possible to evalu-
ate the thermal-mechanical fatigue resistance of materials under well-
defined stress-strain conditions by controlling both thermal and mechani-
cal strain independently. This same objective was used in developing ex-
perimental procedures for independent control of both thermal and
mechanical strain in the present study of AISI 1010 steel.
In recent years, there have been a number of books and symposia
either partially concerned with or devoted to the subject of thermal
fatigue [5-11]. In addition, there are two extensive bibliographies that
list publications on this subject [12,13]. King and Smith [3] have reviewed
many of the more recent pubhcations related to thermal fatigue. Since
such a large amount of information is already available in published
literature, the present brief review will be concerned with only the fatigue
behavior of low-carbon steels under thermal or thermal-mechanical
cycling. A similar review of isothermal-fatigue data on low-carbon steels
is also available [2].
The limited amount of available information on the thermal fatigue
behavior of low-c£irbon steels is summarized in Table 1. Although the
work of Kawamoto et al [14] and Udoguchi and Wada [15] was on steels
with higher amounts of carbon (0.69 and 0.38 percent, respectively) than
were of interest to the present study, it does contain information on be-
havior under in-phase, thermal-mechanical strain cycling. Kawamoto
et al [14] found that, when specimens greater than about 0.64 cm (0.25
in.) diameter were used, the fatigue life decreased, due to triaxial stresses
that were developed in the larger specimens. For in-phase cycling between
100 to 300°C (212 to 572°F) and 100 to 400°C (212 to 752°F), Udoguchi
and Wada [751 found that 0.30 percent carbon steel had about the same
fatigue resistance for both temperature ranges. This thermal-fatigue re-
sistance was slightly less than the isothermal-fatigue resistance at both 2(X)
and 400°C (392 and 752°F).
Lyle [16] presented a limited amount of data on the thermal-fatigue
behavior of a low-carbon steel. Similar results for a 0.16 percent carbon
steel were reported by Taira and Inoue [17] and Taira [18]. All three of
these studies used restrained specimens loaded uniaxially where the
mechanical strain was directly proportional to the change in temperature.
Since separate control of mechanical strain was not employed, the results
have limited application.
Taira and Inoue [17] also obtained multiaxial thermal stresses by
cooling a solid cylinder from one end. They were able to correlate the

- C c

^U.
ctf

E
1)
ai
HSI
111!
I
00
^ CM

o e
<u ••3
o. n
>>
HKJ
o
H
Si
*-•
111
Vl
U
•S9 H
(4-

o
4»
Q,

H>.
II
;-?^'^ ?!
(A
V
00
c:
S ill
2 S2 2
^
0

F
CO

05
u
W
3
"^^ s g -ac
.0
CO CQ

2
(U
8g|8 i
0
c 0
u 00
u f*l
a. 222 2
E
O o
§l§ li B v-1
ffl
H
o
§§§§ §8 lU VO
4S u^

•3
'C
V
4-1
tS u V)

0
o
#
s 00
m i
0
d -J

u
g 16
a
(U
St: Z

u .S
liU

li II
1^
111i! f ^^
tf
O
ki n

M
H
—• —c <N <N <N —. M
^ ^ m r^ OS ^ —^
22 2
<N rs *N r i
—« 0\ r^ O
<N f*^ »0 f*^

\ en ^ r^ S88S|8

g;
«

multiaxial with uniaxial fatigue behavior by carrying out an elastic-

plastic, thermal stress-strain analysis of the multiaxial specimens.
For the special case of thermal cycling with full end restraint, Taira
[IS] found that the thermal-fatigue life of a 0.16 percent carbon steel
could be predicted from isothermal fatigue data. For cycling at tempera-
tures below about 400°C (750°F), he found that the thermal-fatigue life
was about the same as the isothermal fatigue life for the same plastic
strain range at the mean temperature of the thermal cycle. When the
thermal-fatigue cycle had a maximum temperature of about 500 °C (930 °F)
or more, he found that the thermal-fatigue life was close to that for iso-
thermal-fatigue cycling at the same plastic strain range at the maximum
temperature. However, the variety of mechanical strain/thermal cycle
conditions studied were limited by use of the restrained specimen, and
the minimum temperature of thermal cycling was always about 200 °C
(400 °F) or greater.
The only results on out-of-phase thermal-mechanical cycling of a low-
carbon steel were those recently reported by Taira et al [19] for a 0.16
percent carbon steel. The data were well approximated by a Unear rela-
tionship between log Ae^ and log Nf, but the slope of this line was much
less than - 0 . 5 . The fatigue life values obtained from these experiments
were much lower than lives predicted from isothermal-fatigue data using
a linear damage theory.
None of the studies just described carried out both in-phase and out-of-
phase thermal-mechanical strain cycling on a heat of low-carbon steel
with extensive documentation of isothermal-fatigue behavior. Further-
more, cyclic stress-strain data were not reported.

Experimental Procedures
The material for this program was taken from the same heat of special
quality hot-rolled, 1.91-cm (3/a-in.)-diameter AISI 1010 steel bar used
in the previous study of isothermal low-cycle fatigue behavior [2]. The
chemical analysis and tensile properties of this steel were typical of those
expected for this alloy and are reported in that same study.
The thermal-mechanical fatigue specimens were machined to the con-
figuration shown in Fig. 1. This is the same external configuration used
for the isothermal-fatigue specimens. Use of this specimen configuration
allowed the temperature to be controlled at the minimum diameter where
strain was measured and reduced the influence of temperature gradients
on fatigue life. The present specimens were hollow, rather than solid as
for the isothermal studies, to facilitate cooling.
The test section of each specimen was polished with successively finer
grades of silicon-carbide paper to produce a surface finish of 0.4 jim
(16 /iin.) rms or better, with finishing marks parallel to the longitudinal

-9.85(3.88)-
-4.67(1.94)-
_ 3.17 _
(1.25) •0.635 (0.250)
2.23 -0.33 (0.i30)
(0.88) -0.635 (0.250)

Iiniiiuiu/;.
r
in)in'///, •///rniuni
'/yjlllllin
\ '\ in. (i,59 cm)-i8 UNF tiid.
0.4 ^m (ie^in.) rms finish or
Dimensions: cm (in.) better In test section

FIG. 1—Configuration of specimen.

axis of the specimen. Specimens were degreased with trichlorethylene,

followed by reagent grade acetone before being heated to temperature.

Apparatus
The thermal-mechanical fatigue experiments were conducted using the
S2ime basic servocontrolled electrohydraulic test system that was used for
isothermal-fatigue experiments [2]. This type of system has been used in
numerous low-cycle fatigue studies (for example, see Ref 20). Details
of the equipment, method of heating, gripping and alignment procedure,
technique of strain measurement, and general experimental procedures
are given in any one of several reports [1,2,21]. In the present work, the
closed-loop axial strain control system was integrated with a closed-loop
temperature control apparatus, as illustrated schematically in Fig. 2. Thus,
both the axial mechanical strain and the temperature were controlled
throughout each test.
Load was applied to the specimen axially; diametral changes in the
specimen were measured using a special diametral extensometer; and load
was measured by a strain-gaged load cell in series with the specimen.
Diametral mechanical strain was obtained by subtracting the diametral
thermal strain from the total diametral strain using a simple analog cir-
cuit. The thermal strain was computed from the linear coefficient of
thermal expansion and the change in temperature. Diametral mechanical
strain and load signals were combined and converted to an axial mechani-
cal strain using an analog computer [22].
The axial mechanical strain and temperature were controlled to follow
one of the two basic constant-amplitude waveforms shown in Fig. 3. In
Fig. 3a, £max is in phase with Tmax; whereas, in Fig. 3b, Emm is out of phase
with TmsLx. (or £min Is in phase with Tmax). The control program waveform
was provided by an arbitrary function generator. Heating rate was not
constant, but it changed throughout the cycle as shown in Fig. 3. The

--Phase reversal switch

Function
Generator

Signal
Conditioner

? Upper Crosshead S

Thermocouples
Proportional
Temperature
Controller

4 5 0 kHz
2 5 k W Induction
Heoting Unit

(b) Ttiermal and mechanical diametrol

strain
( B ) Mechanical d i a m e t r a l s t r a i n
© Load
(D) Mechanical total axiol s t r a i n

Hydraulic actuator

FIG. 2—Schematic diagram of thermal-mechanical fatigue apparatus.

heating rate was determined by the thermal response of the specimen and
the heating equipment, with the length of the thermal cycle maintained at
about IVi min. Mechanical strain was programmed through appropriate
phasing and scaling of the same waveform used to control temperature.
In two of the tests, the waveform was modified by introducing 6-min.-
long temperature and strain hold periods at maximum temperature.
Before starting each test, controls of the thermal-strain computer were
set to give close to zero diametral mechanical strain when the tempera-
ture was cycled with the load at zero. At the start of each test, the
specimen was heated to the mean temperature to obtain values of com-
pliance and elastic Poisson's ratio needed in the analog computer. These
values at mean temperature were then used throughout the test. Since
these values were actually temperature dependent, assuming them to be
constant introduced errors, as discussed in the next section of this paper.
During each fatigue test, load was continuously recorded using a time-
base strip-chart recorder. Load versus axial extension hysteresis loops
were periodically recorded during each test using an x-y recorder.

Data Acquisition
Use of the hourglass-type specimen (Fig. 1) required that diametral.

lol Maximum strain in-phase with maximum temperature

l-).,T

(bl iVtaximum strain out-of-phase with maximum temperature

FIG. 3—Illustration of control waveforms.

rather than axial, strain be measured. Because axial mechanical strain was
of prime interest, diametral strain had to be converted before the signal
from the diametral extensometer could be effectively used in the control
loop. To accomplish this conversion, it was first necessary to compute
diametral mechanical strain.
Since diametral thermal strain was simply a AT, mechanical diametral
strain was given by the following relation

£dt — £d( a AT (1)

The thermal strain computer (Fig. 2) was designed to compute ta, by Eq
1, using the value of a at the mean temperature of the thermal cycle.
Since a is a function of temperature, use of a constant value of a gave an
error at temperatures other than mean temperature. As mentioned earlier,
the value of a was adjusted experimentally to give minimum error in the
computation. This error was controlled to be of equal magnitude at Tmin

and Jraax. The size of this error, in terms of strain, is discussed after the
introduction of Eq 2.
To compute axial mechanical strain from diametral mechanical strain,
it was then necessary to determine the elastic and plastic components of
diametral strain. Once these were obtained, the elastic and plastic com-
ponents of axial mechanical strain were computed and the results summed
to generate the total axial mechanical strain. The necessary equations for
the computation of axial strain and the analog circuits of the computer
were described in detail by Slot et al [22].
In this study, the value of Poisson's ratio for inelastic deformation was
assumed to be 0.5, and the value of Poisson's ratio for elastic deformation
was determined for each specimen. The specimen was installed in the test
fixture and heated to the mean temperature of the thermal cycle. The
value of specimen compliance, K, was then set into the computer by load-
cycling the specimen for several cycles at low loads in the elastic region
and adjusting the K potentiometer until the value of Af^p was equal to
zero. The value of elastic Poisson's ratio was then determined from the
equation

where the values of Aeae and AP were determined from the slope of the
load versus diametral extension x-y plot recorded during a few load cycles
after setting the value of K. Values of E used in Eq 2 were the average
results of the tension tests reported previously [2]. Since values of AT and
E at the mean temperature of the thermal cycle were used to compute
V, at the mean temperature, an additional error was introduced in the
computation of the total mechanical strain at other temperatures.
Considering both the error introduced in Eq 1 by assuming constant
value of a and the error introduced in Eq 2 by assuming constant v^,
the magnitude of total possible errors were computed for each tempera-
ture range. For 93 to 316°C (200 to 600°F), it was about 0.03 percent
strain; for 93 to 427 °C (200 to 800°F), it was about 0.04 percent strain;
and for 93 to 538 °C (200 to 1000°F), it was about 0.05 percent strain.
Thus, for example, at 0.5 percent strain amplitude, these would be rela-
tive errors of 6, 8, and 10 percent, respectively. Although these are
fairly large errors, they are in the opposite direction for each side of the
mean temperature and cause little error in the total strain range. Also,
they are likely to have only a small influence on fatigue life for strain
ranges with relatively large inelastic strain components. However, if this
test procedure is to be used for low-strain range, long-life tests in the
future, it must be modified to account for variations in a and v^.
Examination of the load-time records revealed changes in the cyclic

deformation resistance of the material. Values of stress amplitude were

determined using the load-time histories and the cross-sectional area of
the specimen at the mean test temperature. The load-time histories also
were used to determine the three measures of fatigue life: No, Ns, and

Results and Discussion

Results of all the experiments conducted in this program are simimarized
in Table 2. Five experiments were conducted with £max at Tmax. However,
because out-of-phase cycUng was more detrimental than in-phase cycling,
the other twelve experiments were with £min a t Tmax. As described pre-
viously, At, was maintained constant with zero mean strain throughout
each test. Values of A^ at Ny/2 were computed from the relation

. . O a t £max O a t £min
A£, = A£. - — ^ — (3)

where E was the value of elastic modulus at which a at £inax o r O a t £min

occurred. For hold-time tests, a at the end of the hold period was used
in Eq 3 instead of o at £max for Specimen 79 and instead of o at £niin for
Specimen 78. Thus, Usted values of A^ are measures of the width of the
half-life hysteresis loop for each test and include both time-dependent
and time-dependent inelastic strain components.

Cyclic Stress-Strain Response

Even though the stredn cycling was fully reversed, the stress cycle was
not symmetric about zero because the temperature was different at each
extreme of the cycle. For example. Fig. 4 shows this type of stress response
for the 92 to 427 °C (200 to 800 °F) temperature range. This is typical of
the cyclic hardening observed in these experiments. In comparison, a
smaller amount of cyclic hardening was observed for the 93 to 538 °C
(200 to 1000 °F) experiments, and a larger amount was observed for the
93 to 316°C (200 to 600°F) experiments. For the in-phase cycle (Fig. 4a),
the magnitude of peak compressive stress was greater than the magnitude
of peak tensile stress, or |an,in| > |omax|. The opposite was true for the out-
of-phase cycle (Fig. 4b) were jomaxl > |omin|.
It is also important to note (Fig. 5) that maximum stress (or minimum
stress) was not necessarily coincident with the maximum strain (or minimum
strain). A detailed tabulation of half-life values of stress at £raax and Emin,
as well as values of Omax and omin, are reported in Ref 1. Examination of
the data records showed that omax (or omin) was practically the same as
a at £max (or a at £min), when £max (or £min) was at 93 °C (200 °F). However,

fnorovorN»nOO*ol>«/-ii^OONi
r^oO'-^>n>nr^fI?w^Qoooooro(Si

I ' v^ w^ m in o (
I fN ^o OS r^ O <
I «ri <^ r- f<i rsi (

?
« 0 — ' O O O O O — - ^ — l O O O O O O
iS -S C!

— CO

^
s
•Sf

i<N—lOOO—• — «—'OO—'O—CO

I Eo
88888888888
OOOOOOOOOOOOOOOOVC'O^
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
88888888888888888

.SO 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

c
«
In
0. 0
u !2—
J. : SB g E 6 S E E g E E E 6 E e E S t=
b. k. c
3 e0
a
S S .S .= .S .S S S .S .3 .= S .E .S S .g .= 0
t- E g g E E e S S e E E g a E E E S
e •03
t 3 ns
0 <u
C 0
0c
«e Z

< s

E
b

in b

Number of Cycles

500

400
1
E
< ._ f M

< ^
300
1i 3
i=
(n
200 i b

oo
e < 7

-200
en E
b
-300 e
E
-400

Number of Cycles

W £ max ^t / max-
( D ) £ min ^t ^max-

FIG. 4—Cyc/fc stress response of AISI 1010 steel under thermal-mechanical strain cycling
with a temperature range of 93 to 427''C (200 to SOO'F).

FIG. 5—Stress-strain hysteresis loop for 93 to 538°C (200 to 1000°F) cycle with £„„ at
Tm„ (specimen 56 at cycle 40).

when £max o r £min W3.S 3.t J max, there was generally a difference between
omax (or omin) and o at £max (or o at Ernin). As shown by the cyclic stress-
strain curves in Fig. 6, this difference was quite significant at higher
temperatures and larger strain ranges. As temperature became lower or
strain range became smaller, such differences became much less pronounced.
For out-of-phase cyclic stress-strain response, the difference between
Omax and a at Emax o r between Omin and o at cmin was only about 3 percent
of Omax or Omin for 93 to 538 °C (200 to 1000 °F) and negligible for the two
lower temperature ranges. However, as shown in Fig. 6, there was a dif-
ference between such data for in-phase cyclic stress-strain response. For
a constant strain amplitude, this difference became larger as the tempera-
ture range increased. It also became larger as the strain amplitude in-
creased for a constant temperature range. At 93 to 316°C (200 to 600°F)
and 93 to 427 "C (200 to 800 °F), values of a at £max or £mi„ were still
above the isothermal cyclic stress-strain curves for the maximum tempera-
ture of that cycle. However, for 93 to 538 °C (200 to 1000 °F), values of
o at £max o r £min were about the same as the isothermal cyclic stress-strain
curve at 538 °C (1000°F).
As shown in Fig. 5, an instabihty in the stress-strain response was en-
countered near 149°C (300 °F). This sort of instability was related to
dynamic strain-aging processes that resulted in discontinuous yielding
behavior. These instabilities tended to become less prominent with cyclic
hardening for all 93 to 316°C (200 to 600 °F) and 93 to 427 °C (200 to
800°F) cycles, and no significant instabilities were noted in the stabilized
hysteresis loops for these temperature ranges. However, such was not the
case for the 93 to 538 °C (200 to 1000°F) cycles, where the instabilities

200 S

OA 0,6 0,8 1.0

Total Strain Amplitude, A « / 2 . percent

FIG. 6—Cyclic stress-strain behavior of AISI 1010 steel under in-phase thermal-mechanical
cycling.

were still observed in the stable loops as shown in Fig. 5. Thus, it was
hypothesized that the time at temperatures near 538 °C (1000°F) was
sufficient to reverse the processes that caused dynamic strain aging at
lower temperatures. On the other hand, such was not the case for lower
peak temperature cycles.
Although Omax (ox Omin) was not always at tmax (or tmin), the values of
Omax and Omin provided a quantitative measure of the cyclic stress-strain
response. Therefore, they were used to characterize cyclic stress-strain be-
havior in the remaining discussion. For the 93 to 427°C (200 to 800°F)
(Fig. 4) and the 93 to 538°C (200 to 1000°F), the cychc stress in phase
with Tmax was relatively stable after the first few cycles and was within 5
percent of the value at Nf/1 for more than 90 percent of the cyclic life.
In contrast, the out-of-phase stress for these temperature ranges showed
continual cyclic hardening and the stress after 10 percent of cyclic life
was about 10 percent less than the value at Nf/2. The stress response
for both in-phase and out-of-phase cycling at 93 to 316°C (200 to 600°F)
showed the same trends as that for the out-of-phase cycling at the two
higher temperature ranges.
Morrow [25] has shown that it is often possible to express the cyclic

Stress-Strain curve in terms of a simple power function relating A^ to Ao.

However, in the present study, it was found to be more convenient to
use the same type of function for relating At, to Ao for A£,/2 < 1 percent

Ao/2 = K' (A£,/2)"' (4)

where K' = £ and « ' = 1 for 0 < A£,/2 < £o. The parameters E, K',
and n' and £o are functions of temperature as listed in Table 3; the
parameter Oo, also listed in Table 3, is directly related to to (oo = £"£0).
The thermal-mechanical, cyclic stress-strain curves defined by Eq 4 are
shown by the dashed Unes in Fig. 7. For comparison, isothermal cyclic
curves from Ref 2 are shown in Fig. 7 as solid Unes to indicate that more
cyclic hardening occurred in the thermal-mechanical than in the isothermal
experiments.

Fatigue Resistance
Results of the thermal-mechanical fatigue experiments are compared
with the previous isothermal fatigue data [2]
{a) In terms of As, in Fig. 8.
(6) In terms of Aa/2 in Fig. 9.
(c) In terms of Omax A£,/2 in Fig. 10.
Based on At, as shown in Fig. 8 (also true for Afp as shown in Ref 2),
the thermal-mechanical results fell below the isothermal results. This dif-
ference became more marked at higher peak temperatures and was greater
for out-of-phase cycling than for in-phase cycling. It was also noted that
curves for out-of-phase cycUng were generally steeper than those for in-
phase cycUng. Thus, the out-of-phase cychng was more damaging than
the in-phase cycling for these conditions.
There may be three main reasons why the thermal-mechanical data fell
below the isothermal data. First, thermal-mechanical cycles were applied
at a variable and lower strain rate than the isothermal cycles. In the
former case, the slowest strain rate was comparable to a frequency of
about 0.1 cpm; whereas, in the latter case, depending on the strain range,
since strain rate was maintained constant, the frequency was about 6 to 60
cpm. At high temperatures, usually greater than 427 °C (800°F), such a
reduction in frequency can cause a reduction in fatigue life. For AISI
1010 steel at 500°C (1112°F), Coffin [24] has shown that the effect of
frequency can be described by the following equation

Lt, = 0.00291 Nf'"-"' V o'" + 3.0N/""'' v «'" (5)

For Ae, = 1.0 percent, this equation predicts (a) A^ = 860 cycles at v
= 0.1 cpm, (b) Nf = 2280 cycles at v = 5.0 cpm, and (c) Nf = 4060 cycles

1^ -^ >n
lO vo —
o o o
t - o oo
r^ oo ' *

I »rj <s \o
"O 0> OO
cs r ) M

I
1
.5
«N (S ^
in ^ ^
m M (S

888

I
z
o

^£
0) o
Eu 888
H . vo p- oo
^ fS m
r<v ^ <n
Jo
E

0.4 0.6 0.8

Strain Amplitude, A«/2. percent

FIG. 7—Comparison of cyclic stress-strain response for thermal-mechanical and iso-

thermal strain cycling ofAISI 1010 steel.

at V = 5.0 cpm. While this effect is significant, it is not nearly large

enough to explain the very short lives observed for thermal-mechanical
cycling in this study. Furthermore, this effect would become smaller at
lower temperatures and would very likely be negligible below 427 °C
(800 °F).
Second, the thermal-mechanical specimens were hollow tubes instead of
solid bars as used in the isothermal studies. Once a significant fatigue
crack developed in the tube, it might propagate more quickly than in a
solid specimen. Although no tubular specimens were included in the pre-
vious isothermal work [2\, it is believed that, in the present case, such
differences were negligible. Low-cycle fatigue experiments using hollow
specimens have not shown significant deviations from soHd-specimen
behavior in work done at other laboratories [5,25,26\. For example, iso-
thermal data for continuous strain cycling of 2l4Cr-lMo steel [26] agree
very well (with about only ± 20 percent difference in cycUc life) with those
developed for the AISI 1010 steel over the 316 to 649 °C (600 to 1200 °F)
temperature range [2].
Third, repeated thermal-mechanical cycling through the 120 to 175 °C

z nr t Z i Z u : iZ _
O Q o
o u. c
ill •fc
1 1 1 1 1
i
8 izc M s=.
" 888888 s %^ /fi 1
o o o o o o
CO 00 N t^ ff) 0 /I / 1ft

srM
ID
ro ro c^ OJ ^ ro
IT) If) f ^ ro 1
1 1 1 1 1
fO ro K !S K ro
a> 01CT)<J) (Tl
o o o o o
/ 1
c
E E E E E
J
•S
t- 1- 1- 1- K
"5 "5 o o o O
S s S .= S c / , ' /// ili /
' / g
1
• ^ E E E E E <m g
g w w w w to / / /// / s;
•% o • < ^ 0 »
i, V ^ / / / /
///

11
o B

1 >l
E
1 = o c 3 // / "S^

1
X
i
o
E
1/ TVrr u- *
•jt 1 i
1i
V V

/
/ / /
1
j iT* f
11// 1 :§|i:
5i
11 / ! 1
y v^ O o <->
00 2
:5
/ /
7
f
/ o
1
a> i

2
Ifi / /

1^-f
x: g ^ . J
3

o
1 t Ap
J i J

/
4^
/ 1
1 11 ,' •5

°
i
if r o
1
^
-1 L V i— L _

g
*»V 'a6uDy ui.DJiS |D!XV |D|OI

,UJ/NW 'Z/*N iD -"V '9pni!|dujv ssaJts

I
s

-TT G

.i
I

I
a
I
d

!s>l '2/N D 2/^V 'apnjjidujv ssoiiS

uj/Ni^ 'Z/*»v''°'"-° 'JataiuDiDd umJiS-ssaJts

••"
%
1 I/ /
17
'o

1
'o

1 I
V / I
/
/
/
tz /
/ /
tzo / i

i1
/ / /
-8 O i
J /
o— 'J" / '
I D /
u.
r-1 O fO. / /
8^1
. o r ' <o
^hj > ro
^
/ ^
/ /
f ^ /
/ // ^ J^
\,

• / As /

/
r
/
/
o (/>
^o
>»
(J
!"
^
cs
•*3

•a
8" lr/\ P / y '' / g g o O o o 3-
Jf"
Z
^^••«
88g§og B ^3
r / £ •Sf
^
u\
1 1 1 t 1 1 U '4— t ^

/ 888888 1 - Zi •a.
^ r £

/
/ / \// -
f /f
/
/
/ } g po oo po f3 (5
O O O < J ^ O
00 GO f^ ^- crt i n
f^* ^ Sii ai S =
£
3
i
«)
.i>
t
o
[1
1
1 a>
1 CT*
1 (T>
1 <T)
1 01
1 K
f / / / ¥ / , 1
0^ E
• «

X X K M M j< O c
/ VJ / Q
O O O O O O

> 10 / » -E KE 1 E
- 1 -E ( -£ K £ .££ ^»
q
tj
01
^
'^1
>
•
>
3

3
0 B ^ B B o
S £ S £ S E 1
1 E E E e ^E ^ -
S
o
•*:

C
« •^ ~y ' £
- o
u E .a o
Is s e
•i2
E <u S

is
1 3

2
C

I
aI
LU1 ,
1
°o o
O
!s>| '2/'»v'"'"'j> 'jajaujDJDd uiDJiS-ssaiiS

(250 to 350 °F) temperature range allowed dynamic strain aging to take
place during each cycle. It is hypothesized that this repeated dynamic
strain aging was responsible for the larger amount of cyclic hardening
which occurred during the thermal-mechanical tests in comparison with
that which occurred during the isothermal tests. If the cycUc hardening
caused the stress to be higher for the same strain range without increasing
the fatigue resistance of the material, then a shorter fatigue life would
be expected for the higher stress level.
To study this hypothesis, the thermal-mechanical data were compared
with the isothermal data on the basis of stress amplitude in Fig. 9. There
was no significant difference between the in-phase and out-of-phase
results when compared on this basis. Data from 93 to 538 °C (200 to
1000 °F) cycling fell slightly above the 427 °C (800 °F) isothermal curve;
data from the 93 to 427 °C (200 to 800 °F) cycUng fell above the 371 °C
(700 °F) curve; and data from the 93 to 316 °C (200 to 600 °F) cycling fell
above the 316°C (6(X)°F) curve. The isothermal curves were taken from
Ref 2 where heavy soUd Unes are shown in Fig. 9 and were extrapolated
as shown by the Ught lines. The curve at 371 °C (700 °F) was obtained
by logarithmic interpolation between the curves at 21 and 427 °C (70 and
800°F). The curves at 482 and 593 °C (900 and llOO'F) were obtained in a
similar manner using the curves at 427 and 538 °C (800 and 1000 °F) and
at 538 and 649°C (1000 and 1200°F), respectively. To get the 371 °C (700°F)
curve, the 21 ° (70 °F) curve was used rather than the 427 °C (800 °F) curve
because it is beUeved that the strength dropped rapidly between 316 and
371 °C (600 and 700 °F) and that such a choice was more realistic. In view
of these comparisons, it appears that the increased stress due to cyclic
hardening was the primary reason for the lower fatigue life observed
previously on the basis of At, or A£p.
Smith et al [27] have used the following empirical parameter to assess
the fatigue resistance of metals

Omax(A£,/2)

This parameter is used in Fig. 10 for comparison of the thermal-mechan-

ical with the isothermal fatigue data. For the 93 to 316 °C (200 to 600 °F)
and the 93 to 427 °C (200 to 800 °F) cycles, the thermal-mechanical fatigue
data (plotted symbols) fell close to the isothermal fatigue curves associated
with the maximum temperature of the cycle. However, for the 93 to 538 °C
(200 to 1000 °F) cycle, the data fell well below the isothermal curve for
538 °C (1000°F). Thus, it appears that this parameter gives a reasonable
correlation between thermal-mechanical and isothermal fatigue data for
peak temperatures near or below 427 °C (800 °F), but not for higher peak
temperatures where time-dependent mechanisms play a significant role
in the damage process. This result was expected because the parameter

is based upon consideration of only time-independent damage processes.

Extension of such a parameter to cover time-dependent damage mech-
anisms should be pursued in future studies.
The thermal-mechanical fatigue data were compared [7] with similar
data from other studies [14-17,19]. Because of the limited amount of
published information, a complete comparison with present data was not
possible. Where comparisons were made, some differences did exist
between results of this study and those of other studies, but the overall
behavior trends were similar. Thus, it is believed that the present results
represent typical thermal-mechanical fatigue behavior of low-carbon steels.

Calculation of Fatigue Life

In order to design or evaluate components operating under combined
thermal and mechanical cycling, it is necessary to be able to make reason-
able calculations of fatigue life. Two different approaches to life cal-
culation may be considered: (a) development of relationships directly
from thermal-mechanical fatigue experiments and (b) development of
methods for correlating thermal-mechanical with isothermal fatigue resis-
tance. Although the first approach is more straightforward than the
second, it requires the conduction of costly experiments for long periods
of time when intermediate or long-life (>10'' cycles) fatigue is of interest.
Even in the short-life range of the present study, it was difficult to develop
enough data to rely on such an approach. These limitations make it desir-
able to use the second approach and develop methods of relating the
results of thermal-mechanical cycling experiments to isothermal experi-
ments. If such methods can be developed, then they can provide a rational
basis for extrapolating the results of relatively short-lived, thermal-
mechanical experiments.
In their studies of a 0.16 percent carbon steel, Taira et al [19] tried to
predict thermal-mechaniceil fatigue hfe using a linear damage theory com-
bined with the isothermal low-cycle fatigue behavior defined in terms of
plastic strain range. This attempt was not very successful in that, in most
instances, the predicted life was greater than the actual life, an uncon-
servative result from a design viewpoint. Inspection of the data showed
that such an approach would give similar results for the present study.
Taira [18] has suggested use of the concept of an equivalent temper-
ature for correlating thermal-fatigue behavior with isothermal-fatigue
behavior. This concept states that, for thermal-fatigue cycling between
two temperatures, the same fatigue hfe will be developed by isothermal
cycling at the equivalent temperature. For thermal fatigue of 0.16 percent
carbon steel, where r„,i„ was greater than or equal to 200°C (392°F), it
was shown that the equivalent temperature was defined as follows

r , s ^"^ '^ -^""", for r„ax < 400 °C (752 °F) (6a)

and

n = T-^ax, for r„ax > 500 °C (932 °F) (66)

where Nf was related to Ae^. This approach probably worked reasonably

well because Tmin was above the temperature where dynamic strain aging
could occur repeatedly. It would not work for the present study because
Tmin was low cnough to allow dynamic strain aging to take place during
each cycle.
In the present program, it was empirically found that the following
expression for equivalent temperature would provide a conservative lower
bound on fatigue life if thermal-mephanical fatigue life is related to Ao/2
at Nf/2

Te = r„ax, for Tn-ax < 316°C (600°F) (7a)

and

Tmax + 316°C (600°F)

n =- ^ ^^ ~, for 316°C (600°F)
(Jb)
<r„ax< 538 °C (1000°F)

This empirical relationship is based only on data for a low-carbon steel

for thermal-mechanical strain cycling over temperature ranges employed
in this study.
To evaluate this relationship, fatigue life calculations were made for
each of the specimens from the present work. The isothermal fatigue
curves of Fig. 9 were used in making these calculations, and the results
are shown by the open points in Fig. 11. In comparing actual with cal-
culated fatigue lives, it can be seen that in all cases (except one) the cal-
culated hfe was less than the actual life. The ratio of actual to calculated
Ufe gave a quantitative measure of the difference. The average ratio was
about 3.4, as indicated in Fig. 11.
Also shown in Fig. 11 (solid points) are calculated fatigue lives using
the parameter, OmaxA£,/2, and the isothermal fatigue curves of Fig. 10.
The ratio of actual to calculated life was about 0.22 (a factor of 4.5) for
the 93 to 538 °C (200 to 1000°F) data and was about 0.77 (a factor of 1.3)

10' 10'
Actual Fatigue Life, cycies

FIG. 11—Comparison of actual thermal-mechanical fatigue life with cyclic life calculated
from isothermal fatigue curves.

for both the 93 to 316 °C (200 to 600 °F) and 93 to 427 °C (200 to 800 °F)
data. Thus, for temperatures of 427 °C (800 °F) or less, this parameter
gave a reasonably accurate assessment of fatigue life, amd it could be
used successfully in this lower temperature range. However, the calcula-
tions were significantly unconservative at 538°C (1000°F). Therefore, life
calculations for this steel and these temperature ranges were more con-
sistent when made using the approach based on Aa/2 and T^, which gave
about the same results for all three temperature ranges 2md was conserva-
tive by a factor of about three on average cyclic life.
At first, it may seem that a factor of three is a large difference, but
it is important to consider how such calculations are used in design applica-
tions. The designer usually follows one of two approaches: {a) he deter-
mines the cyclic stresses and strains in the critical area of the component
under consideration and uses them to assess whether or not adequate
fatigue Ufe is available or (Jb) knowing what fatigue life is desired of the
component in service, he designs it so that the cyclic stresses and strains
will be low enough to assure achievement of desired life. Whichever
approach is used, it is always necessary to apply a safety factor to either
the stress and strain or to the fatigue life. Safety factors of 1.5 to 2.0
are normally applied to stress or strain, whereas factors of 10 to 20 are
usually applied to cyclic Ufe. The reason for larger factors on life than
on stress or strain is that fatigue life is usually treated as a logarithmic

variable and small changes in stress or strain produce much greater changes
in fatigue life.
When actual cyclic lives from the thermal-mechanical experiments were
used to compute Ao/2 values from the isothermal fatigue curves at Te, the
actual stress amplitudes were only about 18 percent greater than the cal-
culated ones on the average. Thus, a factor of three under calculation
of the average number of cycles to failure is not excessive in relation to
the relatively small variation in stress amplitude that can cause a factor
of three variation in cychc life. The most important point is that this
approach gives a consistent method of computing a lower bound on
thermal-mechanical fatigue life of this steel from isothermal fatigue infor-
mation. Computation of such a lower bound is more valuable for design
purposes than is prediction of average fatigue life. Although the actual
thermal-mechanical fatigue data might be considered for use, the limited
amount of these data makes the statistical confidence in them lower than
in the more numerous isothermal data.
It is important to note that either of the two life prediction methods
just discussed required a knowledge of the half-life stress amplitude and
that strain alone did not provide a sufficient correlation between thermal-
mechanical and isothermal fatigue data. In general, a knowledge of both
stress and strain were required to make accurate fatigue life predictions.
Thus, in studies of thermal-mechanical fatigue behavior of materials, it
is necessary to develop information on cycUc deformation behayior as
well as on cyclic life. Also, the methods are based upon information for
uniaxial loading and are not generally valid for multiaxial loading. Develop-
ment of a method for complex multiaxial loading situations that are often
encountered in design appUcations should be pursued in future studies.

Summary and Conclusions

During this study, thermal-mechanical fatigue experiments were con-
ducted on AISI 1010 steel for temperature ranges of 93 to 316°C (200 to
600°F), 93 to 427°C (200 to 800°F), and 93 to 538°C (200 to 1000°F).
Experiments were under fully reversed strain cycling, where the maximum
strain was either in phase or out of phase with the maximum temperature.
Where comparisons could be made with a limited amount of published
data, present results showed similar trends to those of other studies.
Therefore, the present data were considered to be typical of those ex-
pected for low-carbon steels in general.
Evaluating the results of the present experiments in comparison with
results of comparative isothermal fatigue studies [2] leads to six major
conclusions, applicable to the steel and the temperature ranges just men-
tioned.
1. More cyclic hardening was observed in the thermal-mechanical exper-

iments than in isothermal experiments. Cydic hardening behavior was

noted for all temperature ranges and was greater for the out-of-phase
peak stress than for the in-phase peak stress.
2. In terms of At, or Ae^ versus N/, fatigue life was much lower for
thermal-mechanical cycling than for isothermal cycling and out-of-phase
cycling was more detrimental than in-phase cycling.
3. Stress-based fatigue curves for thermal-mechanical cycling were
comparable to those of isothermal studies. Therefore, it was hypothesized
that the added cyclic hardening caused the reduction in fatigue Ufe on a
strain basis.
4. On the basis of Ao/2 at N/2 versus Nf, little difference was observed
between data for in-phase and out-of-phase cycling, and a lower bound
on the thermal-mechanicEil results could be calculated from the isothermal
results using an empirically developed effective temperature concept. This
method of calculation may be useful in design applications.
5. On the basis of OmaxA£,/2 at N/Z versus Nf, the thermal-mechanical
fatigue Ufe for 93 to 316 °C (200 to 600 °F) and 93 to 427 °C (200 to 800 °F)
cycles correlated well with the isothermal fatigue hfe at the peak tempera-
ture. However, no such correlation was obtained for the 93 to 538 °C
(200 to 1000°F) cycle.
6. A knowledge of both cyclic stress and strain was required to obtain
an adequate correlation between thermal-mechanical £md isothermal fatigue
life.

Acknowledgments
The work described in this report was sponsored by the American Gas
Association, Inc. (Project No. HA-3-1), and was performed at Battelle-
Columbus, Columbus, Ohio. Thanks are extended to Dr. Ab Flowers of
the American Gas Association, Inc., for his guidance of the program and
to R. W. Newell, G. C. Groff, and O. C. Davis of the advisory com-
mittee for their technical assistance.
Appreciation is EIISO extended to the following Battelle-Columbus staff
members for their assistance in conducting the experimental work: N. D.
Frey, J. J. Parks, and H. J. Mahk. T. L. Porfilio, formerly of Battelle-
Columbus and presently with Dialight, Brooklyn, New York, assisted
in technical conduct and planning of this program. Dr. L. E. Hulbert
and H. Mindlin of Battelle-Columbus provided direct technical super-
vision of the studies.

References
[1] Jaske, C. E. and Porfilio, T. L., "Thermal-Mechanical, Low-Cycle Fatigue of AISI
1010 Steel," Report to American Gas Association, Inc., Catalog No. H54375, Bat-
telle's Columbus Laboratories, Columbus, Ohio, 14 Feb. 1975.

[2] Jaske, C. E. and Porfilio, T. L ., "Isothermal, Low-Cycle Fatigue of AISI 1010 Steel
at Temperatures Up to 1200 F , " Report to American Gas Association, Inc., Catalog
No. H54975, Battelle's Columbus Laboratories, Columbus, Ohio, 14 Feb. 1975.
[5] King, R. M. and Smith, A. 1. in Thermal and High-Strain Fatigue, Monograph and
Report Series No. 32, The Metals and Metallurgy Trust, London, 1967, pp. 364-378.
[4] Carden, A. E. in Manual on Low-Cycle Fatigue Testing, ASTM STP 465, American
Society for Testing and Materials, 1970, pp. 163-188.
[5] Manson, S. S„ Thermal Stress and Low-Cycle Fatigue, McGraw-Hill, New York, 1966.
[6[ "Thermal Fatigue of Metals in Naval Power Machinery," Wash 1178, "Sudostroyeniye"
Press, Leningrad, 1967, given to the 1970 U.S. Nuclear Power Reactor Delegation.
[7\ "Thermal and High-Strain Fatigue," Monograph and Report Series No. 32, The
Metals and Metallurgy Trust, London, 1967.
[8] Fatigue at High Temperature, ASTM STP 459, American Society for Testing and
Materials, 1969.
[9\ International Conference on Thermal Stresses and Thermal Fatigue, Berkeley, England,
1%9.
[10] Fatigue at Elevated Temperatures, ASTM STP 520, American Society for Testing and
Materials, 1973.
[JJ] "International Conference on Creep and Fatigue in Elevated Temperature Applica-
tions," American Society for Testing and Materials, American Society of Mechanical
Engineers, and Institution of Mechanical Engineers, Philadelphia, 23-27 Sept. 1973.
[12] Carden, A. E., "Bibliography of the Literature on Thermal Fatigue," N68-23652
(NASA CR-94605), University of Alabama, Huntsville, 1967.
[13] Wundt, B. M., Low-Cycle Fatigue Bibliographic Survey, ASTM STP 449, American
Society for Testing and Materials, 1968.
[14] Kawamoto, M., Tanaka, T., and Nakajima, H., Journal of Materials, Vol. 1, No. 4,
1966, pp. 719-757.
[15] Udoguchi, T. and Wada, T., "Thermal Effect on Low-Cycle Fatigue Strength of
Steels," International Conference on Thermal Stresses and Thermal Fatigue, Paper
No. 18, Berkeley, England, 1969.
[16] Lyle, J. M., "Forced Air Furnace Exchangers; Some Design and Reliability Evaluation
Considerations," Gas Appliance Engineers Society Meeting, June 1971.
[17] Taira, S. and Inoue, T., "Thermal Fatigue Under Multiaxial-Thermal Stresses," Inter-
national Conference on Thermal Stress and Thermal Fatigue, Paper No. 3, Berkeley,
England, 1969.
[/*] Taira, S. in Fatigue at Elevated Temperatures, ASTM STP 520, American Society for
Testing and Materials, 1973, pp. 80-101.
[19] Taira, S., Fujino, M., and Takashi, M., Journal of Society of Materials Science, Japan,
Vol.22, 1973, pp. 235-241.
[20] Jaske, C. E., Perrin, J. S., and Mindlin, H., Reactor Technology, Vol. 15, No. 3, Fall
1972, pp. 185-207.
[21] Jaske, C. E., Mindlin, H., and Perrin, J. S., "Low-Cycle-Fatigue Evaluation of Reactor
Materials," in USAEC Report No. BMI-1928, Battelle's Columbus Laboratories,
Columbus, Ohio, July 1972.
[22] Slot, T., Stentz, R. H., and Berling, J. T. in Manual on Low-Cycle Fatigue Testing,
ASTM STP 465, American Society for Testing and Materials, 1969, pp. 100-128.
[23] Morrow, JoDean, in Internal Friction, Damping, and Cyclic Plasticity, ASTM STP
378, American Society for Testing and Materials, 1965, pp. 45-87.
[24] Coffin, L. F., Jr., "A Generalized Equation for Predicting High-Temperature Low-
Cycle Fatigue, Including Hold Times," Air Force Conference on Fatigue and Fracture
of Aircraft Structures and Materials, Miami Beach, Florida, 1969, (Also General
Electric Research and Development Center Report 69-C-401, 1969.)
[25\ Hirschberg, M. H. in Manual on Low-Cycle-Fatigue Testing, ASTM STP 465, American
Society for Testing and Materials, 1969, pp. 67-86.
[26] Manson, S. S., Halford, G. R., and Hirschberg, M. H. in Design for Elevated Temper-
ature Environment, The American Society of Mechanical Engineers, 1971, pp. 12-28.
[27] Smith, K. N., Watson, P., and Topper, T. H., Journal of Materials, Vol. 5, No. 4,
Dec. 1970, pp. 767-778.

Thermomechanical Fatigue Crack

Propagation in an Anisotropic
(Directionally Solidified) Nickel-
Base Superalloy

REFERENCE: Gemma, A. E., Langer, B. S., and Leverant, G. R., "Ther-

momechanical Fatigue Cracl( Propagation in an Anisotropic (Directionally Solidifled)
Nickel-Base Superalloy," Thermal Fatigue of Materials and Components, ASTM STP
612, D. A. Spera and D. F. Mowbray, Eds., American Society for Testing and
Materials, 1976, pp. 199-213.

ABSTRACT: Strain-controlled thermomechanical fatigue tests were performed on two

nickel-base superalloys, directionally solidified (DS) Mar-M200 plus hafnium and con-
ventionally cast B-19<X) plus hafnium. The DS alloy was tested at various angles (8) to
the direction of grain growth. It was found that crack growth rates were a minimum for
the DS alloy tested parallel to the direction of grain growth (crack propagating normal
to the direction of grain growth). The crack growth rate increased for other orientations
in the sequence 6 = IS, 30 or 90, and 45 deg until equivalence was reached with rates for
the conventionally cast alloy, B-19(X) plus hafnium. A linear elastic fracture mechanics
approach was extended to strain-controlled crack growth as a function of 9 for the DS
alloy. Prediction of crack growth rates as a function of 6 was achieved by normalizing
with the elastic modulus. The sequencing of crack growth rates with 6 and a smooth-to-
rough fracture surface transition are explained on the basis of the orientation depen-
dence of crack opening displacement for a given strain intensity range.

KEY WORDS: thermal fatigue, fatigue tests, nickel alloys, crack propagation

Linear elastic fracture mechanics was extended to fatigue when Paris

[7]^ showed that the range of the stress intensity could be used to correlate
crack propagation data. Since the stress intensity factor is only a function
of the applied load, crack, and structure geometry, (that is, it is indepen-
dent of material properties), the crack data can be used to predict the fa-
tigue behavior of any structure that is subject to the same cyclic environ-

' Senior assistant project engineer, senior materials engineer, and group leader, respectively,
Pratt & Whitney Aircraft, Middletown, Conn. 06457 and East Hartford, Conn.
^ The italic numbers in brackets refer to the list of references appended to this paper.

199

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
200 THERMAL FATIGUE OF MATERIALS AND COMPONENTS

mental conditions under which the crack growth data has been obtained.
The empirical parameters, that is, the pre-exponential term and the ex-
ponent of the Paris-type crack growth relationship, encompass the mate-
rial and environmental information. Some investigators [2-4] have studied
the influence of various mechanical properties on fatigue crack propaga-
tion; however, Pearson [5] was among the first to recognize the strong in-
fluence of Young's modulus on the fatigue crack growth rate. Richards
[6] found a correlation between grain orientation and room-temperature
fatigue crack growth rate in silicon iron, and he attributed the effects to
the variation in the ultimate tensile strength and Young's modulus.
Woodford and Mowbray [7] conducted thermal fatigue tests using tapered
disks in a fluidized bed. They found that the crack growth rate correlated
with grain orientation for directionally solidified IN 738. The variation in
crack growth rate was attributed to carbide alignment and dendrite orien-
tation.
Rau, Gemma, and Leverant [8] have shown that, for strain-controlled
thermal-mechanical fatigue cycling of high strength turbine alloys, crack
growth may be analyzed on the basis of a linear elastic fracture mechanics
approach.' This paper will extend this approach to a directionally solidified
nickel-base superalloy tested at various angles to the grains growth direc-
tion. It will be shown that the orientation dependence of the modulus of
elasticity can be used to normalize the power-type crack propagation rela-
tionship. Thus, the dependence of crack growth behavior of an anisotropic
alloy on the angle between the grain growth direction and the tensile axis,
9, can be predicted from a single test of longitudinal (0 = 0 deg) material.

Alloy Characterization and Experimental Procedure

The materials investigated were Mar-M200 plus hafnium (PWA 1422)
and B-1900 plus hafnium (PWA 1455) which are directionally solidified
(DS) and conventionally cast nickel-base superalloys, respectively. The al-
loys are composed of a y solid solution matrix precipitation hardened by
ordered y' particles and contain both bulk and grain boundary carbides.
The microstructures and chemistries for both alloys are shown in Fig. 1.
The Mar-M200 plus hafnium specimens were fabricated from 1 by 4 by
6 in. (2.5 by 10.2 by 15.2 cm) slabs, except the longitudinal (0 deg) specimens
which were machined from 1.250 by 5.5 in. (3.2 by 14.0 cm) bars. This al-
loy was heat treated at 2200°F (1204°C) for 2 h, 1975°F (1079°C) for 4 h,

' The data in Ref 8 and much of the data in the present study are plotted as a function of the
strain intensity range [A^. = A£ V naf (geometry)] rather than the usual stress intensity range.
This is done primarily for convenience since thermal fatigue is a strain-controlled property,
and lifetime prediction of components is more easily accomplished when deaUng in terms of
strain distributions. (As shown later in Fig. 6, the stress intensity factor is still considered to be
the prime driving force for crack propagation.)

Growth Axis

MAB-M200 plus Hafnium

Analyzed Chemistry (Weight Percent)
Al B Co Cr Zr Hf Ma Mb Ti
5.0 0.015 0.11 10.2 8.8 0.08 2.1t6 0.001 1.1 2.0
3i W ' Pb Bl 0
0.10 li!,U P. P.M. F.P.M. P. P.M.
< 0.5 < 12

B-190C' plus Hafnium

Analyze;! Chemistry (Weight Percent)
Al B C Co Cr Fe Hf Mn Mo Ta
6.0 0.013 0.13O 10.0 7.9 0.09 1.2 0.002 5.9
Tl Si /•r Pb Bi 0
1.1 0.06 0.09 P.P.M. P. P.M. P. P.M.
^~2 < 0.5 < 10
FIG. 1—Microstructure and chemistry.

and 1600°F (871 °C) for 32 h prior to machining. The B-1900 plus hafnium
specimens were machined from hollow bars and were heat treated at
1975°F (1079°C) for 4 h and 1650°F (899°C) for 10 h. The Mar-M2(X)
plus hafnium grain size ranged from approximately 0.040 to 0.150 in. (1

to 3.8 mm) and the B-1900 plus hafnium grain size was equal to or less
than 0.032 in. (0.8 mm) with preferred growth in the radial direction.
Thermal-mechanical fatigue tests were performed in strain control on
tubular specimens (Fig. 2a) with a 1 in. (2.54 cm) uniform section gage
length in closed loop testing machines. Axial strain was measured with a
calibrated linear variable differential transformer (LVDT) and its associ-
ated extensometry (coaxial quartz tubes) which rested on ridges machined
on the inside diameter of the specimen gage section, as shown in the cut-
away view in Fig. 2b. Temperature was controlled £md measured with the

Bs^SaT

FIG. 2—Strain control fatigue specimen: (a) macroetched Mar-M200 plus hafnium. The
angle between the grain growth direction and the tensile axis, 9, is 45 deg and (b) cutaway
view of specimen with quartz extensometer in place.

aid of a calibrated infrared temperature comparator, and both the strain

and temperature signals were independently controlled by the closed loop
servohydraulic systems. To produce the desired strain-temperature histo-

ries, the temperature excursions were obtained by induction heating and

forced convective cooling of the outside specimen surface. Total cycle
time was 2'4 min (0.445 cpm, 0.0074 Hz) with a symmetrical cycle shape
maintained for the heating and cooling portions of the cycle.
The fatigue cracks were started from electrical discharge machined slots
(0.45/0.030 by 0.007/0.005 in.) (1.14/0.76 by 0.178/0.127 mm) in the cen-
ter of the gage section. The specimen wall thickness was 0.055 to 0.060 in.
(1.40 to 1.52 mm) except for the longitudinal or 0-deg oriented DS Mar-
M200 plus hafnium specimens where the wall was 0.040 in. (1.02 mm)
thick. Crack extension was monitored with a calibrated micrometer
mounted telescope unit. A curve was then fitted to the plot of crack length
versus number of cycles. The half-crack extension rate (da/dN) versus
strain intensity range (AATJ was calculated by the various methods to be
discussed and plotted. All the strain intensity factors were calculated with
the aid of a curvature correction [9] and a geometrical correction of true
projected length to mean arc length. For all testing, the maximum tensile
strain was applied at the minimum cycle temperature (427 °C, 800 °F) aiid
the maximum compressive strain at the maximum cycle temperature
(1038 °C, 1900°F), Fig. 3a. The crack growth rates were determined for
the conventionally cast B-1900 plus hafnium and for DS Mar-M200 plus
hafnium having the grain growth direction oriented 0, 15, 30, 45, and 90
deg from the loading axis. A schematic of the loading and grain growth
axis is shown in Fig. 3b.
The cyclic strain ranges employed were less than 50 percent of the yield
strain (Table 1), and the appropriate calculations indicate that modifica-
tion to the measured crack length by the crack tip plastic zone size was
negligible.

Experimental Results

Dependence of the Crack Growth Rate of Directionally Solidified Mar-

M200 Plus Hafnium on the Angle Between the Grain Growth Direction
and Tensile Axis
The Mar-M200 plus hafnium thermal-mechanical fatigue crack propa-
gation rate was a function of the angle between the tensile axis and the
grain growth direction. The slowest crack growth rate was obtained for
the longitudinal material (0 = 0 deg) and the most rapid crack growth
rate for 45-deg off-£ixis tests. The 15-deg off-axis crack growth rates were
25 to 80 percent faster than the average 0-deg growth rates. The 30 and
90-deg off-axis crack growth rates fell within the same scatterband and
were faster than the 15-deg off-axis test. The crack growth rates for the
45-deg orientation were similar to those obtained for conventionally cast
B-1900 plus hafnium under identical test conditions (Fig. 4) and were

LOAD AXISl GRAIN GROWTH

(1-1) * DIRECTION
(LONGITUDINAL)

TRANSVERSE
DIRECTION

FIG. 3—(a) Schematic of thermal-mechanical fatigue cycle and (b) schematic of loading
and grain growth axes.

TABLE 1—Elastic modulus, yield strain, and yield stress of DS Mar-M200 plus hafnium and
conventionally cast B-1900 plus hafnium.

DS Mar-M200 plus Hf Orientation (6), deg B-1900 plus Hf

Temperature, °F 0 15 30 45 90 Cast

Elastic Modulus (psi x 10')

800 17 20 26 34 20 25.4
1500 14 17 23 30 17 22.2
1700 12 15 20 26 16 20.8
1900 11.5 13 18 23 12.5 19.4

Yield Strain," %

800 0.97 0.82 0.68 0.60 0.78 0.58

1500 1.02 1.02 0.81 0.65 0.81 0.59
1700 0.85 0.85 0.69 0.52 0.68 0.48
1900 0.66 0.52 0.47 0.41 0.57 0.38

Yield Stress, ksi

800 130 121 123 133 116 118

1500 115 132 131 136 103 107
1700 85 94 92 84 76 74
1900 58 43 46 43 48 42

"Strain required to attain the 0.2 percent yield stress.

1 X10-'

HALF CRACK
GROWTH RATE
da/dN
(INCHES/CYCLE)

IX 10-«

5 I 7 I 910
STRAIN INTENSITY R A N G E , A K C ( V 1 N ! X 1 0 ' ' )

FIG. 4—Dependence of the fatigue crack propagation rate on the angle between the tensile
axis and the grain growth direction.

about a factor of eight faster than those for the longitudinal material
(0 = 0 deg). Conventionally cast Mar-M200 plus hafnium was not tested
in this investigation but would be expected to show crack growth rates
similar to conventionally cast B-1900 plus hafnium.

Fractography
Crack growth was transgranular and perpendicular to the tensile axis
(Stage 2) for all DS Mar-M2(X) plus hafnium orientations. At low strain
intensity ranges (tJC^, the crack propagation path was smooth. With in-
creasing tJCi, Fig. 5, the amount of interdendritic cracking increased,
resulting in a much rougher fracture surface for all orientations. The
strain intensity range for the smooth to rough fracture surface transition
decreased in the following order: 9 = 0, 15, 30 or 90, and 45 deg, Table
2.

Discussion

Background
The stress intensity factor of a crack oriented at an arbitrary angle
to the principal material directions of an orthotropic plate is not meaning-
ful within the now classical Irwin approach to fracture mechanics. Wu
[10,11] has shown that the stress intensity factor for a crack in an ortho-

f T

(b)

(c)

(a) 0 deg off-axis (longitudinal).

(b) 30 deg off-axis.
(c) 45 deg off-axis.
(d) 90 deg off-axis (transverse).
Arrowheads show the location of the smooth to rough transition.
FIG. 5—Fracture surfaces for various angles between the loading axis and the grain growth
direction.

TABLE 2—Values of various fracture mechanics parameters at the transition from a smooth
to a rough fracture surface in DS Mar-M200 plus hafnium.

d (Angle Between Loading Axis and Growth

Direction), deg

0 15 30 45 90
Strain intensity range A*:., sAm. X 10"' 1.76 1.75 1.31 1.23 1.26
1.96 1.34
avg 1.86 1.30
Estimated crack opening displacement
COD, in. X 10-' 6.0 8.8 6.0 6.0 7.2
7.6 6.9
avg 6.8 7.0

tropic plate is reducible to the Irwin approach, that is, uncoupled modes,
only if the crack is oriented along one of the principal material directions,
and Bowie and Freese [12] allude to the difficulties which arise when lines
of elastic symmetry do not coincide with the coordinate axes. Most
published analyses are specialized to avoid these difficulties and so fall
within the Irwin scheme of fracture mechanics [10-14]. Gandhi's analysis
[75] of an arbitrarily oriented crack in a finite orthotropic sheet indicates
a strong dependence of the stress field of an inclined crack on material
properties and the aspect ratio of the finite orthotropic plate. In addition.
Cook and Rau [76] critically examined the concepts of linear elastic
fracture mechanics and concluded that orientation and plasticity effects
were of significance in fatigue crack extension.
In light of these difficulties, a heuristic approach was taken to estimate
the crack propagation behavior of a directionally soUdified nickel-base
superalloy. The objective of the analysis was to retain the linear elastic
isotropic definition of the stress intensity factor by introducing the effects
of grain orientation in the Paris-type crack propagation relationship.

Normalization of the Fatigue Crack Propagation Data

As shown in Fig. 4, crack growth rates increased in the following
sequence: 6 = 0, 15, 30 or 90 and 45 deg. If the driving force for crack
propagation is calculated in terms of effective stress intensity* (obtained
from the load history during strain cycling) rather than strain intensity,

* The effective stress intensity range (A/fo)eff is derived from the load history of a strain-
controlled test and is calculated in exactly the same manner as the standard stress intensity
range for load-controlled cycling. It is not strictly equivalent to the stress intensity range
calculated from a load-controlled test because of the presence of complex time-dependent
plasticity at the crack tip during cycUng at elevated temperature. As a result, {U(o)crf would
not be material independent or independent of orientation for an anisotropic material as
required by the linear elastic fracture mechanics definition of stress intensity range.

the data are significantly condensed, as shown in Fig. 6, suggesting that

the elastic modulus might be utilized for predictive purposes. For example,
it would be desirable to predict all off-axis crack growth behavior from

O.S. MAR-II200 PLUS Hf

CyCLE SHAPE

HALF CRACK
GROWTH RATE,
da/dN
(INCHES/CYCLE)

/ g» = 90'
' 0 6 = 45*
D ClJ=30"
a j A9<a5-
— 9 = 0 * SCATTER >AND
J I I I II

EFFECTIVE STRESS INTENSITY RANGE, (AKo.)gf,(KSl7iN)

FIG. 6—Convergence of the crack growth rates when plotted as a function of effective
stress intensity range.

one test of a specific orientation. The data in Table 1 indicate that there
is some relationship between the anisotropic elastic modulus and crack
growth rate. This approach is developed as follows.
A plate of directionally solidified alloy may be considered to be ap-
proximately transversely isotropic if the transverse columnar grain
orientation (<100> to <110> in the case of directionally solidified nickel-
base superalloys) is random. Thus, any lamina of the plate when loaded
uniaxially in the 1-1 direction, is orthotropic. The ratio of the longitudinal
modulus, EL, to the modulus of the load axis, £•, is [77]

EL I EL \
A"' = — = cos "0 + — sin''0 + ( IVLT ) cos^e sin^e
ET \GLT J
where
0 = angle between the load and grain growth axes,
E-T = transverse modulus (0 = 90 deg),
GLT = shear modulus, and
VLT = Poisson ratio.

A schematic diagram of the loading and grain growth axes is shown in

Fig. 3b.
The crack growth law is then normalized with respect to the grain
growth direction and has the form da/dN = ^ [A (A/Tt)]" where A and n
are constants that are determined from tests for which the loading and
grain growth direction are coincident, that is, 0 = 0 deg and A = 1. The
material constants, E and v, used in the calculation were obtained at
800°F (427 °C) because, at this temperature, the specimen experiences
the maximum tensile strain. (Similar results can be obtained using other
temperatures as well.) The results are shown in Fig. 7. A very good
correlation of the data is achieved with this approach.

B' W 30'WI5V

HALF CRACK
GROWTH RATE
da/dN
(INCHES/CYCLES)
C«ICUI«T10 ( X A K , )
TEStOMH
O S = 0•
^ 9 • IS-
a e = 90-
C) 8 = 30*
O 8 • 45-

4X10
20 30 4.0 5.0 G.O 10 10.0
AK.cyliix 10')

FIG. 7—Predictability of off-axis crack growth rates in DS Mar-M200 plus hafnium by

employing a modulus correction factor (A) where

> - ' = — = co^B +^sin''e + I — - 2vtT Icos^esiw^e.

Ex ')•

It is generally accepted that fatigue crack growth in a single cycle is

proportional to the crack opening displacement (COD) at the crack tip
[18\. This suggests that, for a given crack growth rate, COD should be in-
dependent of 0. In order to make an estimate of COD, the equation for
isotropic and time-independent material's response is employed, for
example

COD = iiKJ/4Ea,

where
AATo = stress intensity range,
E = Young's modulus, and
Oy = yield stress [18].
The values for the estimated COD for various Q calculated with the
800 °F material constants are tabulated in Table 3 for a constant crack
growth rate of 2.5 x 10^* in./cycle. The calculated COD is reasonably
constant, with the exception of the value for 0 = 45 deg.

TABLE 3-—Crack opening displacement (COD) for ar given crack growth rate.

e (Angle: Between Loading Axis and Grain Growth Direction), deg

0 15 30 45 90

Estimated COD at a
constant growth rate
of2.5 X lO-Mn./cycle 1.43 x :lO- 1.73 X 1 0 - 1.53 X 10-" 0.86 X 10"" 1.95 X l O -

The ability of the elastic modulus to normalize the crack growth data
must be associated with the form of the equation for COD. Oy is a weak
function of d (Table 1), and there is a direct relationship between elastic
modulus and stress intensity range experienced by the crack tip. Since
the loading conditions for these tests resulted in no net section yielding
(the one exception being moderate shakedown in the initial portion of
the 45-deg orientation), the elastic modulus and AA",, are directly pro-
portional such that the A/f„^ term in the COD equation dominates.

Fractography
It was reported in the results section that the transition from a smooth
to a rough fracture surface occurred at decreasing AKe in the order
0 = 0, 15, 30 and 90, 45 deg. Table 2. Calculation of the COD at the
transition using the effective stress intensity range shows that the esti-
mated COD is essentially constant (6.0 to 8.8 x 10"' in.) at this location,
Table 2. This is consistent with the observation that COD is the control-
ling factor in determining the crack growth rate for various d.

Comparison with Other Studies

It is interesting to compare the results of this investigation with those of
Pearson [5], Richards [6], and Rau [19] who have examined the influence
of elastic modulus on fatigue crack growth under stress-controlled con-

ditions. In these studies, fatigue crack growth rate decreased with in-
creasing elastic modulus. An example is shown in Fig. 8 for Udimet 700
[19], a superalloy similar to Mar-M200 and B-1900. For stress-controlled
n - ' .Y
1 X10-

7- SMCLE CItVSTtL UOIMET-700 EB 12 X 10' PSI

S 4-

-700 ESI I t X 10' PSI

t- 1 X 10-
as

ixio-'4fi5 i
6 7 t 9 10 IS 20 30
STRESS INTENSITY RANGE, AKJKSiyiN)

FIG. 8—Crack growth rate as a function of stress intensity range for single crystal and
wrought Udimet-700 tested in pulsating tension (R = 0.1) at nOO'F (927°C).

cycling, the lower modulus material has a considerably higher crack

growth rate compared to the higher modulus form of the same alloy. In
contrast, the present work, which was performed under strain-controlled
conditions, showed an increasing crack growth rate with increasing elastic
modulus. Fig. 4.
This result emphasizes the basic difference between strain- and stress-
controlled crack propagation testing at elevated temperatures. Although
Fig. 6 would seem to imply that the crack growth rate is insensitive to
grain growth direction and elastic modulus for load control tests, the
(A^„)eff used to plot that graph is clearly not equivalent to the A^,, cal-
culated in a load-controlled test. The primary information to be derived
from Fig. 6 is the importance of elastic modulus on crack growth rate.
As pointed up previously, because of the complex, time-dependent de-
formation at crack tips in elevated temperature testing, one cannot
deduce that the absence of net section yielding implies that (A/r„)efr is
equivalent to LK„.

The difference in material's response between stress- and strain-con-

trolled cycling can be explained on the basis of COD. In stress-con-
trolled testing, the denominator of the COD equation (4£'o^) dominates
because the numerator (A^,,^) is constant. Higher E results in reduced
levels of COD and reduced crack growth rates. In contrast, as already
discussed, strain-controlled testing of DS Mar-M200 plus hafnium results
in dominance of the A^„^ term in the numerator of the equation for
COD, and this leads to an increase in COD and crack growth rate with
increasing E.
Thermal fatigue crack propagation rates as a function of the angle
between the crack growth path and the grain growth direction have been
measured by Woodford and Mowbray [7] in fluidized bed testing of disks
of DS IN 738. In contrast to the present results, they found that 0 = 45
deg had the most resistance to crack propagation. This orientation
dependence was attributed to aligned MC carbides which significantly
influenced the crack path and apparently the crack growth rate. In ad-
dition, it would be expected that the distribution of strains around the
periphery of the disk would be asymmetrical because of the tendency for
an anisotropic material to distort under such loading. In DS Mar-M2(X)
plus hafnium, high-density-aligned carbides are not an important factor,
and testing for each orientation was performed under uniform loading
conditions.

Summary
It has been established that thermomechanical fatigue crack growth
rates in a directionally solidified nickel-base superalloy are a minimum for
loading parallel to the grain growth direction (0 = 0 deg). Crack growth
rates increase for other orientations in the sequence 0 = 15, 30 or 90,
and 45 deg until equivalence is reached with rates for conventionally-cast
superalloys.
It was demonstrated that a linear elastic fracture mechanics approach
can be extended to the prediction of strain-controlled thermomechanical
fatigue crack growth in an anisotropic material of this type. Prediction of
crack growth rates as a function of the angle between the loading direc-
tion and the grain growth direction can be achieved by utilizing the elastic
modulus as a normalization factor. The relative crack growth rates as a
function of orientation (0) and the smooth-to-rough transition in fracture
surface appearance can be explained by the orientation dependence of the
COD.

References
[1] Paris, P. C , Gomez, M. P., and Anderson, W. E., The Trend in Engineering, Vol. 13,
Jan. 1961, p. 9.

[2] Grosskreutz, J. C , Physica Status Sotidi (b), Vol. 47, No. 1, 1971, pp. 11-31.
[3] Grosskreutz, J. € . , Physica Status Solidi (b). Vol. 47, 1971, pp. 359-396.
[4] Hickerson, J. P., Jr., and Hertzberg, R. W., Metallurgical Transactions, Vol. 3, Jan.
1972, pp. 179-189.
[5] Pearson, S., Nature, Vol. 211, Sept. 1966, pp. 1077.
[5] Richards, C. E., Acta Metallurgica, Vol. 19, July 1971, pp. 583-596.
[7] Woodford, D. A. and Mowbray, D. F., Materials Science and Engineering, Vol. 16,
No. 1/2, Oct./Nov., 1974, pp. 5-43.
[8] Rau, C. A., Jr., Gemma, A. E., and Leverant, G. R., in Fatigue at Elevated Tempera-
tures, ASTMSTP520, American Society for Testing and Materials, 1973, pp. 166-178.
[P] Erdogen, F. and Ratwani, M., International Journal of Fracture Mechanics, Vol. 6,
No. 4, Dec. 1970, pp. 379-392.
[10] Wu, E. M., Journal of Applied Mechanics, Dec. 1967, pp. 967-974.
[77] Wu, E. M. in Composite Materials Workshop, S. W. Tsai, J. C. Halpin, and N. J.
Pagano, Eds., Technomic Publishing Co., Stamford, Conn., 1968, pp. 20-43.
[12] Bowie, O. L. and Freese, C. E., International Journal of Fracture Mechanics, Vol. 8,
1972, pp. 49-58.
[13] Sih, G. C , Paris, P. C , and Irwin, G. R., International Journal of Fracture Mechanics,
Vol. 1, No. 3, 1965, pp. 189-203.
[14] Walsh, P. F., Engineering Fracture Mechanics, Vol. 4, 1972, pp. 553-541.
[15] Gandhi, K. R., Journal of Strain Analysis, Vol. 7, No. 3, 1972, pp. 157-162.
[16] Cook, T. S. and Rau, C. A., " A Critical Review of Anisotropic Fracture Mechanics,"
Proceedings, International Conference on Prospects of Fracture Mechanics, Delft
University of Technology, Netherlands, 24-28, June 1974.
[17] Calcote, L. R., The Analysis of Laminated Composite Structures, Van Nostrand Rein-
hold Co., New York, 1969, pp. 29-31.
[18] Hahn, G. T., Rosenfield, and Sarrate, M., "Elastic-Plastic Fracture Mechanics,"
AFML-TR-67-143, Part III, Air Force Materials Laboratory, Jan. 1970.
[19] Rau, C. A., Pratt & Whitney Aircraft, unpublished data.

Vacuum Thermal-Mechanical Fatigue

Behavior of Two Iron-Base Alloys

REFERENCE: Sheffler, K. D., "Vacuum Thermal-Mechanical Fatigue Behavior of

Two Iron-Base Alloys," Thermal Fatigue of Materials and Components, ASTM
STP 612, D. A. Spera and D. F. Mowbray, Eds., American Society for Testing and
Materials, 1976, pp. 214-226.

ABSTRACT: Isothermal and thermomechanical fatigue tests conducted in ultrahigh

vacuum on Type 304 stainless steel and A286 alloy have shown significant effects of
frequency and combined temperature-strain cycling on fatigue life. Results of iso-
thermal tests at elevated temperatures in the creep range indicate reduced life with
decreasing frequency. Combined temperature-strain cycling further reduces fatigue
life with respect to isothermal lives measured at the maximum temperature of the
applied thermal cycle. Life reductions observed with in-phase thermal cycling (tensile
deformation at high temperature, compressive deformation at low temperature) are
attributed to grain boundary ratcheting effects resulting from unreversed tensile
grain boundary sliding. Results of out-of-phase tests (compressive deformation at
high temperature, tensile deformation at low temperature) were confused in Type
304 stainless steel by the occurrence of geometric instabilities in the hourglass speci-
mens. In the low ductility A286 alloy, where geometric instabilities were not ob-
served, out-of-phase life reductions are attributed to grain boundary cavitation re-
sulting from unreversed compressive grain boundary displacements (ratcheting) which
could not be accomodated by intragranular deformation of the high-strength low-
ductiUty matrix.

KEY WORDS: thermal fatigue, fatigue (materials), iron alloys, tests, stainless steels

Thermal fatigue is a recognized failure mode for structures such as

gas turbine components which operate with combined temperature and
strain cycling. Previous vacuum thermal fatigue work involving indepen-
dently controlled temperature and strain cycUng of two tantalum alloys^''

' Assistant project engineer. Materials Engineering and Research Laboratory, Pratt and
Whitney Aircraft, East Hartford, Conn. 06108.
'Sheffler, K. D. and Doble, G. S. in Fatigue at Elevated Temperatures, ASTM STP 520,
American Society for Testing and Materials, 1973, p. 491.
'Sheffler, K. D. and Doble, G. S., "Influence of Creep Damage on the Low Cycle
Thermal-Mechanical Fatigue Behavior of Two Tantalum-Base Alloys," Final Report, Con-
tract NAS3-13228, NASA-CR-121001, National Aeronautics and Space Administration,
May 1972.

214

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
SHEFFLER ON IRON-BASE ALLOYS 215

showed that tensile deformation imposed at a high temperature in the

creep range reversed by compressive deformation imposed at a lower
temperature (defined as in-phase cycling) causes a significant fatigue life
reduction as compared to isothermal cycUng, consistent with concepts of
strain range partitioning." This life debit was attributed to grain boundary
ratcheting effects involving high-temperature grain boundary sliding in
tension which is not reversed by compressive deformation imposed at
lower temperatures where grain boundary sliding does not occur. The
effect of out-of-phase cycling (compressive deformation at high tempera-
ture reversed by tensile deformation at low temperature) was not clearly
defined in the previous study because of geometric instabiUties which
confused interpretation of the experimental results. The present study ex-
tends the concept of in-phase grain boundary ratcheting to two iron-base
alloys (Type 304 stainless steel and A286) and provides a clearer interpre-
tation of out-of-phase grain boundary ratcheting effects observed in the
A286 alloy which does not exhibit the geometric instability.

Experimental Details
To separate the effects of temperature and environment on fatigue be-
havior, all tests were conducted in an ion-pumped ultrahigh vacuum
chamber at pressures below 1 x 10"' torr. Elevated temperature iso-
thermal fatigue tests were conducted at two frequencies (0.65 and 0.0065
Hz) to evaluate the contribution of frequency induced creep effects to the
fatigue process. Test temperatures for the isothermal tests were 1200°F
(922 K) for the Type 304 stainless steel and 1100 °F (866 K) for the A286
alloy. Tests with combined temperature and strain cycUng were of two
types, as illustrated in Fig. 1. The in-phase cycle involved isothermal
tensile strain imposed at a high temperature reversed by isothermal
compressive strain imposed at a low temperature. The out-of phase cycle
was similar except that the temperature-strain phasing was reversed. The
upper temperature for each material was the same as the isothermal
test temperatures. The lower temperature was 600°F (589 K) for both
alloys. Tests were conducted to failure (defined as a separation of the
specimen into two pieces) over a range of plastic strain amplitudes which
were measured by the width of hysteresis loop at zero load. Fractured
specimens were sectioned longitudinally and examined metallographically
to evaluate the character of the microstructural damage associated with
each of the applied cycle types.
Test procedures were essentially identical to those reported for pre-
vious tests.^^ The apparatus was designed to perform completely reversed
push-pull fatigue tests on hourglass specimens using independently pro-
*Manson, S. S. in Fatigue at Elevated Temperatures, ASTM STP 520, American Society
for Testing and Materials, 1973, p. 744.

1=/TN T™P. — 1/r\

=ii \/ .
ii N^ II II
II 1 1

1fi r i I
II 1 1
M 1 I

Time ——
a .™ b

FIG. 1—Schematic representation of the two types of thermal-mechanical cycles applied in

this study: (a) in-phase cycle (TCIPS) and (b) out-of-phase cycle (TCOPS).

grammable temperature and strain control. Temperature was programmed

using a thyrathron-controlled 50 kV a-c transformer for direct resistance
heating of the specimen, while diametral strain was controlled directly
using a linear variable differential transducer (LVDT) type extensometer
coupled to a programmable closed loop electrohydraulic servosystem. For
tests with combined temperature and strain cycling, the output of the
servoamplifier was electronically compensated for thermal expansion so
that net mechanical strain was controlled directly, which allowed tempera-
ture changes to be accomplished with mechanical strain held constant, as
shown in Fig. 1.

Results and Discussion

Test Results
Typical hysteresis loops generated in Type 304 stainless steel are shown
in Fig. 2. The isothermal loops are essentially symmetrical for both ma-
terials, with the load range developed at a constant diametral strain range
being larger at the higher test frequency. Hysteresis loops observed for
A286 were similar except that the inelastic strain ranges were smaller at
equivalent total diametral strain ranges because of the larger loads and,
consequently, larger elastic strains developed in the higher strength ma-
terial. Asymmetric hysteresis loops were developed with both types of
thermal fatigue cycles as a result of differences in flow stress at different
temperatures. In-phase cycling generated loops having a net compressive
stress, while out-of-phase cycling caused a mean tensile stress to be de-
veloped.

lOk SS
APPROXIMATE
TOTAL LONGITUDINAL
STRAIN RANGE
OF .008 I N . / I N .

a) ISOTHERMAL .65 Hz b) ISOTHERMAL .0065 Hz

+1)0

+30

H—I—h H 1—h
000 3000
DIAMETRAL STRAIN
/4IN./IN.
100

-30

-1*0

c) i n - p h a s e .0065 Ife d) out of phase .0065 ite

FIG. 2—Typical hysteresis loops observed for Type 304 stainless steel tested with various
strain-temperature cycles.

In previous thermal cycled tests on the tantalum alloys, geometry

changes occurred which involved thickening of the specimen in locations
above and below the minimum diameter with isothermal and in-phase
cycling and thinning of the specimen in these same locations with out-of-
phase cycling (Fig. 3). Similar geometry changes were observed in Type
304 stainless steel but not in the A286 specimens tested in the current pro-
gram (Fig. 4). While no definite explanation is available for the occur-
rence of geometric instabilities in one alloy and not the other, it may be
observed that the three materials which have shown pronounced geometry
changes in out-of-phase thermal cycling (two tantalum alloys and Type
304 stainless steel) are basically high ductility alloys, whereas A286 has a
relatively low ductility. It is not possible to determine from the available
results if this qualitative correlation represents a general trend or is merely
coincidental. Additional data on a wider range of materials would be de-
sirable to clarify this point.
The geometry changes did not appreciably influence the test results
from the isothermal and in-phase tests since the change left the area of
controlled maximum strain unchanged. However, the out-of-phase
geometry change had a significant influence on fatigue life since this
change caused failure to occur at a location separate from the original

t ORIGINA). tPECIHEN
3NT0UR ( 1 - 1 / 2 " RADIUS)

ORIGINAL HINIHUH
DIAMETER ( l A " )

ZONE OF HAXIHUH
HICROSTRUCTURAL
DAMAGE

FIG. 3—Schematic illustration of specimen geometry changes produced by thermal-

mechanical fatigue cycling: (a) isothermal and in-phase thermal cycling and (b) out-of-phase
thermal cycling.

minimum diameter (Fig. Ad) where strain was both unknown and uncon-
trolled. This qualification applies only to the out-of-phase Type 304 stain-
less steel results since geometry changes were not observed in the A286
alloy.
Fatigue life results for both alloys are summarized in Tables 1 and 2.
Type 304 stainless steel displayed cycUc strain hardening, while the A286
alloy exhibited cyclic strain softening. For both alloys, the majority of
hardening or softening occurred during the first few cycles of testing. The
stress ranges noted in Tables 1 and 2 represent stabiUzed loop sizes. For
the thermal cycled tests, where asymmetric loops were observed, both
tensile and compressive stress values are noted.
The longitudinal inelastic strain ranges noted in Tables 1 and 2 are
plotted against life in Figs. 5 and 6. These results indicate definite ef-
fects of both frequency and thermal cycling on fatigue life. Decreasing
frequency, which increases the potential for creep effects, reduces the iso-
thermal fatigue life of both materials. Combined temperature and strain
cycling causes further reductions of fatigue life in both materials as com-
pared to isothermal cycling at the maximum temperature associated with
the thermal cycle. For the A286 alloy, in-phase cycling is more damaging
than out-of-phase cycling. For Type 304 stainless steel, the reverse at
first appears to be true; however, the out-of-phase Type 304 results are
confused by the previously noted geometry change and are therefore not
considered meaningful.

I
If
If
II
il
!i|1
ft
S.i

i
5I

i
Copyright by ASTM Int'l (all rights reserved); Sun Dec 27 13:13:08 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
220 THERMAL FATIGUE OF MATERIALS AND COMPONENTS

-O r- (N ^o w-i I
<N -"J- "j-i M (*) >0

— 00 r- (S I
oo 00 o f*! <*)

00 r-i v"! 1^ ^ oo
(j\ oo r- <N — CT\
r^ (N (N r-) rs —

O 0\
r^ » ^ r4
3 S

\0 ^O ^ \0 '"O

fl — 0\ »N —< 00
** ^ "1 m m (N

"•O ^O r*! f«i oo

3
.50

o •= I S
iiiiiiiiiii
ON 0 \CT\ON 0^

I
BQ
<

888 88 8 8 8 888

tr\ *r\ tn

mmm
^i < <
< < <
< < M
< <
<< << << < < -t
Z.

^ O ^ 0\ w^
irj ry <*) r- >0
. . '. . : o o oo r- r~-

*N Q ^ •* r^
0\ '•D a^ tri —
r- ^ I/-, o oo

W^ Tf C-I -^ —

v^ <n \ 0 f^ oo
-H Ov oo K-l —

I
I o8S
d d o
08
do
088
o d d
08
dd

T ^ O^ ON
00 00 00
u-i ir\ in
O^
00 00
*n v~i
ON

0 0 0 0 0 0 OOOO OOOOOO 0 0 0 0

888 88 8 8 8 88

«i «i w-i in vt

0 0 0 0 0 0 0 0 0 0

II mHiiiii
11

"T 1 1 1 1 I I ~i—I—I I I IJ

In-phase 60O-I20OF c y c l e , O.OO65 Hz

ISOTHERMAL 1200°? Cycle Q0065 HZ

SOTHERMAL I 2 0 0 ° F c y c l e , 0 . 6 5 Hz

Out-of-phase 600-1200F c y c l e , O.OO65 m

I I I I I II ' ' 'I

lo'
CYCLES TO FAILURE

FIG. 5—304 stainless steel fatigue life results.

"1 1—I—I I I I I I 1 1—I—I I I I I

ISOTHERMAL 1 lOO^F c y c l e , ( 8 6 6 ' K ) O.OO65 HZ

ISOTHERMAL IIOflOF c y c l e , 0,65 HZ

In-ph>ise 600-llOOF cycle

0.0065 Hz

J I I ' l l J I I I I I I Ii J I I I I I I

CYCLES TO FAILURE

FIG. 6—A286 fatigue results.

Microstructural Observations
Microstructural damage observed in the fatigue tested specimens varied
with both cycle and test material, as shown in Figs. 7-9. In high fre-
quency cycling, both alloys exhibit mixed intergranular and transgranular
cracking (Fig. 9) with minimal microstructural damage localized at grain
boundaries (Figs. So and 9a). As shown in Fig. 9d, some of the secondary
cracks found in A286 are associated with carbide inclusions. While car-
bide cracking is assumed to have some influence on the cyclic life of

*>l^i^

20K.n

(a) Isothermal 0.65 Hz.

(b) Isothermal 0.0065 Hz.
(c) In-phase cycle.
(rf) Out-of-phase cycle.
FIG. 7—Microstructural damage found in Type 304 stainless steel tested with various
cycles. Tensile axis vertical.

tf^mil

^ 3 ^ w - • •;•;*.•

Em.*' ^r.
(a) Isothermal 0.65 Hz.
(b) Isothermal 0.0065 Hz.
(c) In-phase cycle.
(rf) Out-of-phase cycle.
FIG. 8—Microstructural damage found in A286 tested with various cycles. Tensile axis
vertical.

A286, it is not thought to influence the life differences caused by dif-

ferent types of cycles, since carbide associated cracking was observed in
A286 with all cycle types studied.
Reduced frequency isothermal cycling causes a greater degree of local-
ized grain boundary damage in the Type 304 stainless steel (Fig. lb), with
the fracture mode being exclusively intergranular in both alloys.
In-phase thermal cycling causes severe grain boundary cavitation in
both alloys (Figs. Ic and 8c). This effect is attributed to the previously

m^:.

(a) Type 304 stainless steel, 0.65 Hz.

(b) Type 304 stainless steel, 0.0065 Hz.
(c) A286, 0.65 Hz.
(d) A286, 0,65 Hz.
FIG. 9—Secondary cracks found in various fatigue specimens. Tensile axis vertical.

discussed grain boundary ratcheting mechanism, which is assumed to be

responsible for the significant life reduction associated with in-phase
cycling.
Significant differences were found between the types of damage devel-
oped in out-of-phase A286 and Type 304 stainless steel specimens. Dam-
age in Type 304 alloy is characterized by extensive etch pitting concen-
trated near grain boundaries (Fig. Id). This pitting presumably repre-
sents areas of high dislocation density associated with large localized

intragranular strains required to accomodate accumulated compressive

grain boundary displacements (compressive ratcheting). In the higher
strength A286 alloy, where unreversed compressive grain boundary dis-
placements cannot be fully accomodated by intragranular deformation,
significant intergranular cavitation is found (Fig. &d). As with in-phase
cycling, this cavitation is presumed to be the cause of the life reduction
caused by out-of-phase cycUng of A286.

Summary
Elevated-temperature, low-cycle thermal-mechanical fatigue tests con-
ducted in vacuum on Type 304 stainless steel and A286 alloy show sig-
nificant effects of frequency and of combined temperature-strain cycling
on fatigue life. At temperatures in the creep range, fatigue lives of both
materials are lower at 0.0065 than at 0.65 Hz. Metallographic examina-
tion of fractured specimens indicates mixed mode (intergranular and
transgranular) fracture at the higher frequency and exclusively inter-
granular fracture at the lower frequency. In-phase thermal cycling (ten-
sion at high temperature and compression at low temperature) causes
large life reductions in both materials. These Hfe reductions are attributed
to grain boundary cavitation caused by unreversed grain boundary sliding
(grain boundary ratcheting). Out-of-phase thermal cycling (tension at low
temperature and compression at high temperature) also causes large cyclic
life reductions in both materials, but the results from Type 304 stainless
steel are confused by the occurrence of geometric instabilities. In the
A286 alloy, out-of-phase life reductions are attributed to cavitation dam-
age resulting from unreversed compressive grain boundary displacements
(compressive ratcheting) which cannot be fully accommodated by intra-
granular deformation.

A cknowledgments
Support for this work was provided by the NASA Lewis Research Cen-
ter, Contract NAS-3-6010, with Dr. G. R. Halford as program manager.
Assistance with the experimental effort, provided by J. W. Sweeney, is
gratefully acknowledged.

Instability Effects in
Thermal Fatigue

REFERENCE: Coffin, L. F., "Instability Effects in Thermal Fatigue," Thermal Fa-

tigue of Materials and Components, ASTM STP 612, D. A. Spera and D. F. Mowbray.
Eds., American Society for Testing and Materials, 1976, pp. 227-238.

ABSTRACT: This paper considers shape changes which develop progressively in

hourglass specimens subjected to fully reversed cyclic plastic strain applied in a variety
of mechanical and thermal waveforms. While the strain at the minimum diameter is
controlled, off-center locations subjected to cyclic plastic strain undergo ratcheting.
Two major off-center shape changes have been observed: barreling and necking. The
paper discusses the reasons for these changes.

KEY WORDS: thermal fatigue, strains, necking

It is observed that uniaxially loaded, cyclically strained hourglass speci-

mens change their shape during the progress of fatigue testing, particu-
larly at high-cyclic strains [1].^ At elevated temperature with mixed ther-
mal and mechanical cycling, distinct differences in shape occur, depend-
ing on the phase relationship between the mechanical and thermal effects
[2]. This paper is addressed to the question and discusses the broader im-
plications to the thermal fatigue problem.
Earlier [7], the author investigated the phenomenon in room-tempera-
ture tests. The softer metals, notably 1100 aluminum and, to a lesser ex-
tent, OFHC copper, were particularly prone to progressive off-center bar-
reUng when the minimum diameter was cycled between prescribed limits
[3]. From an analysis [I] of the stress and strain limits at both the mini-
mum diameter and off center, it was shown that the instability came from
unequal loads in tension and compression. When the minimum diameter
was controlled between equjil diametral strain limits of opposite size, a
compressive stress bias and a progressive diameter increase developed off
center at plastically deforming sections.
' Mechanical engineer and Coolidge Fellow, Corporate Research and Development, Gen-
eral Electric Co., Schenectady, N.Y. 12301.
^The italic numbers in brackets refer to the list of references appended to this paper.

227

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
228 THERMAL FATIGUE OF MATERIALS AND COMPONENTS

Conway, Stentz, and Berling [4] have performed continuous cycling

and hold time tests on hourglass specimens of zirconium-copper in argon
at 538 °C, also utilizing diametral strain control. Figure 1 shows the occur-
rence of off-center necking during a tension hold and off-center barreling

tt Oil-- -

m"^'-^
a-2-28 •i-2-1^7 H-2-50
538°v
6.1,-^ i.ti:^
no hold time • 56 s e c ( t e n s . ) 56 sec(eoinp.}
Nf= 176 Nf= 1,152 N^= 3,180

FIG. 1—Test specimens of zirconium-copper tested in argon at 538°C. Continuous cycling

and strain-hold time tests. See RefA.

resulting from compression hold. Here rfo refers to the position of the
diametral extensometer. Using similar specimen geometries and strain
control procedures, the author has shown that AISI 304 stainless steel,
tested at 650 °C with unequal (sawtooth) total longitudinal strain ramp
rates, developed similar shapes [5]. With a plastic strain range, A£p, of
0.02, a period of 10 min, and a ramp whose tension going time was 0.1
min, followed by a compression going ramp time of 9.9 min, off-center
necking was found, while reversing the ramp rates produced off-center
barreling.
Shape instability was found in thermal fatigue tests performed on axi-

ally constrained, uniform gage length tubular specimens subjected to al-

ternate heating and cooling [6,7]. Those tests revealed a progressive thin-
ning of the wall near the ends of the specimen as cycling proceeded, par-
ticularly under the extreme conditions of cyclic temperature (100 to 600°C).
Garden and his associates have also studied this problem [8-10]. Sheffler
[2] reported shape instabilities in hourglass geometries of AISI 304 stain-
less steel and A286 at elevated temperatures in high vacuum. (The temper-
ature-mechanical strain program was as shown later in Fig. 7a (in-phase) or
7b, (out-of-phase.) Figure 2 shows the shape changes observed for AISI
304 stainless steel. A286, on the other hand, showed no shape change.
Sheffler reports that prior tests [77] on two tantalum alloys showed simi-
lar instabilities.

ORIGINAI. SPECIMEN
CONTOUR ( 1 - 1 / 2 " RADIUS)

ORIGINAL HINIHUn
DIAMETER ( l A " )

ZONE OF MAXIMUM
MICROSTRUCTURAL
DAMAGE

a) ISOTHERMAL AND IN-PHASE b) OUT-OF-PHASE

THERMAL CYCLING THERMAL CYCLING

FIG. 2—Schematic of shape changes observed with in-phase and out-of-phase thermal and
mechanical cycling of AISI 304 stainless steel. See Refl.

It is clear that, aside from being a curious observation, the phenomenon

of shape instability has practical importance in the interpretation of test
results. It may have other important implications, as will be discussed
later.

Monotonic Strain Distribution in Hourglass Geometries

In developing the mechanism for shape instability under cyclic strain.

we consider first the strain distribution in an hourglass-shaped geometry

under monotonic loading as a function of such parameters as strain hard-
ening, strain rate, and temperature. In Fig. 3a, the axial plastic strain dis-
tributions are compared for two distinct materials at the same tempera-
ture and strain rate, but of differing strain hardening coefficients. In Fig.
3b, a given material is tested at two different strain rates, while, in Fig.
3c, the response of the material at two temperatures is considered. As a

• (a) EFFECT OF
Z STRAIN HARDENING

^ (b) EFFECT OF
Z STRAIN RATE
SENSITIVITY

• - ( c ) EFFECT OF
Z TEMPERATURE

FIG. 3—Strain distributions in hourglass-shaped specimen with monotonic loading, show-

ing effect of strain hardening, strain rate, and temperature.

result of the changes in the flow stress curves for the parameters consid-
ered here, the axial strain distribution may be diffuse or locaUzed with ap-
plication of axial stress. The locaUzation or diffuseness can also be ob-
served by comparing, in each case, the axial strains at Position 2 for equal
strains at 1.

Off-Center Dimensional Changes

Assuming now an hourglass-shaped specimen subjected to controlled
diametral strain at Position 1, Fig. 3, we now consider how dimensional
changes can occur at the off-center Position 2. In doing so, we shall relate

the axial strain distributions shown in Fig. 3 to the separate legs of spe-
cific hysteresis loops. Three cases will be considered: (a) isothermal bal-
anced cycling (Fig. 4), (b) isothermal unbalanced cycling (Figs. 5,6), and
(c) thermomechanical cycling (Fig. 7c).

IT, IC

I^'PI
BALANCED ISOTHERMAL LOOP

•'ZCi ^-ZTl

OFF-CENTER BARRELING

FIG. 4—Cyclic stress-strain behavior and hysteresis loop for balanced isothermal cycling
showing conditions for off-center barreling.

Isothermal Balanced Cycling

Consider the stable hysteresis loops of stress versus plastic strain and
the waveform illustrated in Fig. 4 for a specimen controlled at its mini-
mum diameter Position 1. The two legs of the loop are closely similar
since the temperature, the strain ramp rates, and the strziin Umits for each
leg are of equal magnitude. We next consider the off-center Position 2.
As indicated earlier, the tensile load on the specimen is less than the com-
pressive load in order to equate the tensile and compressive true stresses at
Position 1. Because the area at 2 is larger and the stress is less, the area
correction for stress at 2 is small. Hence, the tensile stress at 2, 2T, is less
than the corresponding compressive stress, 2C. Referring to the loop of

ISOTHERMAL SLOW-FAST LOOP

OR
IN-PHASE THERMAL FATIGUE LOOP

'•2CI ' II'^Parl

Pari
OFF-CENTER BARRELING

FIG. 5—Cyclic stress-strain behavior and hysteresis loop for unbalanced cycling produced
by isothermal slow-fast or in-phase thermal cycling showing conditions for off-center barrel-
ing.

Fig. 4, these points are identified. Also shown in Fig. 4 is the cyclic stress-
cyclic plastic strain amplitude response of the material, |oJ versus le^pi.
The Points 2T and 2C are identified. The cyclic stress-strain curve shown
is the locus of stable loop End Points 1, either in tension or compression,
for all strain ranges. Since the loop is balanced, the tension and compres-
sion curves are identical. Now since |2C| > |27l, Itzpzcl > k^pzrl- Thus, the
loop for Point 2 does not close, but rather a slightly excessive axial com-
pressive strain is produced. This increment of strain accumulates with
each cycle leading to progressive off-center barreling.
The ratcheting process described is difficult to predict quantitatively.
As the compressive strain accumulates at 2, the flow stress increases and
the increment of plastic strain decreases because of strain hardening. Cy-
clic strain softening limits strain hardening, and a steady-state ratcheting
rate is reached.
We can qualitatively determine the response of various materials to

/HIGH €
'T \ L O W TEMP

LOW €
IC 1 HIGH TEMP

l^'J
ISOTHERMAL FAST-SLOW LOOPS
OR
OUT-OF-PHASE THERMAL FATIGUE LOOP

ISTI > I'^PacI

OFF-CENTER NECKING

FIG. 6—Cyclic stress-strain behavior and hysteresis loop for unbalanced cycling produced
by isothermal fast-slow or out-of-phase thermal cycling showing conditions for off-center
necking.

shape changes under these conditions. First, a high strain hardening coef-
ficient is desirable in order to enhance the off-center strain range. Sec-
ondly, pronounced cyclic strain softening is necessary in order that the
strain hardening effects which would otherwise limit the ratcheting strains
be minimized. Because of these considerations, 1100 aluminium is particu-
larly sensitive to shape changes [7,5].
Isothermal Unbalanced Cycling
We now consider specimen shape changes for isothermal loops of un-
equal ramp rates, called slow-fast (Fig. 5) or fast-slow (Fig. 6). We first
note the hysteresis loop generated for the slow-fast ramp condition illus-
trated and observe that its unbalanced nature arises from the material's
response to strain rate. Fig. 3b [5], The cyclic stress-plastic strain ampli-

MECHANICAL STRESS
STRAIN OR TEMR
"ITEMP

TIME
,STRAIN

IN PHASE
STEPPED TEMPI

I 1 I TEMP
I

.STRAIN
L_J I I
OUT-OF-PHASE
STEPPED TEMP

DECREASING
TEMP
v

^STRAIN

OUT-OF-PHASE^
RAMPED TEMP

FIG. 7—Waveshapes for mixed mechanical and thermal cycling for in-phase and out-of-
phase conditions.

tude curves for the tensile and compressive stress portions of the loop (lo-
cus of Points i r and \C) are also shown. Since the tensile portion of the
cyclic stress-strain curve is controlled by the slow tension-going part of
the loop, it differs distinctly from the compressive portion, as indicated.
Values of stresses and plastic strain are shown for the off-center Position
2 as indicated, the stress assumed to be 90 percent of the stress at Position
1. No correction has been made for load differences at Position 1 between
tension and compression nor for differences in strain rate between Posi-
tions 1 and 2. Strain rate difference effects between 1 and 2 can be
avoided if the cyclic stress-plastic strain curve is obtained under condi-
tions of constant frequency. From Fig. 5 we note that |£,p2cl > kzparl; that
is, for each closed cycle of strain at Position 1, there is a net compressive
strain increment developing at Position 2, causing the specimen to become
progressively fatter there. Simultaneously, there will be a net shortening
of the specimen.
Conversely, the fast-slow loop produces just the opposite effect. Fig. 6.
The unbalanced cyclic stress-strain loci for Point 1 are now reversed, the
tensile curve being generated at high strain rate. The off-center stresses
are located at 90 percent of the end point values and, because of the slope
differences, |£z„2rl > k zp2C\ Thus, at Position 2, there is a progressive

elongation and necking, causing the configuration described earlier. The

behavior may not be as pronounced as for off-center barreling, for when
the stresses at Position 2 are corrected for the load differences between
tension and compression at Position 1, IC increases and IT decreases,
causing less strain unbalance in each cycle. In fact, with fast-slow loops
which are only slightly unbalanced, this later correction may control, and
off-center barreling may occur.
Thermal-Mechanical Cycling
A similar approach can be used to describe the instability that develops
when stepped temperatures but similar strain rates are employed in the
two legs of the loop, shown in Figs, la and b. Referring to Figs. 5 and 6,
substitution of high temperature for low strain rate and low temperature
for high strain rate leads to the same behavior as mentioned previously.
For in-phase cycling, that is, for high-temperature tension going, the cy-
clic stress-strain amplitude curve for tensile stresses is lower and flatter
than that for compression. Hence, at remote positions, that is, 2, the
strain values, indicated by 27 and 2C, are such that \trp2c\ > l^zpiA, and
off-center barreUng will occur. On the other hand, during out-of-phase
cycling. Fig. 6, the tensile part of the cycUc stress-strain curve is higher
and steeper such that, at Position 2, |£,p2rl > \^zp2c\- Consequently, off-
center necking develops at Position 2. These predictions fit the observa-
tions of Sheffler, Fig. 2.
The case of differing strain rates as well as temperatures can be simi-
larly treated. From the arguments presented, the largest imbalance is de-
veloped for the combination of high temperature and low strain rate in
one leg and low temperature and high strain rate in the other. The shape
change (off-center necking or barreling) will depend on the phasing with
mechanical strain. Combinations of high temperature and high strain rate
for one leg and low temperature and low strain rate are not predictable
because of the offsetting effects of temperature and strain rate.

Special Cases
Earlier, the case of end thinning in axially constrained and thermally
cycled tubular specimens was referred to [6]. This situation requires a dif-
ferent treatment since the end thinning (and central thickening) was occur-
ring in a region of uniform stress. Nonuniform temperature along the
tube length, especially near the ends, was found and was particularly
severe at the higher temperature (600 "C) but was insignificant at the lower
temperature (100 °C). The end thinning and center thickening can be
traced to this axial temperature distribution. Since compression exists at
high temperature, greater thickening in the wall occurs there rather than
at the ends where the flow stress is greater. At the lower temperature and

the accompanying tensile stress, the ends are thinned further because of
the now developed nonuniform wall thickness. There is little to prevent
the process from continuing.
Another case mentioned earlier is that reported by Conway, Stentz, and
Berling [4] for isothermal hold-time tests on zirconium-copper in argon at
538 °C. It should first be noted that this temperature is very high for cop-
per alloys. Secondly, the shape changes were reported to occur at 1.4 per-
cent total strain range, but not at 5.0 percent. Hence the behavior is not a
general phenomenon. Because of the high temperature and very low strain
hardening, most of the plastic deformation during the ramping of the
strain is localized to a region very near the specimen center, as illustrated
by Fig. 3. Thus, at off-center Position 2, the cyclic plastic strain com-
ponent is very small. The tensile hold period requires that the total com-
puted longitudinal strain at Position 1 remain constant. The specimen is
subjected to a prolonged, relaxing tensile stress, allowing tensile creep de-
formation to occur at Position 2. Correspondingly, there will be elastic
unloading at 2 consistent with the constant total (elastic plus plastic)
strain conditions at Position 1. With the next strain reversal, again little
plastic deformation occurs at 2, but another creep increment develops
during the hold period. Hence, progressive necking develops at 2. As the
diameter becomes smaller and the stress increases at 2, the creep effects
accelerate, although no indication of this event is evident at the control
position. However, the specimen becomes progressively longer.
In the case of compression hold, the same process takes place, but in
compression. Off-center barreUng develops as a result of repetitive incre-
ments of compressive creep. Here, the stress decreases with increasing
barreling, so that the process tends to slow down at 2. However, the
shape change of the specimen causes the bulging process to move down
closer to the specimen center. Eventually the barreling extends almost to
the center, and the profile becomes highly localized. The highly concen-
trated necking and barreling is clearly seen in Fig. 2, together with spec-
imen shortening.
In the case of the hold experiments performed at higher strain ranges,
the zone of plastic deformation extends further from the center. The dif-
ferences in rapid straining in one direction and creep in the other direction
are now less, and the shape changes are smaller. Additionally, there are
fewer cycles.^

' Note added in proof: attention should also be called to the work of Skelton [J3] in which
uniform gage length test specimens of 20Cr/25Ni/Nb steel were subjected to longitudinal
strain control at 750 °C with tension and compression hold periods. It was found that tension
hold periods led to barreling at the midgage length position and necking near the end fillets,
while compression holds produced midlength necking and barreUng near the end fillets.
These observations are consistent with those of Conway, Stentz, and Berling [4].

Discussion
The general condition that appears necessary for shape changes and in-
stabilities in mechanical and thermal cychng is a positional variation or
gradient in the cyclic stress-strain field caused by geometry, temperature,
or material variations. We have cited several examples of geometry varia-
tions where hourglass specimen geometries Eire used and one example of
temperature variation, that of the axially constrained and thermally cycled
tube. Instability can be minimized by maintaining axial uniformity both in
geometry and in temperature distribution wherever cyclic plastic strain is
present.
Although the observations and interpretations cited here relate to spec-
ific bulk geometries, it is interesting to speculate as to whether similar
geometric changes can occur on a much finer scale. Two intriguing ques-
tions come to mind. One relates to whether the intense strain locahzation
associated with slip-band extrusion and crack initiation develops as a con-
sequence of progressive microstructural changes in shape of surface
grains. Another deals with crack growth and the shape of the crack tip as
a function of the wave shape employed. Relative to the latter question,
substantial differences in the shape of mildly notched specimens occur de-
pending on the phase relationship between thermal and mechanical load-
ing (Fig. 5). Similar shape changes can be envisaged when the notch
sharpens sufficiently to become a propagating crack. Various investigators
[2,5] have reported substantial differences in life as a function of wave
shape, other conditions being equal. Can these differences be related to
differentes in crack tip shape changes and thus on crack propagation?
Relative to crack nucleation, McClintock's [12] quantitative model of in-
stability to explain surface roughness and crack nucleation should be re-
called. That work was inspired by observations of roughenmg on cyciea
plasticine models. Northcott and Baron's studies on thermal fatigue can
also be cited, where narrow edges of wedge-shaped specimens were alter-
nately heated and cooled to produce a regular pattern of deep grooves,
initiating crack propagation. Crazing in thermeil fatigue and local surface
roughening in isothermal fatigue may be closely related to the hourglass
specimen shape instabilities reported here. The craze-cracking problem is
unquestionably more compHcated, dealing with cyclic phase transforma-
tions, interplay with oxidation processes, two-dimensional stresses, etc. It
is, however, basic to many thermal fatigue problems and should be
studied in more detail.

References
[/] Coffin, L. F., Transactions, American Society for Mechanical Engineers, Journal of
Basic Engineering, D, Vol. 82, 1960, p. 671.

[2] Sheffler, K. D., "Vacuum Thermal-Mechanical Fatigue Testing of Two Iron Base Al-
loys High Temperature Alloys," NASA-CR-134524, National Aeronautics and Space
Administration, Lewis Research Center, Cleveland, Ohio, Jan. 1974.
[3] Raymond, M. H. and Coffin, L. F., Acta Metallurgica, Vol. 11, 1963, pp. 801-807.
[4] Conway, J. B., Stentz, R. H., and Berling, J. T., "High Temperature, Low-Cycle Fa-
tigue of Copper-Base Alloys in Argon," NASA-CR-121260, National Aeronautics and
Space Administration, Lewis Research Center, Cleveland, Ohio, Aug. 1973.
[5] Unpublished General Electric Co. research.
[6] Coffin, L. F., Transactions, American Society of Mechanical Engineers, Vol. 76, 1954,
pp. 931-950.
[7] Coffin, L. F., Transactions, American Society of Mechanical Engineers, Vol. 79, 1957,
p. 1637.
[8] Carden, A. E., "Thermal Fatigue—An Analysis of the Experimental Method," ORNL
TM-405, Oak Ridge National Laboratory, Oak Ridge, Tenn., 1963.
[9] Thomas, W. S., "An Investigation of the Stages of Deformation of a Thin-Wall Tube
Caused by Cyclic Thermal Stress," M.S. Thesis, University of Alabama, University,
Ala., Aug. 1961.
[10] Hernandez, J. A., "An Analysis of the Thermal Fatigue Characteristics of Inconel,"
M. S. Thesis, University of Alabama, University, Ala., 1964.
[77] Sheffler, K. D. and Doble G. S. in Fatigue at Elevated Temperature, ASTM 520, Amer-
ican Society for Testing and Materials, 1973, pp. 491-499.
[12] McClintock, F. A. in Fracture of Solids, Interscience Publishers, New York, 1963, pp.
65-102.
[13] Skelton, R. P., Materials Science and Engineering, Vol. 19, 1975, pp. 25-30.

Life Prediction of Thermal-

Mechanical Fatigue Using
Strainrange Partitioning

REFERENCE: Halford, G. R. and Manson, S. S., "Life Prediction of Thermal-

Mechanical Fatigue Using Strainrange Partitioning," Thermal Fatigue of Materials
and Components, ASTM STP 612, D. A. Spera and D. F. Mowbray, Eds., American
Society for Testing and Materials, 1976, pp. 239-254.

ABSTRACT: This paper describes the features of the method of strainrange partition-
ing that make it applicable to the life prediction of thermal-mechanical strain-cycling
fatigue. An in-phase (230 to 760 °C) test on Type 316 stainless steel is analyzed as an
illustrative example. The method utilizes the recently proposed step-stress procedure
of experimental partitioning, the interaction damage rule, and the life relationships
determined at an isothermal temperature of 70S°C. Implications of the present study
are discussed relative to the general thermal fatigue problem.

KEY WORDS: thermal fatigue, intergranular fracture, plastic strain, creep strain,
life prediction, strainrange partitioning, stainless steels

Several of the papers in this volume relate to fatigue data obtained by

thermal-mechanical strain-cycling (TMSC) involving the simultaneous
independent variation of both strain and temperature. TMSC simulates
many of the features of the general thermal fatigue problem, yet retains
the relative simplicity and ease of data gathering and interpretation associ-
ated with axial strain-cycling of smooth laboratory specimens. In a TMSC
test, the temperature, strain, and stress can be measured directly at any
point within a cycle. Furthermore, any two of these variables can be con-
trolled independently with time.
Cyclic lives associated with thermal fatigue and TMSC are typically
observed to be lower than those associated with comparable isothermal
cycles. Many investigators have concluded that some form of creep dam-
age is responsible for such adverse effects. As a result, a number of high-
' Engineer, Lewis Research Center, Nationed Aeronautics and Space Administration,
Cleveland, Ohio 44135.
^Professor, Case Western Reserve University, Cleveland, Ohio 44106.

239

Downloaded/printed
Copyright 1976 b y Aby
S TM International www.astm.org
University of Washington (University of Washington) pursuant to Li
240 THERMAL FATIGUE OF MATERIALS AND COMPONENTS

temperature fatigue life prediction approaches [J-Sy have evolved over

the past two decades in attempts to deal rationally with the pernicious
nature of creep damage. The earliest attempts did not consider that creep
damage might be encountered in more than one form.
In 1968, Spera [5] proposed upper and lower bound criteria for creep
damage for use in calculating isothermal and thermal fatigue life. The
lower bound criterion based on "monotonic rupture" best fit thermal
fatigue results since they tended to exhibit creep damage in only one direc-
tion. Isothermal results involving creep damage in both directions usually
followed the upper bound or "cyclic rupture" criteria. At that time,
however, a methodology was not considered for blending between mono-
tonic and cyclic damage in a general cycle.
The method of strainrange partitioning [8] was proposed in 1971 and
provides a methodology for handling any strain cycle composed of creep
and plastic strains, including those cycles which give rise to monotonic
and cycUc creep damage. Within the framework of this method, damage
is contributed by each of four partitioned strainranges in proportion to
the amount of each present within a cycle. A damage rule, such as the
interaction damage rule [9], provides the means for synthesizing these
damage contributions. The partitioned inelastic strainranges are defined
as:
Afpp = tensile plastic strain that is balanced by compressive plastic
strain.
A£p<. = tensile plastic strain that is balanced by compressive creep strain.
A£<.p = tensile creep strain that is balanced by compressive plastic strain.
Ae^c = tensile creep strain that is balanced by compressive creep strain.
These strainranges are commonly encountered in a number of applica-
tions considered by others [1,5-7,10-14] in dealing with a variety of
creep-fatigue problems. Strainrange partitioning simply recognizes these
components as generic in that the strain response to any arbitrary history
of reversed strain can be synthesized from contributions of these com-
ponents.
Several features of the method are of special advantage in the treatment
of thermal-mechanical cychng. The purpose of this paper is to discuss
these features and to illustrate possible application with an example.

Features of the Strainrange Partitioning Method

Although Atpp, Aspc, l^cp, and Ae^,; are the four generic strainrange com-
ponents which comprise all hysteresis loops, the amount of each com-
ponent present in any one hysteresis loop must be determined by a process
referred to as "partitioning." Several procedures have been developed for
' The italic numbers in brackets refer to the list of references appended to this paper.

accomplishing this process [9,15]. In the present discussion we shall limit

consideration to only the "step-stress" method which is an experimental
procedure whereby the hysteresis loop is traversed, stopping at a selected
number of stress levels to determine creep rates. In Ref 15, this procedure
was developed in connection with hysteresis loops at isothermal con-
ditions. However, the same basic partitioning procedure can be applied
in thermal-mechanical cycles involving arbitrary temperature variations
within the cycle. Several advantages accrue when the analysis is performed
according to the strainrange partitioning framework and through the step-
stress partitioning procedure.

Accounting for Details of Cycle, Including Creep Induced by Compres-

sive Stress
An important aspect of strainrange partitioning relevant to the subject
of thermal fatigue analysis is that it is possible to distinguish between
the severity of the different types of strainranges that can be commonly
generated during such cycling. For example, if the strain is "in-phase"
with the temperature (rising tensile strain while temperature is increasing),
the strain-range component developed will be of the AECP type. "Out-of-
phase" cycling (rising compressive strain with increases in temperature)
is conducive to the development of a Acp^ component of strainrange.
Since the method associates different effects with each of these strain-
range types, it not only provides a means for recognizing the importance
of the compressive portion of the cycle (which some alternative methods
do not; for example, that currently specified by the American Society
of Mechanical Engineers (ASME) Boiler and Pressure Vessel Code Case
1592 [16]), but it also provides a basis for distinguishing between the
details of strain and temperature variation within the cycle that can have
significant effects on fatigue Ufe.

Separation of the Two Distinct Roles of Temperature

Within the framework of strainrange partitioning, temperature is as-
sumed to exert an influence on low-cycle fatigue in two independent ways:
its strong influence on flow behavior (constitutive relationships), and, to
a much lesser degree, its influence upon failure behavior (life relation-
ships). A more detailed discussion of these two aspects is given in Ref 17.
Any influence the temperature has on the flow behavior is manifested
in the size and shape of the hysteresis loop that develops during cycUng.
When such a loop is partitioned, the amount of creep strain present will
be a direct function of the time available and the thermal activation pro-
vided by the temperature. Once the hysteresis loop has been partitioned
and the pertinent components of strainrange Acpp, Atcc, and AE^P, or Atpc

determined, the method makes use of an appropriate damage rule to

combine the effects of two or three strainranges present. In applying this
rule, the life relations for each component strainrange become the impor-
tant material characterization par2mieters.
For certain materials—for example, Type 316 stainless steel and 2'4Cr-
IMo, which are materials of great technological importance in nuclear
pressure vessel applications—the partitioned strainrange versus life rela-
tionships are not sufficiently dependent on temperature [17] to require
special consideration. Advantages accrue in these cases. First, there is a
reduction in the amount of material testing needed to characterize the
life relationships. Although this is not too important when only short
time results are needed, it is of great importance when very long test times
are deemed necessary. Second, temperature insensitivity removes the
question of selecting a temperature within the spectrum covered by a
temperature variation throughout a cycle that is to be regarded as the
most important for the choice of pertinent life relationships.
In treating materials for which there may be a strong dependence of the
life relationship on temperature, this advantage is, of course, lost. For
such materials, alternate criteria could be assumed. For example, one
could use the Ufe relationships at the maximum temperature [16], the
relationships at the temperature which produces the lowest lives [18,19],
or the temperature at which the dominant strainrange type is imposed.

Provision of Upper and Lower Bounds on Life

The method lends itself well to the determination of bounding values
on life. Thus, once the inelastic strainrange ha^ been determined, a lower
bound on cycUc Ufe can be obtained by assuming that all of the strain-
range is of the most damaging type (for most materials, Asop, but for
some, Atpe). Should the life relationships vary with temperature, the lower
bound could, of course, be determined at the temperature producing the
lowest lives. Likewise, an upper bound on life can be ascertained by as-
suming that the entire strainrange is of the most benign type (usually Atpp)
and evaluated at the temperature producing the greatest Uves. If the
bounding values are acceptable, a detailed analysis to determine the actual
hysteresis loops and to partition them becomes unnecessary; if not, the
need for a more detailed study is indicated. This feature has special merit
to designers who may wish to apportion design effort according to the
needs of a particular situation rather than always making detailed analy-
ses. It is also of significance to codifying bodies in specifying under what
conditions minimal analyses will be required (that is, when the most pes-
simistic estimate of life associated with the lower bound can be tolerated)
and when detailed analyses are needed (that is, when the designer is moti-
vated to take full advantage of the potential life of the part).

Example Analysis

Description of the Problem Treated

As a demonstration of the applicability of the method of strainrange
partitioning to the thermal-mechanical strain-cycling problem, an axially
loaded, tubular specimen of AISI Type 316 stainless steel was subjected
to the in-phase conditions listed in Table 1.
TABLE 1—Material, specimen, and test conditions for sample problem.

Material: AISI Type 316 stainless steel, hot rolled and mill annealed, 19 mm (0.75 in.) di-
ameter bar stock supplied by Viking Steel Co., Cleveland, Ohio. The heat number,
exact chemical composition, and annealing details are unknown. Tensile and creep-
rupture properties at 705 "C are given in Ref 9. An indication of the grain structure
can be found in Fig. 5a.
Specimen: Tubular, hourglass test section [20\, specimen number AYY-304.
Extensometer: Diametral [20].
Strain: Diametral strain cyclically ramped linearly with time between limits resulting in an
' axial total mechanical strainrange of 0.00923. The apparent thermal component of
strain (0.01381) was measured by temperature cycling the specimen under a cori-
trolled zero load condition. This thermal component was subtracted from the total
measured strain to determine the mechanical component. The elastic strainrange
was subtracted from the total mechanical strain range to get the inelastic strain
range of 0.00616. The cyclic period was 30 min.
Temperature: Silicon carbide internal heating element, temperature ramped linearly with
time between temperature limits of 230 and 760 °C with a period of 30 min. The
temperature was phased with the strain such that the maximum temperature oc-
curred at the peak tensile strain. This is referred to as an "in-phase" thermal-
mechanical strain cycle.

CycUc strain hardening caused an increase in the stress range by about

15 percent during the first 10 to 15 percent of the cyclic life. Thereupon,
the hysteresis loop stabilized, and the stress range remained essentially
constant until approximately 90 percent of the Ufe was reached. Macro-
cracking became evident at this time, producing a drop in the stress range
and introducing a cusp in the compressive portion of the hysteresis loop.
Failure occurred on the 307th cycle. A hysteresis loop, ABCDEFGHIJKLA,
is reconstructed in Fig. 1, using the stress and strain values taken from
hysteresis loops on cycles 38, 85, 86, 88, 89, 173, 174, 175, and 176. These
were the cycles during which the inelastic strains were partitioned into
their creep and plastic components, using the step-stress method.

Separation of the Creep and Plastic Strains

Following the suggestion of Ref 4, the component of steady-state creep
for the entire period of the time interval considered is taken as the "creep"
strain for use in strainrange partitioning analysis. All the remaining in-

AXIAL STRESS, MPa

.006 .006 .008

AXIAL SreAINiMECH.)

A e , - A H = 0.00923
Aejn = 61 = 0.00616
TEMP: 230°-760° C
PERIOD • 1800 SEC
AISI TYPE 316 STAINLESS STEEL

FIG. 1—Stress-strain hysteresis loop for in-phase thermal-mechanical strain cycle illustrat-
ing stress levels and creep strains associated with experimental partitioning of strains by the
step-stress method.

elastic strain, whether instantaneous or whether occurring as first stage

(primary) creep, is taken to be "plasticity."
The partitioning procedure followed is to temporarily halt the tempera-
ture and strain programmers at a selected point, such as F, switch the
servocontroUer from strain to load control, and hold the stress and tem-
perature constant at the stabilized values associated with Point F while
creep strain is measured as a function of time. This condition is held until
a reasonably linear creep rate is estabUshed. This is taken as an approxi-
mation to the steady-state creep condition. At Point F ' , the controller is
switched back to strain control, and the strain and temperature programs
are resumed. Before stopping at the next step-stress level, the hysteresis
loop is restabilized by traversing one or more cycles. Stability is indicated
by observing the hysteresis loop to repeat the path of the previous loops.
In the present case, stability was established in the very next cycle follow-
ing a period of creeping. Table 2 summarizes the steady-state creep rate
information measured at each of the stresses and temperatures indicated.
Figure 2 illustrates the creep curves measured at various points around the

111
HU55

a
>•„ 2

t!
llf
5 u
i5 a3 1<s3 -
t^os'»

n^
>-i *0 • 00 O t>
oooo ' —vdodooooo
!lis
<^
so •t-i t«
c
•§ 1^12 :88 :S8S :S8
1s ^11
^ap a
•S e£:
§ — 5
•J- m
5> Tu? -H ( S <S <S| — —I — (S <S f l
-^
-*::
C
*-•
* i cj I I I I
^=4. M >.
U

c =»

1 IS o « vo <s 00 -^ •
»0 IS <S -" — I
^ ^ w% vo r^ I ills fS8 r^ 00

w 2«^
0»/^fnoo^o\»nQ(Sas'<tooo

H PI 4>

.s'R <pQOQWfcODa^-^^J<
o og
ft-

a> .
"y o
^z

TYPE 316 STAINLESS STCEL

500 1000 1500 2000 3000

TIME HELD AT STRESS, SEC

FIG. 2—Creep curves generated during application of step-stress method to partition

strains in thermal-mechanical strain cycle (230 to 760°C).

hysteresis loop. Comparison of these cyclic creep data with conventional

monotonic data would probably show significant differences in behavior.
However, equivalent monotonic data are not presently available. Experi-
mental partitioning procedures have as yet to be worked out for materials
that do not exhibit cyclic stability in the sense described here.
A plot of the steady-state creep rate versus time within the thermal-
mechanical cycle that each stress and temperature was encountered results
in the curve shown in Fig. 3. By integrating the area under this curve,
the amounts of tensile and compressive steady-state creep strain can be
determined. Areas above the horizontal axis represent tensile creep; those
below the horizontal axis represent compressive creep. For the example
shown in Fig. 3, the compressive steady-state creep strain is 0.00006.
Since the compressive inelastic strain is 0.00616, the compressive plastic
strain is 0.00610. Similarly, the tensile creep strain as determined from
Fig. 3 is 0.00145. Since the tensile inelastic strain is also equal to 0.00616,
the tensile plastic strain is 0.00471. Hence, the partitioned strainranges
[8] are

A£„ 0.00471, A£„ = 0.00006, A£„ = 0.00139 (1)

where

A£,„ Afpp + Af, + A£„ = 0.00616

9i-

-1800

u
<:
A R E A - a 00006
(STEADY STATE
COMPRESSIVE CREEP STRAIN)

^ ^
J
200 400 600 800 1000 1200 1400 1600 1800
TIME IN CYCLE, SEC

FIG. 3—Accumulation of tensile and compressive steady-state creep strain during in-phase
thermal-mechanical strain cycle (230 to 760°C) applied to AISI Type 316 stainless steel.

An assessment of the significance of experimental errors or other sources

of variation in the magnitude of these strainranges is given in a later sec-
tion. The strainrange fractions are

0.00471

0.00006
(2)
F,, = A£„/A£,„ = 0.010
0.00616
0.00139
F „ = A£„/A£,.„ = = 0.225
0.00616

Damage Rule
The strain fractions are used in the interaction damage rule which is
written as follows

N N N N N
•* ^pp ^ ^ cc ^ ^ cp •» ^pc •'• ^pr

where
Npp, Ncc, Ncp, and Np^ = cycUc lives determined from entering the life
relationships at a strainrange equal to the en-
tire inelastic strainrange of the cycle of inter-
est, and
Npr = predicted life.
This damage rule is based on assumptions described in detail in Ref 9.
Although all of the terms in the equation are Unear, we have continued to
use the term "interaction" in its designation in order to distinguish it
from the Unear damage rule used in Ref 8 and because of certain inter-
action assumptions entering the original derivation.

Life Relationships
The life relationships used in the life prediction of the example thermal-
mechanical strain-cycling problem are shown in Fig. 4. These curves are
based upon the use of the interaction damage rule and the interpretation
of creep strain as being only the steady-state (secondary) portion of the
time-dependent strain. All of the tests used to establish these life relation-
ships were conducted at a single isothermal temperature of 705 °C.

Prediction of Thermal-Mechanical Strain-Cycling Life

For an inelastic strainrange of 0.00616, the Npp, Ncc, Ncp, and Npc lives
from Fig. 4 are 1330, 405, 94, and 410, respectively. The predicted life,
Npr for this cycle, if uninterrupted, is computed from the interaction
damage rule as follows
F F F 1
/ PP^ , ^ cc I "P _
N N N N
^^PP ^^cc •'•^cp *^pr

0.765 0.010 0.225 1

1330 + 405 + 94 Npr
(4)
0.000575 + 0.000025 + 0.002395 = 0.002995 = ^
Npr

Npr = 334 cycles

.100 AISI TYPE 316

STAINLESS STEEL
705" C
IN-PHASE TMSC TEST
V 2300-760° C
PERIOD = 1800 SEC

105
CYCLES TO FAILURE

FIG. 4—Strainrange partitioning life relationships based on use of interaction damage rule
and steady-state creep strain.

As mentioned earlier, one of the advantages of strainrange partitioning

is its ability to recognize the importance of differences in the details of
strain cycles. In the sample problem, the considerable amount of tensile
creep strain is held responsible for reducing the life from a maximum of
1330 cycles down to the observed life of only 307 cycles. Had the same
in-phase thermal-mechanical cycle been run in such a way as to avoid
accumulation of creep strain by using a higher frequency or by shifting
the temperature range to a lower level, no loss in Hfe would be expected
since all of the strain would be of the Atpp type. With no creep in the
cycle, there would also be no expected difference in Ufe between an in-
phase and an out-of-phase test. As a case in point, Coffin [70] observed
no marked differences in thermal-mechanical fatigue lives of specimens of
AISI Type 347 stainless steel tested using in-phase and out-of-phase cycles.
Coffin's experiments were run at significantly higher frequencies and at
lower temperature levels than those used in the present investigation.
However, for the sample problem here, a frequency of 1/30 cycle/min
and a temperature range from 230 to 760 °C is severe enough to produce
significant creep. The method of strainrange partitioning therefore in-
dicates a substantial difference between lives expected for in-phase and
out-of-phase thermal-mechanical strain cycles.
Had the sample test been conducted with the temperature and strain

out-of-phase, one could expect exactly the same amount of creep and
plastic strain as measured previously but with the role of tension and
compression reversed. That is, the term Fcp/Ncp in Eq 4 would be replaced
with FpJNp^ where Fp<, is 0.225 and Np^ is 410 cycles. The predicted life
for out-of-phase cycling would be 870 cycles which is 2.6 times greater
than the predicted life of 334 cycles for in-phase cycling. The result of
this calculation is in line with our experience in conducting a number of
other thermal-mechanical strain-cycling tests on AISI Type 316 stainless
steel.''
Application of the time and cycle fraction approach as used in ASME
Code Case 1592 [16\ does not differentiate between in-phase and out-of-
phase thermal-mechanical straining since compressive creep and tensile
creep are taken to be equally damaging. This approach would predict the
same life for these two extremes in cycling which can lead to the following
undesirable situation: consider that the design curves were established
with just the proper amount of conservatism to handle an in-phase thermal-
mechanical cycle such as the one just discussed. By applying the same
criteria and design curves to an out-of-phase cycle, the designer would
predict an unnecessarily conservative life. An inefficient design results.
The foregoing discussion helps to illustrate the emphasis that strain-
range partitioning places upon some of the details within thermal-mechan-
ical strain cycles.

Assessment of Error
It should also be pointed up that these calculations do not include an
assessment of the added damage to the specimen as a result of the step-
stress partitioning procedures. A series of calculations have been made
to determine the damage done by having held a constant creep stress for
the time intervals listed in Table 2. Each cycle during which the creep
' A n out-of-phase test conducted at an inelastic strainrange of 0.00484 failed at 1449
cycles. The frequency and temperature range were the same as listed in Table 1. Since this
particular test was conducted prior to the development of the step-stress method, the parti-
tioning of the creep and plastic strains was not accomplished. However, since the strainrange
is only 20 percent less than that of the in-phase test, it is reasonable to assume that the
partitioning would have been approximately the same. Hence, the predicted life is given by

F F F 1
PP ' cc ^ pc *
A^ N N ~ N
*^ PP ^^ cc ^^ pc ^^ pr
0.765 0.010 0.225 1
2022 + 524
... + 540 N,r
1
0.000378 -t- 0.000019 + 0.000417 = 0.000814 =

N., = 1229 cycles

Stress was tensile produced a component of Ae^p strainrange which was

greater than would normally have been experienced during that cycle.
Likewise, during the cycles with compressive creep stress, there was usually
a compressive creep strain greater than the 0.00006 amount normally
experienced. In this latter case, the result was to increase the Lz^c strain-
range at the expense of the At^p strainrange. Since A£„ straining is less
damaging than Accp, there was less damage done during these particular
cycles. The net effect of the hold periods during experimental partitioning
was to induce damage in the amount of 0.0240. Thus, 2.4 percent less
damage was available for the thermal-mechanical strain cycling. Hence,
the Npr value should be reduced by 2.4 percent to a value of 326 cycles to
failure. This life can be compared directly with the observed Ufe of 307
cycles to failure for this example problem.
It is also informative to note that the predicted Ufe would have differed
by less than 3 percent had an error as large as 100 percent been made in
the determination of the compressive creep strain. Had an error of 10
percent been made in the determination of the tensile creep strain, the
predicted life would have been altered by less than 8 percent. Further-
more, a 10 percent variation in the inelastic strainrange would produce
only an 8 percent change in the predicted life. For the particular problem
at hand, a percentage error in determination of strainrange translates
into an error in predicted life that is slightly less than that percentage
figure.

Bounding the Life Values

Instead of quantitatively determining life by detailed partitioning, an
alternate approach is to estimate bounding values on life from a knowl-
edge of the total inelastic strainrange and an estimate of the types of
strainrange likely to be present. Here, because of the nature of the cycling,
the most damaging type of strainrange is of the Ltcp type. If the entire
cycle produced a A£<,p type strainrange, and the Ae^p life relationship is
indeed independent of temperature, the life would be 94 cycles to failure.
This value is a lower bound on life and can be determined without detailed
partitioning. The upper bound on life would have been the N^p value of
1330 cycles to failure. For the out-of-phase test mentioned in the earlier
footnote, the upper bound would be the Npp life of 2022 cycles, and the
lower bound would be the Np^ Ufe of 540 cycles for the inelastic strain-
range of 0.00484.

Some Metallurgical Aspects

As was shown in Ref 8, application of a A£c.p type strainrange is con-
ducive to intergranular creep cracking in AISI Type 316 stainless steel.

The photomicrographs of the failed specimen shown in Fig. 5 also reveal

intergranular cracking (particularly evident in Fig. 5a), thus substantiating
qualitatively the measured presence of a significant amount of Ae^p type
strainrange in the present test. Figure 5c shows the wall thickness of the
tubular specimen on the side of the tube from which the major crack
originated and propagated. This cracking was along grain boundaries.
Intergranular secondary cracks are also noted, although there is an indica-
tion that these may have initiated in a transgranular mode, possibly by
the action of the Aspp component of strainrange. Figure 5b is a photo-
micrograph of the wall thickness on the side of the tube opposite to Fig.
5c. The 45 deg shear angle across the thickness illustrates the final tearing-
off of the specimen into two pieces on the 307th cycle. This tearing action
opened up a number of internal intergranular cracks that had formed

INTERGRANUUR
CRACKS OPEWO
BY LAST CYCLE

(a)

aiMM

mRECTION
Of AXIAL
STRAIN
i:

INSire OlAM
OUTSIM OlAM
WALL THICKNESS WALL THICKNESS

MACROGRAPH OF HOUR-GLASS TCST SECTIW

FIG. 5—Photomicrographs of failed specimen of AISI Type 316 stainless steel subjected
to thermal-mechanical cycling (230 to 760°C) in phase.

during the thermal-mechanical strain-cycHng. It should be noted that the

failed specimen is free of cycle-dependent buckling or other geometric
instabilities (Fig. 5d).

Conclusions
We have applied the method of strainrange partitioning to a thermal-
mechanical strain-cycling problem on AISI Type 316 stainless steel. The
sample problem can be used as an initial guide in applying the method to
the general thermal fatigue problem. To date, the approach has been
limited to cycles in which the inelastic strains can be determined with
some degree of confidence. Extension of the method would be required in
order to deal with cycles for which the elastic strains dominate and the
inelastic strains are too small to calculate or measure accurately.
Although the sample problem involved a material whose life relation-
ships are insensitive to temperature, some suggestions were given for
dealing with materials for which this is not the case. Obviously, further
experimentation is needed to test out the applicability of these suggestions.
Additional research is also needed to guide in the selection of appropriate
techniques for analyzing materials that exhibit metallurgical instabilities
such as dynamic strain-aging or that experience strong interactions with
the environment.
Extension of the method to predict the thermal fatigue life of an actual
component would not be a difficult task, provided analyses were available
for the temperature and equivalent axial strain histories at critical loca-
tions. An axially loaded laboratory specimen could then be programmed
to follow these histories. From that point on, the determination of the
creep and plastic strains, the partitioned strainrange components, etc.,
could follow the procedures described in this paper.
Rather than measuring the creep and plastic strains within a cycle as
illustrated in this paper, they could be calculated using procedures and
constitutive equations such as those proposed by Spera and Cox [19] and
Mowbray and McConnelee [27]. Of course, upper and lower bounds on
life can be estimated as discussed earlier without having to resort to either
experimental or analytical partitioning.

References
[7] Taira, S. in Creep in Structures, N. J. Hoff, Ed., Springer-Verlag, Berlin, 1962, pp.
96-119.
[2] Manson, S. S. and Spera, D. A., Quarterly Transactions of the American Society for
Metals, Vol. 58, 1965, pp. 749-751.
[3] Manson, S. S., IntemationalJoumal of Fracture Mechanics, Vol. 2, 1966, pp. 327-363.
KJ Manson, S. S. and Halford, G. R. in Thermal and High Strain Fatigue, The Metals
and Metallurgy Trust, London, 1967, pp. 154-170.

[5] Spera, D. A., "A Linear Creep Damage Theory for Thermal Fatigue of Materials,"
Ph.D. thesis. University of Wisconsin, Madison, 1968.
[6] Coffin, L. F., Proceedings, 2nd International Conference on Fracture, Chapman and
Hall, London, 1969, pp. 643-654.
[7] Ellis, J. R. and Esztergar, E. P. in Symposium on Design for Elevated Temperature
Analysis, American Society of Mechanical Engineers, 1971, pp. 29-43.
[8\ Manson, S. S., Halford, G. R., and Hirschberg, M. H. in Symposium on Design for
Elevated Temperature Analysis, American Society of Mechanical Engineers, 1971, pp.
12-23.
[P] Manson, S. S. in Fatigue at Elevated Temperatures, ASTM STP 520, American Society
for Testing and Materials, 1973, pp. 744-782.
[10] Coffin, L. F., Jr., Transactions of the American Society of Mechanical Engineers,
Vol. 79, No. 7, Oct. 1957, pp. 1637-1649.
[11] Morrow, J. and Halford, G. R., Proceedings, Joint International Conference on Creep,
Institution of Mechanical Engineers, 1963, pp. 3-43 to 3-47.
[12] Swindeman, R. W., Proceedings, Joint International Conference on Creep, Institution of
Mechanical Engineers, London, 1963, pp. 3-71 to 3-76.
[13] Manson, S. S., Halford, G. R., and Spera, D. A. in Advances in Creep Design, A. 1.
Smith and A. M. Nicolson, Eds., Applied Science Publications Ltd., London, 1971, pp.
229-249.
[14] Halford, G. R., Metallurgical Transactions. Vol. 3, No. 8, 1972, pp. 2247-2256.
[15] Manson, S. S., Halford, G. R., and Nachtigall, A. J. in Advances in Design for Ele-
vated Temperature Environment, American Society of Mechanical Engineers, 1975, pp.
17-28.
[16] Code Case 1592, Boiler and Pressure Vessel Code, American Society of Mechanical
Engineers, April 1974.
[17] Halford, G. R., Hirschberg, H. M., and Manson, S. S. in Fatigue at Elevated Tempera-
lures, ASTM STP 520, American Society for Testing and Materials, 1973, pp. 658-669.
[75] Manson, S. S., Thermal Stress and Low-Cycle Fatigue, McGraw-Hill, New York, 1966.
[19] Spera, D. A. and Cox, E. C , this publication, pp. 69-85.
[20] Hirschberg, M. H. in Manual on Low Cycle Fatigue Testing, ASTM STP 465, American
Society for Testing and Materials, 1969, pp. 67-86.
[21] Mowbray, D. and McConnelee, J., this publication, pp. 10-29.

Summary

The objective put forward by the committee which organized the Sym-
posium on Thermal Fatigue was to exchange ideas on methods for deter-
mining and improving the thermal fatigue resistance of both materials and
components. A total of 16 technical papers were presented at the sym-
posium, of which one was a review of terminology and historical develop-
ments in the field of thermal fatigue, thirteen were devoted to research on
materials, and only two dealt directly with components. Because of the
large quantity and variety of information assembled by those who con-
tributed to this symposium, it is appropriate to summarize here the major
conclusions offered, to evalute the success of the symposium in achieving its
goals, and to offer some suggestions for future thermal fatigue research.

Summary of Major Conclusions of the Contributed Papers

In the opening address, Spera concluded that it would be convenient and
historically correct to view thermal fatigue as one of two categories of low-
cycle fatigue, the other being isothermal fatigue. Further subdivision of
thermal fatigue into thermal-stress fatigue and thermal-mechanical fatigue
was suggested, depending on whether constraint of free thermal expansion
was internal or external, respectively. After a brief review of historical
developments from 1838 to the present, it was hypothesized that we now
have quantitative understanding of the mechanisms of thermal fatigue and
that we can make Ufe predictions with factor-of-two accuracy, at least for
conventional metals.

A nalysis Papers
Mowbray and McConnelee compared the results of two-dimensional
finite-element analysis of temperature, stress, and strain (elastic and
inelastic) in a thermal-stress fatigue specimen with similar results using a
simpler uniaxial, elastic strain invariance (UESI) method and found that
stress-strain behavior trends were the same. Moreover, in most temperature
regions of the specimen, identical behavior was calculated by the two
methods. Therefore, it was concluded that the simpler and less costly UESI
method was adequate for further parametric studies of three damage
calculation methods.

255

Downloaded/printed
Copyright 1976 b y Aby
S I M International www.astm.org
University of Washington (University of Washington) pursuant to Li
256 THERMAL FATIGUE OF MATERIALS AND COMPONENTS

Gonyea, in one of the two symposium papers devoted to components,

proposed that thermal-stress concentration factors (TSCF) can be defined
for large, nonuniform shafts which are relatively independent of time and
heating rate. Charts for estimating these factors were offered, based on
finite-element thermal-stress analysis. Significant differences between ther-
mal and mechanical (bending) stress concentration factors were discussed.
Tepper, in the only other component-oriented paper, proposed a 14-step
design approach to achieve reUability in a component susceptible to thermal
fatigue failure. It was concluded that, for the purpose of comparing design
alternatives, a simple life-fraction rule is sufficient for estimating creep and
plastic strain damage. Considerable effort would be required to improve
upon this simple analysis. Improvements are usually obtained by compiling
a library of creep-fatigue test data. This is an effort which can only be
justified under certain conditions, for example, when reliability is marginal
or a large number of production units are to be made.
Hasselman, Badaliance, and Chen demonstrated that a fracture-
mechanics approach is feasible for predicting the thermal fatigue life of
ceramic materials. They concluded that estimating the proper initial flaw
depth by use of WeibuU theory introduces the largest uncertainty in
estimating fatigue life.
Spera and Cox offered a computerized method for predicting thermal
fatigue life, using a generalized damage fraction theory. Many of the
elements of this computerized analysis also appear in Mowbray and Mc-
Connelee's UESI method and Tepper's preliminary LCF analysis. It was
claimed that sufficient accuracy is retained in spite of simplifying assump-
tions to make this computer program a useful tool for calculating thermal
fatigue resistance. Application of this method to both thermal-stress fatigue
and thermal-mechanical fatigue problems was illustrated.

Papers on Thermal-Stress Fatigue

Howes concluded that the two dominant mechanisms in the formation of
a thermal-stress fatigue crack are local plastic strain and oxidation. After an
incubation period, cracking almost invariably occurs in areas of highest
plastic strain. Oxidation is often preferential at grain boundaries, forming
wedges which grow in size until a crack is initiated and begins to propagate.
Oxidation also increases the amount of local plastic strain by reducing an
alloy's strength through depletion of strengthening elements and phases.
Bizon and Spera used fluidized-bed tests to compare the relative resist-
ances to thermal-stress fatigue of a wide variety of nickel- and cobalt-base
alloys. They showed that directional solidification and proper coating are
very effective techniques for increasing the thermal fatigue life of these
types of alloys, often by two orders of magnitude.

Beck and Santhanam demonstrated that the thermal-stress fatigue re-

sistance of a cast cobah-base alloy can be significantly improved by
changing the casting variables so as to increase the spacing of the dendrite
arms in the microstructure. Crack initiation periods were lengthened, and
propagation rates were reduced by this technique. This paper and the
previous one by Bizon and Spera were the only papers presented at the
symposium which dealt directly with methods for improving the thermal-
fatigue resistance of materials.
Laub stated that thermal fatigue failures in furnace heat exchangers
fabricated from mild steel sheet can now be predicted with confidence. At
present.the prediction technique is primarily experimental, combining
strain gage data from prototype heat exchangers with thermal fatigue data
on tubular specimens. It has not yet been demonstrated that strain gage
measurements can be replaced with a finite-element strain analysis. The lat-
ter must be capable of modeling the large transverse plate deflections
present in furnace heat exchangers.

Papers on Thermal-Mechanical Fatigue

Thermal-mechanical fatigue is primarily a laboratory testing procedure in
which external forces are applied to test specimens to produce stresses which
simulate the internal thermal stresses in an actual component. According to
Hopkins, equipment is now available with which almost any strain-
temperature-time cycle of practical interest can be reproduced in the
laboratory. Tubular test specimens are useful for both crack initiation and
crack propagation studies. Computer control is almost a necessity for
achieving acceptable accuracy because of the need to continuously adjust
the strain equipment to compensate for small temperature errors.
Jaske found that, in order to correlate in-phase, out-of-phase, and
isothermal fatigue test data for a low-carbon steel, a knowledge of both
cyclic stress and cyclic strain was necessary. On the basis of stabilized stress
range versus cycles to failure, little difference was observed in the behavior
of specimens subjected to in-phase and out-of-phase cycles of temperature
and strain. However, the cyclic hardening characteristics of the alloy tested
were complex, indicating the difficulty in predicting the stabilized stress
range.
Gemma, Langer, and Leverant demonstrated that the effect of elastic
modulus on crack growth rates during strain-cycUng tests is opposite to that
during stress cycling. Tests on specimens of a directionally solidified nickel-
base alloy with various grain orientations (and therefore moduli) showed
that crack growth rates increased with increasing modulus in the loading
direction under thermal-mechanical strain-cycling conditions. Good
correlation of data for this alloy in all grain orientations was achieved by
plotting crack growth rate versus a pseudostress intensity range. This latter

parameter is equal to the product of the modulus in the load direction and
the strain intensity range.
Sheffler demonstrated the drastic reductions in Ufe which thermal-
mechanical cycling can produce in two iron-base alloys in comparison to
isothermal cycling at the maximum cycle temperature. In most cases, the
Hfe reductions are attributed to grain-boundary creep damage. A few
specimens of a high ductility alloy subjected to out-of-phase cycling showed
pronounced geometry changes which led to premature failure. This same
behavior has been observed previously in two very ductile tantalum alloys,
suggesting that high-ductility alloys may be more susceptible to geometric
instability.
Halford and Manson demonstrated the feasibiUty of extending the
techniques of strainrange partitioning to include the calculation of thermal-
mechanical fatigue life. It was proposed that further development of this
analysis will remove present restrictions such as requirements for
measurable inelastic strain, temperature insensitivity of life relations, and
metallurgical stabiUty.
According to Coffin, geometric instabilities such as those observed
during thermal-mechanical strain cycUng must be preceded by a gradient in
the cyclic stress-strain field. This gradient may be caused by geometry, tem-
perature, or material variations. It was also suggested that macroscopic
geometry changes occurring during thermal-mechanical fatigue may be
analogous to microscopic surface roughening and crazing observed during
thermal-stress and isothermd fatigue.

Achievement of Symposium Goals

The symposium was successful in fostering an exchange of information
on a variety of subjects, all related to thermal fatigue, and, in this respect,
its major goal was achieved. The combination of these papers with those
recently pubUshed in Fatigue at Elevated Temperatures, ASTM STP 520
provides an excellent set of references on the state of the art in both the
isothermal and thermal fatigue categories of low-cycle fatigue. It is hoped
that papers on methods for improving the thermal fatigue resistance of ac-
tual components will be presented at a symposium in the near future.

Critical Experiment
An important experiment which remains to be performed is the direct
comparison between the life of a thermal-stress fatigue specimen and that of
a thermal-mechanical fatigue specimen subjected to identical cycles of
strain and temperature. While it is convenient to assume that these two lives
are roughly equal, correlation between thermal-stress fatigue and thermal-
mechanical fatigue has not been critically evaluated in the literature.

However, all the necessary tools are now available, as evidenced by papers
presented at this symposium. These tools are (a) equipment for well-
controlled thermal-stress fatigue tests, (b) analyses for determining cycles of
temperature and stress at crack locations in a thermal-stress fatigue
specimen, and (c) thermal-mechanical fatigue machines for faithfully
reproducing these cycles in a material test specimen. This critical experiment
is important, not only for the technical information which it would
produce, but also for the unifying effect it would have on research and
development in the field of thermal fatigue.

D. A. Spera
Fatigue Branch, National Aeronautics and
Space Administration Lewis Research
Center, Cleveland, Ohio; symposium co-
chairman and also coeditor of this publi-
cation.

Index
A Thermal stress, 3, 10, 30, 38, 55,
. , . , , , . ^- . . 69,86,106,123
Analysis (calculation, prediction) ci j - j u j ir> o^ in/: m
r> \ • 1 10 Ir. iin Fluidized bed, 10, 86,106,123
Creep strain, 1, 38, 69, 239 Fracture 55 199
Fatigue life, 38, 55, 69, 239 i^racture, 55, 199
Finite element, 11, 32, 40, 71, 93,
155 H
Plastic strain, 1,38, 69, 239 Heat exchangers, 141
Stress, 10, 30, 69 Heat resistant alloys, 106
Thermal (temperature), 10, 59, 88
I
Intergranular fracture, 239
Ceramics, 55 Iron alloys, 86, 141, 170, 215, 229,
Coatings, 70, 106, 143 239
Cobah alloys, 12, 106,123
Computer program, 11, 32, 42, 62,
69, 166 L
Crack initiation, 69, 157 Life prediction, 38, 55, 69, 239
Crack propagation, 55, 86, 157, 199
Creep strain, 38, 69, 239
M
Materials
D Ceramic
Definitions, 3 Silicon nitride HS-130, 57
Directional solidification, 107, 199 Soda-lime-silica glass R-6, 56
Disk Cobalt alloys, 106
Test specimens, 10, 87 FSX414, 12
Turbine, 31, 38 Mar-M509, 123
Copper, zirconium, 228
P Iron alloys
A286,215
Fatigue AISI 1010, 170
Thermal mechanical, 3, 69, 141, H-13, 86
157, 170, 199, 214, 227,239 Low-carbon mild steel, 141

261

Downloaded/printed
Copyright 1976 b y Aby
S T M Internationa] www.astm.org
University of Washington (University of Washington) pursuant to Lic
262 THERMAL FATIGUE OF MATERIALS AND COMPONENTS

Type 304, 215,229 Silicon nitride, 55

Type 316, 239 Specimens, test
18Ni maraging, 86 Bar, 76, 217, 228
Nickel alloys, 86 Notched, 76
Astroloy, 39 Disk, 12, 87
B-1900, 76 Notched, 87
B-1900 + Hf, 199 Rod, 56
Mar-M2{X) + Hf (DS), 199 Tube, 144, 160, 177, 202, 243
Tantalum alloy, T-111, 76 Notched, 144, 203
Mechanical properties, 106 Wedge, 77, 109, 126
Mechanical tests, 157 Steel, 86, 141, 170, 215, 229, 239
Microstructure, 101, 103, 123, 131- Strainrange peirtitioning, 239
135, 152, 201,223-225 Stress analysis, 10,30, 69
Stress (or strain) concentration, 30,
50, 73, 76, 87
N
Substructures, 38
Necking, 227
Nickel-base alloys, 39, 76, 86, 106,
157, 199
Tantalum alloy, T-111, 76
Test methods
O
Fluidized bed, 10, 86, 106, 123
Oxidation, 106 Quench, 55
Thermal-mechanical, 4, 141, 157,
p 170,199,214,227,239
Thermal (temperature) analysis, 10,
Plastic strain analysis, 11, 38, 69, 239 59, 88
Plastic deformation, 86 Thermal stress, 3, 10, 30, 38, 55, 69,
Precipitation hardening, 170 86, 106
Turbines, 38
S
V
Shaft, turbine, 30
Silica glass, 55 Vacuum fatigue testing, 214

Design of Structural Steel Pipe Racks
100% (2)
Design of Structural Steel Pipe Racks
12 pages
Overview of LNG Storage Tank
100% (4)
Overview of LNG Storage Tank
23 pages
ASTM - Thermal Fatigue of Materials and Components
100% (7)
ASTM - Thermal Fatigue of Materials and Components
270 pages
Advances in Fatigue Science and Technology
100% (1)
Advances in Fatigue Science and Technology
982 pages
Earth Science (Second Quarter)
67% (3)
Earth Science (Second Quarter)
3 pages
PERMAS Online Manuals (EDU)
No ratings yet
PERMAS Online Manuals (EDU)
1 page
Combustion Related Fault Diagnosis of Diesel Engine Based On Harmonic Analysis of Ias
No ratings yet
Combustion Related Fault Diagnosis of Diesel Engine Based On Harmonic Analysis of Ias
8 pages
Levenberg Marquardt in Excel Excel VBA
0% (3)
Levenberg Marquardt in Excel Excel VBA
3 pages
Stratification and Rollover in LNG Tanks
100% (2)
Stratification and Rollover in LNG Tanks
14 pages
Al Safat - Dubai Green Building System 2nd Editionjan2023
100% (1)
Al Safat - Dubai Green Building System 2nd Editionjan2023
75 pages
Spitler Mcquiston Lindsey 93
No ratings yet
Spitler Mcquiston Lindsey 93
9 pages
Stp1231-Eb 20688 PDF
No ratings yet
Stp1231-Eb 20688 PDF
668 pages
STP 338-1963
No ratings yet
STP 338-1963
274 pages
Design and Calculation Ofsteam-Turbine D Isk W Heels: Ntroduction
100% (1)
Design and Calculation Ofsteam-Turbine D Isk W Heels: Ntroduction
16 pages
Thermal Fatigue
No ratings yet
Thermal Fatigue
39 pages
Rotordynamics 2: Problems in Turbo Machinery
No ratings yet
Rotordynamics 2: Problems in Turbo Machinery
6 pages
ASTM International - Site Search
No ratings yet
ASTM International - Site Search
6 pages
Artigo - Schijve
No ratings yet
Artigo - Schijve
24 pages
STP 556-1974
No ratings yet
STP 556-1974
192 pages
A.A.D. Brown Engineering Design Guides Mechanical Springs 42 PDF
No ratings yet
A.A.D. Brown Engineering Design Guides Mechanical Springs 42 PDF
48 pages
Thermal Fatigue of Cast Irons For Automotive Application
No ratings yet
Thermal Fatigue of Cast Irons For Automotive Application
7 pages
STP1337 Eb.1415051 1 PDF
No ratings yet
STP1337 Eb.1415051 1 PDF
203 pages
STP 648-1978
No ratings yet
STP 648-1978
300 pages
A New Approach For Sizing Finger-Type (Multiple-Pipe) Slug Catchers
No ratings yet
A New Approach For Sizing Finger-Type (Multiple-Pipe) Slug Catchers
9 pages
Fatigue of Nitinol
No ratings yet
Fatigue of Nitinol
27 pages
STP 595-1976
100% (1)
STP 595-1976
344 pages
Welding of P91
No ratings yet
Welding of P91
10 pages
Battery Storage For Off-Grid Systems
No ratings yet
Battery Storage For Off-Grid Systems
47 pages
Surface Vehicle Information: Rev. AUG2002
No ratings yet
Surface Vehicle Information: Rev. AUG2002
29 pages
Thermal Fatigue - An Analysis of The Experimental Method
No ratings yet
Thermal Fatigue - An Analysis of The Experimental Method
76 pages
Fatique 1 PDF
No ratings yet
Fatique 1 PDF
116 pages
L-2 Design of Steel Roof Truss As Per SP38888
No ratings yet
L-2 Design of Steel Roof Truss As Per SP38888
21 pages
An Investigation of The Failure Mechanisms in High Temperature Materials Subjected To Isothermal and Anisothermal Fatigue and Creep Conditions
No ratings yet
An Investigation of The Failure Mechanisms in High Temperature Materials Subjected To Isothermal and Anisothermal Fatigue and Creep Conditions
6 pages
STP 1546-2012
100% (2)
STP 1546-2012
390 pages
Ch2 FluidFlow 2
No ratings yet
Ch2 FluidFlow 2
56 pages
STP 569-1975
No ratings yet
STP 569-1975
336 pages
Silfit-Z-91 (HOFFMANN)
No ratings yet
Silfit-Z-91 (HOFFMANN)
2 pages
Homework Solutions For Chapter 4
No ratings yet
Homework Solutions For Chapter 4
9 pages
Applications of DOE in Engineering and Science 2019 Revised-LYE Sept 17 PDF
100% (1)
Applications of DOE in Engineering and Science 2019 Revised-LYE Sept 17 PDF
214 pages
Fourier PHYC20011 2015
No ratings yet
Fourier PHYC20011 2015
8 pages
2 Research in 20th Century PDF
No ratings yet
2 Research in 20th Century PDF
24 pages
Victrex - Material Properties Guide 3 - 7 - US
No ratings yet
Victrex - Material Properties Guide 3 - 7 - US
26 pages
Fatigue - A Complex Subject - Some Simple Approximations
No ratings yet
Fatigue - A Complex Subject - Some Simple Approximations
107 pages
Book Pract RDFFBD Chapter-6
No ratings yet
Book Pract RDFFBD Chapter-6
90 pages
An Introduction To Statistical Learning With Applications in R (ISLR Sixth Printing)
No ratings yet
An Introduction To Statistical Learning With Applications in R (ISLR Sixth Printing)
3 pages
Dura Seal Iom Manual
No ratings yet
Dura Seal Iom Manual
13 pages
Downloaded/printed by Universidad de Chile Pursuant To License Agreement. No Further Reproductions Authorized
No ratings yet
Downloaded/printed by Universidad de Chile Pursuant To License Agreement. No Further Reproductions Authorized
224 pages
Practise Paper - Chemistry - Class XI 2023-24
No ratings yet
Practise Paper - Chemistry - Class XI 2023-24
6 pages
Li Kolar Li Et Al Research Development of Preload Technology On Angular Contact Ball Bearing of High Speed Spindle A Review (2020) PUBV 340215
No ratings yet
Li Kolar Li Et Al Research Development of Preload Technology On Angular Contact Ball Bearing of High Speed Spindle A Review (2020) PUBV 340215
23 pages
Stress in Rotating Disks and Cylinders
No ratings yet
Stress in Rotating Disks and Cylinders
34 pages
Determination of Stress Concentration Factors
No ratings yet
Determination of Stress Concentration Factors
5 pages
Cambridge Material Data Book
0% (1)
Cambridge Material Data Book
40 pages
Modulus of Elasticity Applications
No ratings yet
Modulus of Elasticity Applications
8 pages
Module 5 Balancing of Rotating Masses NM Repaired
No ratings yet
Module 5 Balancing of Rotating Masses NM Repaired
25 pages
IPACK2007-33984: Heat Pipes For High Temperature Thermal Management
No ratings yet
IPACK2007-33984: Heat Pipes For High Temperature Thermal Management
8 pages
GT2007 27064 PDF
No ratings yet
GT2007 27064 PDF
10 pages
Design Built A Young Jackfruit Skin Peeler Machine
No ratings yet
Design Built A Young Jackfruit Skin Peeler Machine
7 pages
Decision Trees
No ratings yet
Decision Trees
53 pages
STP566 Foreword PDF
No ratings yet
STP566 Foreword PDF
11 pages
Establishing Ultrasound Testing As A CBM Pillar - Reliabilityweb - A Culture of Reliability
No ratings yet
Establishing Ultrasound Testing As A CBM Pillar - Reliabilityweb - A Culture of Reliability
3 pages
MP 19000
No ratings yet
MP 19000
15 pages
Boiler Blowdown Types and Boiler Blowdown Calculation
No ratings yet
Boiler Blowdown Types and Boiler Blowdown Calculation
3 pages
Titanium and Titanium Alloys: Josef Stráský
No ratings yet
Titanium and Titanium Alloys: Josef Stráský
25 pages
Types of Friction
No ratings yet
Types of Friction
3 pages
Allowable Nozzle Loads
No ratings yet
Allowable Nozzle Loads
6 pages
Fatigue and Fracture Mechanics, 34 TH Volume: Steven R. Daniewicz, James C. Newman, and Karl-Heinz Schwalbe, Editors
No ratings yet
Fatigue and Fracture Mechanics, 34 TH Volume: Steven R. Daniewicz, James C. Newman, and Karl-Heinz Schwalbe, Editors
599 pages
On The Fatigue Behaviour of Epoxy-Based Nanostructured Composite Materials
No ratings yet
On The Fatigue Behaviour of Epoxy-Based Nanostructured Composite Materials
15 pages
Sec15 031210 NOTES Example
No ratings yet
Sec15 031210 NOTES Example
6 pages
Compressor Theory
No ratings yet
Compressor Theory
4 pages
Astec Concrete Testing
No ratings yet
Astec Concrete Testing
4 pages
Michael R. Mitchell and Kenneth L. Jerina, Editors - Fatigue and Fracture of Medical Metallic Materials and Devices (ASTM Special Technical Publication, 1481) (2007)
No ratings yet
Michael R. Mitchell and Kenneth L. Jerina, Editors - Fatigue and Fracture of Medical Metallic Materials and Devices (ASTM Special Technical Publication, 1481) (2007)
125 pages
AIR Distribution Data Sheet
No ratings yet
AIR Distribution Data Sheet
3 pages
Thermal Mechanical Fatigue Behavior of N
No ratings yet
Thermal Mechanical Fatigue Behavior of N
444 pages
UACE Physics Paper 1 2004 Guide
No ratings yet
UACE Physics Paper 1 2004 Guide
17 pages
Rital User Guide 2019
No ratings yet
Rital User Guide 2019
198 pages
Fatigue Testing Handbook
100% (1)
Fatigue Testing Handbook
223 pages
STP 9-C-1951
No ratings yet
STP 9-C-1951
25 pages
Reliability and Diagnostic Accuracy of 5 Physical Examination Tests and Combination of Tests For Subacromial Impingement
No ratings yet
Reliability and Diagnostic Accuracy of 5 Physical Examination Tests and Combination of Tests For Subacromial Impingement
6 pages
Abaqus Phase Emilio
No ratings yet
Abaqus Phase Emilio
12 pages
Ductility Test
No ratings yet
Ductility Test
4 pages
Quantum Gravity As Gravitized Quantum Theory: Tristan H Ubsch and Djordje Minic
No ratings yet
Quantum Gravity As Gravitized Quantum Theory: Tristan H Ubsch and Djordje Minic
60 pages
Robotic Arm
No ratings yet
Robotic Arm
28 pages
Fatigue in Composites
0% (2)
Fatigue in Composites
10 pages
Ferrari Purosangue - April 2024
No ratings yet
Ferrari Purosangue - April 2024
14 pages
GRADE10 - UnitPlan - Term 1 - Chemistry
No ratings yet
GRADE10 - UnitPlan - Term 1 - Chemistry
4 pages
Interpretation OF: NMR Spectrum
100% (1)
Interpretation OF: NMR Spectrum
5 pages
Types of Notches PPT 2
No ratings yet
Types of Notches PPT 2
17 pages
9.ray Optics
No ratings yet
9.ray Optics
108 pages
Module 5
100% (1)
Module 5
71 pages
How To Use Spss A Stepbystep Guide To Analysis and Interpretation 4th Edition Brian Cronk PDF Download
No ratings yet
How To Use Spss A Stepbystep Guide To Analysis and Interpretation 4th Edition Brian Cronk PDF Download
23 pages