Probabilistic Assessment of Fracture Mechanics of Low Pressure Turbine

Disk Keyway
Lieh Chen ()
1
, Lih-jier Young()
2
and Shih-teng Lin ()
2
1
Department of Industrial Engineering and Management, Ta Hwa Institute of Technology,
No. 1, Dahua Rd., Qionglin Shiang, Hsinchu County, 307, Taiwan, R.O.C.
Tel: 03-592-7700-2755, Fax: 03-592-6848
Email: ietch@thit.edu.tw
2
Department of Applied Mathematics, Chung-Hua University,
No. 707, Sec. 2, Wufu Rd., Hsinchu, 30012, Taiwan, R.O.C.
Tel: 03-518-6392, Fax: 03-518-6435
Email: young@chu.edu.tw, steng.lin@msa.hinet.net
Abstract
This study presents the development of a fracture mechanics-based probability analysis which can be used to
estimate the remaining life of the low pressure turbine against unstable fracture in the rotor disk keyway.
Analysis focuses on the stability of crack growth of a semi-elliptical flaw in the rotor disk keyway. The crack
stability is judged on the basis of the linear elastic fracture mechanics. Then the Monte Carlo technique is
introduced to deal with several random variables. Based on the evaluation results, the probability of failure of
observed crack indications can be determined. The inspection interval could be justified.
Keywords: Monte Carlo simulation, probabilistic fracture mechanics, stress corrosion cracking, fracture
appearance transition temperature
1. Introduction
The electric power and the electric component
demand have been increasing sharply. However, the
new materials which are fully capable of
withstanding stress corrosion cracking for large low
pressure turbines and rotors have not yet been
developed. As an observation, many low pressure
turbines are subjected to stress corrosion cracking in
wet bore, keyway, and dovetail area experienced
wetness in operation. Catastrophic failure of the disk
could occur if the cracks grow to critical size. It is,
therefore, one of the major concerns for the power
industry to limit the probability that disk will burst
and generate turbine missiles.
Many variables are involved in estimating the
remaining life of low pressure turbine disks such as
fatigue crack growth rate, fracture toughness, stress
intensity factor as a function of crack size, material
stress corrosion cracking rate as function of
environment, and loading conditions. To examine the
integrity of the cracked low pressure turbine disk,
both deterministic and probabilistic fracture
mechanics are used. Deterministic fracture mechanics
(DFM) method is used to ascertain allowable flaw
size, while probabilistic fracture mechanics (PFM) is
an alternative for determining inspection interval. The
Monte Carlo technique is one of numerical
probabilistic analysis approaches that are amendable
to statistical problems governed by a certain amount
of random variables.
There are many literatures focusing on choosing
random variables in Monte Carlo simulation.
However, only a few studies adopted Monte Carlo
method in analyzing the structural integrity of
cracked structures. Viswanathan (1984) estimated the
remaining life of the rotor with Monte Carlo
simulation [1]. Sire and Kokorakis (1991) assessed
the fatigue factor of container ship [2]. Zhu and Lin
(1992) studied the aviation structure of fatigue crack
growth under random loading [3]. Liao and Yang
(1992) performed a probabilistic evaluation on
fatigue crack growth of aluminum-alloy [4]. Rosario
and Roberts [5] conducted a remaining life evaluation
for a low pressure turbine disk in 1997.
Low pressure turbine disks are exposed to a
highly wetted environment, whose damaging
mechanisms include high cycle fatigue, low cycle,
and stress corrosion cracking. These conditions can
cause cracking of blade-fit area and keyway at bore
of disks. Therefore, our study tries to integrate
deterministic fracture mechanics with probabilistic
analysis in a consistent and convenient manner to
facilitate remaining life evaluation. Fig. 1 is a
schematic drawing of a typical low pressure turbine
disk with a keyway and a rotor.
2. Theoretical Investigations
The theory proposed by Griffith [6] in 1920
explained that it was the total energy release causing
the crack growth. Then the stress intensity factor, KI,
developed by Irwin [7], is a one parameter
representation of the stresses in the area of a crack
tip. It is a purely numerical quantity providing
complete knowledge of the stress field at the crack
tip, which is defined as follows:
( ) ..........
2
+
i j i j
K
f
r

(1)
Where r is a distance from crack tip, and is the
angle with respect to crack plane.
Equation (1) indicates that the local stress near a flaw
depend on the product of the nominal stress and the
square root of the flaw size a. This product is called
stress intensity factor to emphasize this
fundamental relationship.
The crack growth of the disk due to fatigue or
stress corrosion is predicted in accordance with the
theory of the conventional fracture mechanics. The
dependence of the fatigue crack growth rate [8, 9, 10]
on stress intensity factor can be conveniently
represented as follows:

n3
1
2
1-
( )
1
n
th
m
n
max
c
K
da K
C K
dN
K
K
_
_



,


_

, ,
(2)
where / da dN is the fatigue crack growth rate,
th
K
is the threshold value of stress Intensity factor,
c
K is
critical value, and n and C are constants depending on
the material and environmental conditions. This
typically results in a graph of the form in Fig. 2. It is
noted that the graph exhibits a threshold stress
intensity factor below which cracks will not
propagate.
2.1 Finite element stress analysis
This section is not an integral part of fracture
analysis method, and will not be treated in detail.
However, the determination of stress profile through
the disk constituting one of the basic inputs to the
fracture mechanics analysis. The general purpose
finite element computer program ANSYS [11], was
utilized to perform finite element analysis.
2.1.1 Loading and Boundary Conditions
Two-dimensional axisymmetric finite element
model of the subject disks was constructed to
determine near bore disk stresses due to centrifugal
effects using ANSYS computer program. It is noted
that an equivalent traction is applied on the disk OD
to simulate the blade centrifugal loading, which is

centrif
b
contact
N F
A

(3)
where N is the number of blades,
centrif
F
is the
centrifugal force per blade, and
contact
A is rim contact
area.
The axial displacements at the center of the disk
are all fixed to simulate half-symmetry as the
imposed boundary condition. The resulting disk/shaft
finite element model is shown in Fig. 3.
2.1.2 Stress Concentrating Effect of Keyway
In order to perform stress intensity factor
calculation for cracks emanating from the high stress
gradient area in the disk keyway/bore, a stress
concentration factor (
t
K ) of 2.2 [12] for GE axial
keyway design was utilized. The stress concentration
factor profile can be obtained by varying radial
distance (
x
) from the crown of the keyway which is
expressed as:
( ) ( )
2 4
1 1 0 25 0 75
t t
R R
K x K . .
R x R x
1
_ _
+ + 1

+ +
, , 1
]
(4)
Where R is the keyway depth. Then the stress
profile acting normally to the plane containing the
crack can be immediately determined.
2.2 Stress Intensity Factor
Stress intensity factor expressions for surface flaw
in a cylinder shown in Fig. 4 have been determined
using finite element models [13], the stress intensity
factor is expressed as:
, , , ,
I hoop i j
b b b t b
K F G
Q a t R t

_ _

, ,
(5)
where
( )
1 65
1 1 464
b
Q
a
+
.
.

hoop is the hoop stress, a is the surface length, b is

radial depth, R is outside diameter, t is the thickness,

is the angle,
i
F is the boundary-correction factor,
and j
G
is the influence coefficients corresponding to
the jth stress distribution. The coefficients of the
polynomial expression are used in equation (5) to
calculate the stress intensity factor corresponding to
the crack depth, b, and to the stress profile
represented by both finite element solution for

hoop
and stress concentration factor given in equation (4).
The same procedure can then repeated at various
crack aspect ratios. Fig. 5 shows stress intensity
factor profiles obtained at various crack aspect ratios.
2.3 Fracture Toughness
The material property pertinent to the present
fracture mechanics analysis method is the plane strain
fracture toughness which is a function of temperature.
The method predicts that crack instability occurs
when the stress intensity factor
I
K equals the
material fracture toughness
IC
K .
The great majority of low pressure turbine
materials are built from ASTM A471 (3.5% Ni-Cr-
Mo-V) forgings for which a large number of
toughness data have been generated. Since disk-
specific fracture mechanics (
IC
K ) data is not
available, generic disk toughness for GE low turbines
were utilized to estimate
IC
K [14]. Fig. 6 is a
schematic drawing of disk toughness versus
temperature for typical low pressure turbine
materials. They can be fitted to a hyperbolic tangent
cure such that,

0
tanh
IC
T T
K A B
C
_
+

,
(6)
Where
0
T is fracture appearance transition
temperature (FATT), T is disk temperature, A, B and
C are constants.
2.4 Crack Growth Rate
Crack growth due to stress corrosion cracking
(SCC) is the dominant mechanism in this analysis.
The most widely accepted SCC crack growth
equation used in the analysis of cracking in LP
turbines is given by Clark et al. of Westinghouse [15]
:

1
7302
ln - 0.0278
y
da
C
dt T

_ _
+

, ,
(7)
where
1
C is a material constant with a mean value of
-4.968, and a standard deviation of 0.587, T is the
operating temperature of the disk in
R

, y

is the
yield strength in ksi, and da dt / is the growth rate in
inch/hour.
2.5 Critical crack size determination
Initiation crack depth can be determined during
the service time at the intersection between the stress
intensity factor
I
K and fracture toughness
IC
K .
From the determination of
cr
a , it should be possible
to conclude that the low pressure turbine disk bore
integrity would be maintained in the case of severe
event such as the overspeed resulting from an
abnormal operation. However, the critical flaw crack
size is essentially dependent on stresses, flaw
geometry and disk material fracture toughness. These
parameters are dependent one another, and have
strong influence on critical crack depth and therefore
remaining life.
2.6 Estimating remaining life
Remaining life of LP turbine disks for keyway
cracking is calculated by the initial crack size (
i
a ),
the critical size (
cr
a ) and the crack growth rate (
da dt / ), expressed by the following relationship,

/
cr i
rem
a a
t
da dt

(8)
3. Monte Carlo Simulation
Monte Carlo simulation is used to analyze the
uncertainty growth from each input variable to stress
corrosion cracking life. The proposed method
combines deterministic fracture mechanics with
probabilistic fracture mechanics to develop
relationships in matrix form between the various key
parameters. The essence of this approach is to assign
mean values and statistical distributions to all of the
key variables affecting the problem. Some key
parameters used in deterministic fracture mechanics
analyses of LP turbine rotor disks are known to vary
significantly, and can be assumed to behave in a
random manner. These random variables adopted in
this paper are hoop stress, overspeed possibility,
fracture appearance transition temperature, operating
temperature, fracture toughness with and without
prewarming, crack model, initial crack size, stress
growth coefficient, and yield strength of the material.
The stress intensity factor and crack growth are
calculated deterministically for a few chosen standard
deviation of the above mentioned random variables.
These results were put into matrix form for use in the
probabilistic failure evaluation of low pressure
turbine. Performing each Monte Carlo iteration, a
random number is generated for each random
variable. The corresponding standard deviation is
calculated and is used to determine the stress
intensity factor at any time between initial crack
growth and final failure. Therefore, the time for the
crack to propagate from an initial flaw size to the
critical flaw size is obtained.

3.1 Linear Congruential Generator
In this paper, the Random Number Generator is
made by Linear Congruential Generator, the equation
is

[ ]
1
MOD
i i
Y AY C M
+
+ (9)
where
0
Y is the seed, A is the constant multiplier, C
is the increment, and M is the modulus.
3.2 Flow diagram for life assessment
By applying the input data to the analysis scheme
as shown in Fig. 6, the critical crack size, the
probabilities of crack initiation, crack growth and
failure could be assessed.
4. Results and conclusions
A stress and probabilistic fracture mechanics
analysis for LP turbine disk keyway was performed.
Through this investigation, the remaining life of the
disk has been assessed, and the crack growth of the
disk has also been predicted. The remaining life for
the disk keyway with and without prewarming and
each with an initial crack depth of 0.12 inch are
shown in Fig. 8 and 9, respectively. Fig. 10 and 11
are the plots for the disk keyway with and without
prewarming and each with an initial crack of 0.25
inch, respectively. Fig. 12 and 13 show the
remaining life for the disk keyway with and without
prewarming and each with an initial crack of 0.34
inch, respectively. By carrying out this investigation,
three major findings are:
1. LP turbine with prewarming can significantly
increase the remaining life of LP turbine disk.
2. With limiting probability of catastrophic disk
failure of
4
10

, continued operation for another

9 and 8 year-period can be justified for the case
of 0.12inch initial crack depth with prewarming
and without prewarming respectively.
3. With limiting probability of catastrophic disk
failure of
4
10

, continued operation for another

8.5 and 7.5 year-period can be justified for the
case of 0.25 inch initial crack depth
withprewarming and without prewarming
respectively.
4. With limiting probability of catastrophic disk
failure of
4
10

, continued operation for another

8.5 and 7.0 year-period can be justified for the
case of 0.34 inch initial crack depth with
prewarming and without prewarming
respectively.
References
[1] R. Viswanathan, Lift Assessment and
Improvement of Turbogenerator Rotors for Fossil
Power Plants, EPRI CS4160, Pergamon Press,
New York, 1984.
[2] R. A. Sire, J. E. Kokarakis, C. H. Wells, R. K.
Taylor, Probabilistic Structure Life Prediction
System for Container Ship Repair and
Inspection, International Journal of Pressure
Vessels and Piping, International Colloquium on
Aging of Materials and Lifetime Assessment, Vol.
50, No. 1-3, Apr. 1991, pp.297-315.
[3] W. Q. Zhu, Y. K. Lin, Y. Lei, On Fatigue Crack
Growth Under Random Loading, Engineering
Fracture Mechanics, Vol. 43, No. 1, Sep, 1992,
pp.1-12.
[4] M. Liao, Q. X. Yang, Probabilistic Model for
Fatigue Crack Growth, Engineering Fracture
Mechanics, Vol. 43, No. 4, Nov, 1992, pp.651-
655.
[5] Darryl A. Rosario, Blaine W. Roberts
Probabilistic Assessment of Failure by Stress
Corrosion Cracking in Shrunk-On Disks of Low
Pressure Turbines, EPRI Steam Turbine Stress
Corrosion Cracking Conference, March 19-20,
1997.
[6] A. A. Griffith. Philos. Trans. R. Soc. London, Vol.
A221, 1920, p.163.
[7] G.. R. Irwin, Analysis of Stresses and Strains
Near The End of a Crack Transversing a Plate,
Trans. of ASME, Journal of Applied Mechanics,
Vol. E24, 1957, pp.361-364.
[8] D. W. Heoppnre and W. E. Krupp, Predication
of Component Life by Application of Fatigue
Crack Growth Knowledge, Engineering Fracture
Mechanics, Vol.6, 1974, p.47.
[9] J. W. Barsom, Fatigue Crack Propagation in
steels of Various Yield Strengths, Transaction
ASME, Journal of Engineering Industrial, Series
B, No. 4, November 1971, p.1190.
[10] R. G. Forman and V. E. Kearney and R. M.
Engle, Numerical Analysis of Crack Propagation
in Cyclic Loaded Structures, Transaction ASME,
Journal of Basic Engineering, Vol. 89, No. 3,
1976, p. 459.
[11] , ANSYS 7.0 , 2004,2.
[12] H. G. Pennick and C. H. Wells, Stress and
Fracture Analysis of Shrink-On Steam Turbine
Disks, EPRI Report NP-3340, Rrsearch Project
1398-5, January 1984.
[13] I. S. Raju and J. C. Newman, Jr., Stress-Intensity
Factors For Internal and External Surface Cracks
in Cylindrical Vessels, Trans. ASME, Ser. J, J.
Pressure Vessel Technology, Vol.104, 1982,
pp,293-298.
[14] Stienstra, David I. Auen, Stochastic
Micromechanical Modeling of Cleavage Fracture
In the Ductile-Brittle Transition Region, UMI, A
Bell & Howell information company, 1990.
[15] W. G. Clark, B. B. Seth, and D. M. Shaffer,
Procedures for Estimating the Probability of
Steam Turbine Disc Rupture from Stress
Corrosion Cracking, presented at Joint.
ASME/IEEE Power Generation Conference,
October 1981.
Fig. 1 Schematics of Axial Rotor-Disk Keyway
Fig. 2 Schematic representation of fatigue crack
Fig. 3 Two-dimensional FEM model for Rotor-Disk

Fig. 4 Crack model
x
y
z
Rotor
Disk
Keyway
1
R
2
R
I
II
III
th
K
K
K
c
Log Scale
Log Scale
t
1
R
b
2a

.408 .384 .360 .336 .312 .288 .264 .240 .216 .192 .168 .144 .120 .096 .072 .048 .024
ai
140
120
100
80
60
40
20
1.0
0.4
0.2
K
Fig. 5 Stress intensity factor profiles obtained at
various crack aspect ratios (900 rpm).
Fig. 7 Flow diagram for a scheme of remaining life
evaluation
Fig. 6 Hyperbolic tangent curve showing the
graphical interpretation of the parameters.

Dot/Lines showMeans
0.00 5.00 10.00 15.00 20.00
YEAR
0.00
10.00
20.00
30.00
40.00
50.00
N
O
_
T
I
M

Fig. 8 Probabilistic frequency versus remaining

life without prewarming (0.12 in)
Dot/Lines showMeans
0.00 5.00 10.00 15.00 20.00
YEAR
0.00
10.00
20.00
30.00
40.00
S
O
_
T
I
M

Fig. 9 Probabilistic frequency versus remaining

life with prewarming (0.12 in)
Input data
Select var. min/max value
Read random var.
Calculate the physical property
Simulate the crack model
Calculate the crack size
Stress Corrosion Cracking
Estimating remaining life
Output data
Times 1~10000
Temperatur
e
K
IC
A
B
B
C
Dot/LinesshowMeans
0.00 5.00 10.00 15.00 20.00
YEAR
0.00
20.00
40.00
60.00
N
O
_
T
I
M

Fig. 10 Probabilistic frequency versus remaining

life without prewarming (0.25 in)
Dot/Lines showMeans
0.00 5.00 10.00 15.00 20.00
YEAR
0.00
10.00
20.00
30.00
40.00
S
O
_
T
I
M

Fig. 11 Probabilistic frequency versus remaining life

with prewarming (0.25 in)
Dot/Lines showMeans
0.00 5.00 10.00 15.00 20.00
YEAR
0.00
20.00
40.00
60.00
N
O
_
T
I
M

Fig. 12 Probabilistic frequency versus remaining

life without prewarming (0.34 in)
Dot/Lines showMeans
0.00 5.00 10.00 15.00 20.00
YEAR
0.00
10.00
20.00
30.00
40.00
S
O
_
T
I
M

Fig. 13 Probabilistic frequency versus remaining

life with prewarming (0.34 in)