International Journal of Fracture 102: 43–69, 2000.

© 2000 Kluwer Academic Publishers. Printed in the Netherlands.

Constraint effects on the ductile-to-brittle transition temperature

of ferritic steels: a Weibull stress model

X. GAO and R.H. DODDS, JR.

Department of Civil Engineering, University of Illinois, Urbana, IL-61801, U.S.A.
e-mail: r-dodds@uiuc.edu

Received 20 November 1998; accepted in revised form 28 April 1999.

Abstract. This study examines crack front length and constraint loss effects on cleavage fracture toughness
in ferritic steels at temperatures in the ductile-to-brittle transition region. A local approach for fracture at the
micro-scale of the material based on the Weibull stress is coupled with very detailed three-dimensional models
of deep-notch bend specimens. A new non-dimensional function g(M) derived from the Weibull stress density
describes the overall constraint level in a specimen. This function remains identical for all geometrically similar
specimens regardless of their absolute sizes, and thus provides a computationally simple approach to construct
(three-dimensional) fracture driving force curves σw vs. J , for each absolute size of interest. Proposed modifi-
cations of the conventional, two-parameter Weibull stress expression for cumulative failure probability introduce
a new threshold parameter σw−min . This parameter has a simple calibration procedure requiring no additional
experimental data. The use of a toughness scaling model including σw−min > 0 increases the deformation level at
which the CVN size specimen loses constraint compared to a 1T SE(B) specimen, which improves the agreement
of computational predictions and experimental estimations. Finally the effects of specimen size and constraint
loss on the cleavage fracture reference temperature T0 as determined using the new standard ASTM E1921 are
investigated using Monte Carlo simulation together with the new toughness scaling model.

1. Introduction

Macroscopic values of fracture toughness (Jc , KJ c ), measured experimentally over the lower-
end of the ductile-to-brittle transition (DBT) range of ferritic steels consistently exhibit a large
amount of scatter (for examples, see Wallin, 1984a; Sorem et al., 1991). In this temperature
range, transgranular cleavage triggers the brittle fracture event often in the presence of sig-
nificant plastic deformation along the crack front. Models to describe the observed scatter
generally adopt a weakest link approach consistent with common observations of a single
carbide that initiates cleavage along the crack front (see for example Wallin, 1984a; 1984b;
Lin et al., 1986). The volume of highly stressed material along the crack front thus plays a
crucial role in driving the fracture process. Elastic-plastic stress fields along the crack front
depend strongly on the specimen/component geometry, size, loading mode (e.g. tension vs.
bending) and material flow properties. When plastic regions ahead of the crack front interact
with nearby traction free boundaries, the single parameter characterization of the crack front
stresses, in terms of the J-integral, breaks down. Under such conditions, a complex (non-
linear) relationship develops between the three-dimensional crack front stresses, the applied
J-value which varies along the crack front, and the corresponding stressed volume of material
at the crack front that drives the weakest link mechanism (Moinereau, 1996; Wiesner and
Goldthorpe, 1996; Ruggieri et al., 1998; Gao et al., 1998).
44 X. Gao and R.H. Dodds

To reduce these complexities, testing standards for the measurement of cleavage fracture
toughness in the DBT range (e.g., ASTM E1921; 1998) require selection of a specimen type,
size and loading mode to insure that essentially plane strain, small-scale yielding (SSY) con-
ditions exist along the crack front at fracture. Specimens routinely used include single-edge
notched bends SE(B)s, and compact tensions C(T)s, with crack depth (a) to specimen width
(W ) ratios, a/W > 0.5. These deep-notch geometries eventually lose SSY conditions at J -
values large compared to the specimen size and material flow properties (Mudry et al., 1989).
Consequently, testing standards impose deformation limits of the form M = bσ0 /Jc > Mlimit,
where b denotes the length of the remaining ligament, σ0 represents the yield stress and Mlimit
defines the non-dimensional deformation limit. The specified value for Mlimit thus plays a key
role in defining the minimum allowable specimen size for testing a given material. ASTM
E1921 for example, specifies Mlimit = 30. This value enables testing of Charpy size (0.4T)
SE(B) specimens in the lower transition region for common structural and pressure vessel
steels. The Mlimit = 30 value has been selected largely on the basis of experimental data. Com-
putational studies by Nevalainen and Dodds (1995) which used detailed three-dimensional
finite element analyses of SE(B) and C(T) specimens found an Mlimit ≈ 60–80 needed to
maintain SSY conditions on the centerplane for moderately hardening materials. The issue of
a suitable deformation limit becomes especially significant for the testing of small size spec-
imens which more likely have low M values at fracture. Current applications that must rely
upon testing small specimens include material extracted from surveillance capsules of nuclear
pressure vessels, samples of experimental alloys under development and trepan samples taken
from structures currently in service.
The Weibull stress model originally proposed by Beremin (1983) provides a framework to
quantify this complex interaction among specimen size, deformation level and material flow
properties in a fully three-dimensional setting. They introduced the scalar Weibull stress (σw )
as a probabilistic fracture parameter, computed by integrating a weighted value of the principal
tensile stress over the fracture process zone (i.e., the crack front plastic zone). The Beremin
model adopts a two-parameter description for the cumulative failure probability Pf (σw ) =
1 − exp[−(σw /σu )m ], where m denotes the Weibull modulus and the scale parameter σu sets
the value of σw (the material’s microscale toughness) at 63.2 percent failure probability. The
model assumes the parameters (m, σu ) describe inherent properties of the material that imply
the formation of a certain distribution of metallurgical scale cracks once plastic deformation
occurs as the precursor to cleavage. The dependence of the model on just two parameters to
capture complex metallurgical processes raises concerns about the dependence of (m, σu ), for
example, on temperature, loading rate, irradiation damage, pre-straining, etc. (Mudry, 1987).
When implemented in a three-dimensional finite element code, the Beremin model predicts
the evolution of σw with applied loads (and J ) to define conditions leading to (local) mate-
rial failure. Recent efforts in this area have focused on developing transferability models for
cleavage fracture toughness. Bakker and Koers (1991), Minami et al. (1992), Ruggieri et al.
(1995), Koppenhoefer et al. (1997) assess the effects of specimen thickness, crack length and
loading rate on elastic-plastic fracture toughness. Gao et al. (1998b) propose a new procedure
to calibrate the Weibull parameters (m, σu ) based on the toughness transferability model. The
new procedure eliminates the inability to calibrate unique values of the parameters using only
fracture toughness data measured under SSY conditions. They also modify the failure proba-
bility to include a minimum value (σw−min ) for cleavage fracture which reflects an approximate
model for the conditional failure probability. Still other studies by Ruggieri and Dodds (1996),
Xia and Shih (1996), and Xia and Cheng (1997) generalize the Weibull stress for stationary
 Constraint effects 45

and growing cracks in a two-dimensional setting to include the coupled effects of constraint
loss and ductile tearing on Jc -values.
Here we apply the Weibull stress model, as modified by Gao et al. (1998b), to quantify the
synergistic effects of specimen size, constraint loss and material flow properties on cleavage
toughness for deep-notch SE(B) specimens. Very detailed, three-dimensional finite element
analyses support the development of a non-dimensional function g(M). Derived from the
Weibull stress density along the crack front, g(M) describes the evolution of constraint in
a specimen under increased loading. This function depends on the material flow properties
and the Weibull modulus (m), but remains identical for all geometrically similar specimens
regardless of their absolute size. Given g(M), the corresponding σw vs. J curve for a specimen
follows without further finite element analyses. The g(M) functions are presented for a range
of flow properties representative of structural and pressure vessel steels.
To synthesize results derived from applications of the (modified) Weibull stress model, we
adopt the framework known generally as the ‘master curve’. This concept, largely developed
by Wallin (1991) and incorporated in ASTM E1921, characterizes fracture toughness data
in terms of KJ c -values over the DBT region. The process relies on the applicability of a
normalized curve of median fracture toughness vs. temperature (the ‘master curve’ for 1T
size specimens) demonstrated experimentally to hold for a wide spectrum of ferritic pressure
vessel and structural steels having yield stress in the range 275 to 825 MPa. To generate
the actual (1T) median toughness vs. temperature curve for a specific material, conventional
fracture tests are conducted at one temperature in the DBT using replicate deep-notch SE(B) or
C(T) specimens. The toughness values, coupled with the master curve, provide a statistically
bounded estimate of the reference
√ temperature T0 at which the median fracture toughness
(KJ c(med) ) equals 100 MPa m for 1T specimens √ under SSY conditions (KJ c -values are
determined from measured Jc -values using KJ c = EJc ). The median fracture toughness
at any temperature in the DBT region can be obtained by ‘un-normalizing’ the master curve
using T0 . ASTM E1921 describes the testing and data analysis procedures to determine T0 .
The master curve methodology theoretically requires that plane strain, SSY conditions
prevail at fracture in each of the replicate test specimens. This requirement enables simple
computation of the statistical estimate for KJ c(med) and the adjustment of KJ c -values obtained
with other than 1T specimens to their 1T equivalent values using a simple weakest link model.
Combinations of small specimen sizes and (relatively) high testing temperature may easily
lead to varying levels of constraint loss in each of the measured KJ c -values. The KJ c(med)
determined from these values leads to a corresponding T0 -value that differs from the inherent
material T0 -value (in SSY). Ruggieri et al. (1998) conducted a parameter study to estimate
the effects of constraint loss on T0 using Monte Carlo simulation together with the toughness
scaling model based on the Weibull stress. Here we build on this earlier work by incorporat-
ing improvements in the Weibull stress model, expanding the specimen sizes considered and
exploring the implications from measured differences in T0 -values.
The plan of the paper is as follows. Section 2 reviews briefly the Weibull stress as a local
fracture parameter and the thickness scaling model applicable under plane strain, SSY condi-
tions. Section 3 outlines numerical procedures to conduct the finite strain plasticity analyses
of the SE(B) specimens and the SSY boundary layer model. Section 4 develops the non-
dimensional function g(M) to describe constraint evolution in a convenient form. Section
5 describes the three-parameter Weibull model for the conditional probability that uses the
proposed threshold value σw−min . Section 6 describes the Monte Carlo simulation process and
provides numerical results to estimate the effects of constraint loss on T0 .
46 X. Gao and R.H. Dodds

2. The Weibull stress as a local fracture parameter

At temperatures in the transition region above the lower shelf, but prior to the appearance
of stable ductile tearing, deep notch (bend) fracture specimens fail by transgranular cleavage
in the presence of significant plasticity (often a plastic hinge). Transgranular cleavage frac-
ture develops from slip-induced cracking of carbides, generally located on grain boundaries,
followed by unstable propagation of the resultant cracks into the surrounding ferrite matrix
(McMahon and Cohen, 1965). Experimental studies reveal that the process becomes strongly
driven by a (often single) critical cleavage event at the metallurgical scale that triggers macro-
scopic brittle fracture. Carbide particles ‘eligible’ for cleavage initiation include those (Wang,
1991; Hahn, 1984):
(a) under sufficient tensile stress,
(b) with orientations favorable for nucleating a crack and
(c) producing a sufficient local energy release rate to sustain propagation.
Due to microstructural inhomogeneity of the material (carbide distribution, shape and orien-
tation, etc.), volume sampling effects play a paramount role to quantify the large, observed
scatter in fracture toughness data measured in the transition region. For initially sharp (fa-
tigue) cracks, the probabilistic sampling effect becomes tightly coupled with the near-front
constraint which governs the volume of material subjected to stresses sufficiently large to
initiate fracture. When constraint levels vary under increased loading (for example, due to the
influence of nearby traction free boundaries on crack-front plastic zones), there develops a
complex, nonlinear interaction between the constraint levels and the probabilistic modeling.
The development of a tractable model to quantify the failure probability follows by assum-
ing a random distribution of microscale flaws; their size and density constitute properties of the
material. The implicit distribution of cleavage fracture stress becomes explicit by introducing
a dependence between the critical microcrack size ac and stress in the form ac = (K 2 /Y σ 2),
where Y represents a geometry factor and σ denotes a tensile (opening) stress acting on the
microcrack plane. In the local approach to cleavage fracture, the probability distribution (Pf )
for the fracture stress of a cracked solid at a global load level KJ or J is assumed to follow a
two-parameter Weibull distribution (Beremin, 1983; Mudry, 1987; Minami et al., 1992) in the
form
 Z  m    m 
1 σ1 σw
Pf (σw ) = 1 − exp − dV = 1 − exp − , (1)
V0 V̄ σu σu
where V̄ denotes the volume of the cleavage fracture process zone, V0 defines a reference
volume for dimensional correctness and σ1 is the maximum principal stress acting on material
points inside the fracture process zone. The process zone includes the volume of material
inside the contour of σ1 > λσ0 , where σ0 is the yield stress and λ ≈ 2 is a constant (the
material points must have yielded at some point in the loading history). Parameters m and
σu appearing in (1) define the Weibull modulus and the scale parameter of the Weibull dis-
tribution. In particular, m defines the shape of the probability density function describing
the microcrack size a which is of the form f (a) ∝ a −γ with m = 2γ − 2. Following the
suggestion of Beremin (1983), the stress integral over the fracture process zone is denoted as
σw , and is termed the ‘Weibull stress’, where
 Z 1/m
1
σw = m
σ dV . (2)
V0 V̄ 1
 Constraint effects 47

The Weibull stress thus emerges as a local, crack-front parameter to couple remote loading
with a micromechanics model that incorporates the statistics of microcracks in a weakest link
philosophy. Under increased remote loading described by KJ or J , differences in evolution
of the Weibull stress σw = σw (J ) reflect the potentially strong variations in crack-front
stress fields due to the effects of constraint loss and volume sampling. The inherently three-
dimensional formulation for σw defined by (2) readily accommodates variations in KJ or J
along the crack front.
When plane strain and SSY conditions prevail, the relationship between σw and the applied
loading J takes on the simpler form (Gao et al., 1998b)
σwm = CBJ 2 , (3)
where C denotes a constant dependent upon the material flow properties, the Weibull parame-
ters (m and V0 ), and B is the thickness of the specimen. In plane strain SSY, the volume of the
cleavage process zone (i.e., ≈ the plastic zone) scales with thickness (B)×J 2 . For real fracture
specimens, the stress and deformation fields vary over the thickness, with stronger variations
occurring when the specimen thickness diminishes relative to the in-plane dimensions. Gen-
erally, mid-plane stresses have the largest values with sharp reductions at the traction-free
outside surfaces. Thus, the contribution to σw from each material point in the thickness direc-
tion may vary significantly; the simple effect of specimen thickness on cleavage probability
suggested by (3) no longer applies.
To quantify the non-uniformity of stress and deformation fields along the crack front direc-
tion (Z-direction, see Figure 1), we introduce the Weibull stress density function (σ̄w ) defined
as
Z
1
σ̄w (Z) =
m
σ m dĀ, (4)
V0 Ā(Z) 1
where Ā(Z) represents the area of the cleavage fracture process zone at the plane Z =
constant, such that the process zone volume is given by
Z B/2
V̄ = Ā(z) dZ. (5)
−B/2

For the general three-dimensional case, (3) then becomes

Z B/2
σwm = σ̄wm (Z) dZ, (6)
−B/2

where σ̄wm depends implicitly on the local J at crack front position Z through σ1 .
Gao et al. (1998a) computed the distributions of Weibull stress density along the crack
front for a series of specimens containing surface cracks tested in the lower transition region.
For each specimen, the post-test examination revealed close agreement between the location
of the cleavage initiation site on the crack front and the location of maximum Weibull stress
density.
48 X. Gao and R.H. Dodds

Figure 1. Quarter symmetric, three-dimensional finite element mesh for a plane-sided CVN size, SE(B) specimen.

3. Numerical procedures and computational models

3.1. F INITE ELEMENT PROCEDURES

Nonlinear analyses were performed on very detailed models of fracture specimens using the
three-dimensional research code, WARP3D (Koppenhoefer et al., 1998). This code executes
in parallel via explicit message passing and achieves very high levels of performance on
multi-processor computers. The code provides a Mises constitutive model in a finite-strain
framework and employs a continuously updated formulation naturally suited for solid el-
ements having only translational displacements at the nodes. Using equilibrium equations
linearized about the current configuration, the global solution proceeds in an incremental-
iterative (implicit) manner with nodal equilibrium stringently enforced at n + 1. To advance
the solution from n → n + 1, each Newton iteration utilizes the (consistent) tangent stiffness
computed for the current estimate of the solution at n + 1. Final increments of logarithmic
strain over n → n + 1 are then evaluated using the linear strain-displacement matrix evaluated
on the converged mid-increment configuration x n+1/2 .
Fracture models are constructed using three-dimensional, 8-node tri-linear hexahedral ele-
ments. The element is evaluated using a conventional 2 × 2 × 2 Gauss quadrature. Use of the
B̄ formulation (Hughes, 1980) minimizes mesh lock-ups as the deformation progresses into
a fully plastic, incompressible mode. J -integral values are evaluated with a domain integral
procedure (Moran and Shih, 1987) using domains defined outside material having highly non-
proportional histories at the crack front. The computed J -values exhibit a strong domain
(path) independence. The WARP3D code supports computation of both through-thickness
(average) J -values and pointwise J -values at each node location along the crack front. The
average J -values provide a convenient parameter to characterize the average intensity of far
 Constraint effects 49

field loading on the crack front. Moreover, they agree with J -values derived from estimation
schemes based upon η-factors for deformation plasticity, which supports a direct connection
of analysis values with experimentally measured Jc -values.

3.2. C ONSTITUTIVE MODELS

The material model employed in the parametric studies follows a J2 flow theory with con-
ventional Mises plasticity. All constitutive computations are performed within the finite-strain
framework. The uniaxial true stress-logarithmic strain curve obeys a simple power-law hard-
ening model preceded by a purely linear response before plastic flow,
 n
ε σ̄ ε σ̄
= for ε 6 ε0 ; = for ε > ε0 , (7)
ε0 σ0 ε0 σ0

where σ0 and ε0 denote the reference (yield) stress and strain, and n is the strain hardening
exponent. A small transition region eliminates the discontinuous tangent modulus at ε = ε0 ;
nonlinearity of the σ − ε curve actually begins at ε = 0.95ε0 . The transition region improves
the convergence rate of the global Newton iterations.

3.3. T HREE - DIMENSIONAL FINITE ELEMENT MODEL FOR SE ( B ) SPECIMEN

This study analyses the nonlinear response for a series of geometrically similar, plane-sided
SE(B) specimens (B = W, S = 4W and a/W = 0.5). The absolute specimen dimensions
scale with the thickness, i.e., all the dimensions for a 2T specimen equal twice those for a 1T
specimen. Figure 1 shows a typical, quarter-symmetric finite element model for the Charpy
V-notch size specimen (B = W = 10 mm), i.e., a 0.4T size. The model includes a standard
V-notch above the fatigue crack although the notch has a negligible effect.
A conventional mesh configuration having 44 focused rings of elements in the radial direc-
tion surrounds the crack front. The crack tip has a small, initial radius to enhance convergence
of the finite-strain solutions. To limit effects of the initial root radius on Weibull stress calcu-
lations, the analyses use finite element models having two different root radii of 0.25 µm and
2.5 µm. The model with a 0.25 µm radius provides Weibull stress results early in the loading
history at relatively low J -values (the deformed radius must exceed 2.5× the initial radius
before results are considered valid). At larger J -values, element distortions along the crack
front reach unacceptable levels with this model. The second model, having the larger initial
root radius (2.5 µm), permits loading to much higher J -values. At intermediate J -values,
both models generate essentially identical results, providing a verification of this strategy. For
all specimen sizes, meshes in the crack front region have a fixed element density and element
sizes fixed at those for the CVN specimen. This eliminates possible near-front meshing effects
as the specimen sizes increase. The typical mesh contains 34,200 nodes and 30,500 elements,
with 20 variable thickness layers defined over the half-thickness. The thickest layer lies at
Z = 0 with thinner layers defined near the free surface (Z = B/2) to accommodate strong
Z variations in the stress distributions. We apply displacement boundary conditions to model
the loading of the SE(B) specimen: 1u = 0 is imposed on nodes along the line of support and
1u < 0 is imposed uniformly on all nodes along the load line. At load levels well beyond the
formation of a plastic hinge, the outermost loaded node does develop a very small uplift force
(< 1 percent of the downward force at the adjacent node).
50 X. Gao and R.H. Dodds

Figure 2. (a) Small-scale yielding (SSY) model. (b) Near-tip mesh, where the initial notch root radius is 2.5 µm.

3.4. S MALL SCALE YIELDING MODEL (SSY)

The plane-strain, SSY solution establishes the reference, crack front conditions to quantify the
evolution of constraint loss in SE(B) specimens. In plane-strain SSY, the local J and the stress-
deformation fields remain identical at each crack front location over the thickness, which leads
to the simple form of (3). This idealized model defines a very severe (conservative) constraint
level attainable in real fracture specimens only at low loading levels. The severity of this
reference condition plays a key role in the proper interpretation of subsequent computational
results.
Use of the modified boundary layer model (Larsson and Carlsson, 1973; Rice, 1974) sim-
plifies the generation of numerical solutions for stationary cracks under SSY conditions with
varying levels of constraint governed by the T -stress, see Figure 2. With the plastic region
limited to a small fraction of the domain radius (R) included in the model Rp < R/20, the
general form of the asymptotic crack-tip stress fields well outside the plastic region is given
by (Williams, 1957)
K
σij = √ fij (θ) + T δ1i δ1j , (8)
2π r
where K is the stress intensity factor fij (θ) defines the angular variations of in-plane stress
components, and the non-singular term T represents a tension (or compression) stress parallel
to the crack plane. Numerical solutions for T = 0 are generated by imposing displacements
of the elastic, Mode I singular field on the outer circular boundary (r = R) that encloses the
crack
r
1+ν R
u(R, θ) = K cos( 12 θ)(3 − 4ν − cos θ), (9)
E 2π
r
1+v R
v(R, θ) = K sin( 12 θ)(3 − 4v − cos θ). (10)
E 2π
The SSY model used for analysis (Figure 2) has one layer of 2,800 three-dimensional elements
with plane-strain constraints (w = 0 imposed at all nodes). Under these conditions J =
K 2 (1 − v 2 )/E.
By coupling finite element results for the SSY model with (3), accurate values of the
Weibull stress may be obtained for any loading level (J ) in SSY. For an assumed value of
 Constraint effects 51

m, pairs of computed Weibull stress and applied J values from the finite element analysis
enable a simple computation for the constant C in (3). Finite element results are used once
the deformed root radius exceeds 3× the initial radius. This scheme minimizes the effects of
initial notch root radius and near tip meshing, and it provides a closed-form expression for the
SSY Weibull stress values. Simple scaling of the SSY Weibull stress values to the thickness of
an SE(B) specimen as indicated by (3) provides the values for assessment of constraint loss.

3.5. N UMERICAL COMPUTATION OF THE W EIBULL STRESS

Numerical evaluation of the Weibull stress, defined by (2), using the finite element results pro-
ceeds as follows. Let |J | denote the determinant of the standard coordinate Jacobian between
the deformed Cartesian coordinates xi and the parametric coordinates ηi . Then using standard
procedures for integration over element volumes, the Weibull stress has the form
" Z Z Z #1/m
1 X 1 1 1 m
σw = σ |J | dη1 dη2 dη3 , (11)
V0 n −1 −1 −1 1
e

where ne is the number of elements inside the fracture process zone. The process zone used
here includes all material inside the loci σ1 > λσ0 , with λ = 2 (material points must also
experience plastic deformation). Values for λ other than 2 have only very minor effects on
the Weibull stress values (see Minami et al., 1992; Ruggieri and Dodds, 1998). In view of
the very refined meshes, an element lies in the fracture process zone if the σ1 computed at
η1 = η2 = η3 = 0 exceeds λσ0 . This expression for the Weibull stress represents the integral
form in parametric space of Beremin’s formulation (Beremin, 1983).
The reference volume (V0 ) equals 1 mm3 for convenience in all calculations. As (2) indi-
cates, the Weibull stress scales with this reference volume. A change of the reference volume
from V0 to V̄0 requires scaling of σu to σ̄u such that V0 × σum = V̄0 × σ̄um to maintain identical
failure probabilities.

4. Constraint correction model based on Weibull stress

When coupled with nonlinear analyses of test specimens, measured values of macroscopic
fracture toughness (KJ c , Jc ) yield corresponding sets of Weibull stress values at fracture
(assuming m has been calibrated). For different specimens (type and /or size) constructed of
the same material, and tested under similar conditions (temperature, loading rate), critical J -
values that generate the same critical Weibull stress values generally differ in the specimens
due to the combined effects of constraint loss and absolute size. From (1), equal values of
cleavage probability require equal values of the Weibull stress. This equivalence of local frac-
ture parameters enables development of size and constraint corrections to scale macroscopic
toughness between different crack geometries and /or applied loading conditions at identical
σw -values.

4.1. A NON - DIMENSIONAL CONSTRAINT FUNCTION – g(M)

To introduce a new approach that aids in quantifying constraint loss, this section describes
key results derived from detailed three-dimensional finite element analyses of a 0.4T SE(B)
52 X. Gao and R.H. Dodds

Figure 3. (a) Distribution of the opening mode stress ahead of the crack front on θ = 0 at the centerplane of 1T
SE(B) and CVN specimens at three deformation levels (M = 100, 50, 30). (b) Normalized Weibull stress density
distribution at four deformation levels in 1T SE(B) (solid line) and CVN (dash line) specimens. The material flow
properties are n = 10, E/σ0 = 500 and v = 0.3, and the Weibull modulus is m = 20.

specimen (CVN) and a 1T SE(B) specimen, both with B×B cross-sections. The material prop-
erties used in the computations are: E/σ0 = 500, v = 0.3 and n = 10. The through-thickness
(average) value of J quantifies the level of applied loading. Figure 3a plots the opening mode
stress ahead of the crack front at the center-plane of the specimen with distances normalized by
the similarity length-scale. The parameter M = bσ0 /J defines the (non-dimensional) loading
level scaled to the specimen size (b denotes the remaining ligament length). These results
show that:
(1) in non-dimensional coordinates, stress distributions for the CVN specimen and 1T SE(B)
specimen are essentially identical at each relative deformation level (M);
(2) the peak stress in SE(B) specimen decreases just slightly due to constraint loss as defor-
mation increases; and
(3) for SE(B) specimens, the shape of σ1 /σ0 vs. X/(J /σ0 ) curve changes sharply as the
deformation increases due to increased gradient of the bending field.
For reference, Figure 3a also includes the opening mode stress for the plane strain, SSY
(T = 0) solution. The first point demonstrates that the opening mode stress in the CVN
size specimen agrees with the crack field solution for an initially sharp tip that blunts a small
amount (compared to the remaining ligament) during loading. However, further decreases in
specimen size must show a gradual change in the stresses to those of a finite size notch, where
the ligament exceeds only a few multiples of the notch radius.
Figure 3b shows the evolution of Weibull stress density across the crack front under in-
creased loading for a representative Weibull modulus of m = 20. When normalized appro-
priately, the Weibull stress densities remain invariant of specimen size at each loading level.
The SSY result for σwm given by (3) suggests the proper normalization with the constant C
taken from the finite element solution of the SSY model for m = 20. The density remains
nearly constant over 60 percent of the thickness for these plane-sided specimens. At higher
loading levels M < 100 the peak density occurs away from the centerplane due to anticlastic
 Constraint effects 53

bending effects in these B × B specimens (the uncracked section thickness equals twice the
ligament length). Analyses for specimens with standard W = 2B show maximum densities at
the centerplane for all loading levels.
Dimensional analysis suggests, and numerical experiments confirm, the validity of these
results for all SE(B)s having B = W, S = 4W and a/W = 0.5. The response curves in
Figure 3b make possible a generalization of the Weibull stress expression for all similar SE(B)
specimens. Let s = Z/(B/2) denote the non-dimensional location of a point on the crack
front. Then each curve f in Figure 3b has the form f (M, s) = σ̄wm /(CJ 2) such that
Z 1
σwm = CBJ 2
f (M, s) ds = CBJ 2 g(M), (12)
−1

where
Z 1 Z 1
g(M) = f (M, s) ds = 2 f (M, s) ds. (13)
−1 0

g(M) depends on material flow properties and the Weibull modulus but not on the absolute
size of the specimen. Larger values of g(M) indicate higher levels of crack front constraint.
For plane strain SSY (T = 0), g(M) = 1 regardless of material properties. For fracture
specimens, the deformation level (M) at which g(M) falls below unity does depend on the
flow properties.
Knowledge of the g(M) function for a specimen geometry enables construction of σw vs.
J curves for all geometrically similar specimens without additional finite element analyses.
For example, by modeling a 1T SE(B) specimen with B = W, S = 4W and a/W = 0.5, the
generated g(M) function remains valid for all SE(B) specimens having sizes proportional to
this 1T specimen. For other sizes of SE(B) specimens, application of (12) yields the σw vs. J
curves. The σw vs. J relationship reveals explicitly the size effect on fracture toughness.
Figure 4 shows g(M) vs. M relationship for SE(B) specimens with B = W, S = 4W and
a/W = 0.5 and having material properties of E/σ0 = 500, ν = 0.3, n = 10 and m = 20. The
value of g(M) exceeds unity for M > 200 because of the positive T -stress in the deep-notch
SE(B) specimen early in the loading (the SSY reference defining g = 1 has T = 0). The value
of g(M) decreases as deformation increases, reflecting constraint loss in SE(B) specimen.

4.2. S CALING OF FRACTURE TOUGHNESS FOR CONSTRAINT LOSS AND SIZE

Deviations of the g(M) function from unity demonstrate that the Weibull stress for different
size specimens, loaded to different J -values, does not follow the simple scaling model defined
by (3) for plane strain SSY. In real specimens, there exists a complex interaction between
in-plane and thickness effects on the crack front stress fields that changes continuously un-
der increased plastic deformation. Figure 5 illustrates these effects and the potential error
introduced through use of the simple scaling model of (3). Here, the use of unnormalized
deformation levels makes clear the effects of absolute specimen size on constraint loss, once
SSY conditions no longer prevail. Figure 5a shows the evolution of Weibull stress densi-
ties at the centerplane of the CVN, 1T SE(B) specimens and plane strain SSY. Symbols
on the SE(B) curves indicate specific normalized deformation levels (M) of interest. For
m
convenience, values of the Weibull stress density are normalized by a constant (σ̄w,SSY at
√
KI = 100 MPa m). Both the CVN and 1T SE(B)s show the quadratic increase of Weibull
SSY
54 X. Gao and R.H. Dodds

Figure 4. The non-dimensional constraint function, g(M), for SE(B) specimens with B = W, S = 4W and
a/W = 0.5. The material flow properties are n = 10, E/σ0 = 500 and v = 0.3, and the Weibull modulus is
m = 20.

stress density with J early in the loading which agrees with the plane strain SSY model. At a
higher toughness level, e.g., J = 60 kJ/m2 , the CVN size specimen loses significant constraint
on the centerplane (arrow A), while the 1T SE(B) just begins to lose centerplane constraint
(arrow B). Compared to plane strain SSY, the centerplane of the SE(B) specimens all begin
losing constraint when the relative deformation level (M) falls below . 100.
Figure 5b compares the total Weibull stress for the crack front (σwm) with J for different
configurations. Early in the loading, the stress distributions remain nearly uniform at SSY
levels over the whole thickness; σwm then scales linearly with specimen thickness as in (3). At
higher loadings, the 1T SE(B) Weibull stress falls below that for the reference configuration of
1T SSY. Constraint loss alone, starting at M = 150, creates the growing discrepancy between
the two curves as indicated by the arrow labeled (A). Now consider the CVN response curve
shown on the figure. If the CVN response matched plane strain SSY, the 1T SSY curve could
be predicted by the simple scaling model based on crack front length, i.e., the 1T SSY values
would equal 1.0/0.4 = 2.5× the CVN values. This clearly does not occur as indicated by
the dotted line of CVN values scaled by 2.5 and the arrow labeled (C). Moreover, the scaled
CVN values do not even agree with the 1T SE(B) values that already exhibit constraint loss.
The arrow labeled (B) shown between the two curves arises from constraint differences in the
CVN and 1T SE(B) when loaded to the same J -value (which would vanish if both followed
SSY). Finally the combined (A) and (B) arrows indicate the degree of constraint loss in the
CVN specimen relative to plane strain SSY.
The nonlinearity present in the scaling of Weibull stresses demonstrated in Figure 5 pre-
cludes use of the simple SSY scaling except early in the loading. The approach described
here derives from the toughness scaling model of Anderson and Dodds (1991) and Dodds
et al. (1991). From (1), equal probabilities of cleavage fracture occur when different size
and/or type of specimens are loaded to the same Weibull stress, even though J -values may
differ significantly due to constraint loss and absolute size. The requirement of equal Weibull
stress values at fracture in the specimens enables construction of a curve that relates directly
the different J -values. Figure 6 displays the toughness scaling model for deep-notch SE(B)
specimens relative to the reference condition of the plane strain SSY model having the same
thickness as the SE(B) specimen. Each point on the curve defines a pair of J -values, J for
the SE(B) and J0 for SSY, that produce the same σw -value. The J -values on both axes are
 Constraint effects 55

Figure 5. (a) Comparison of Weibull stress densities at the centerplane of 1T SE(B) and CVN specimens with
plane strain SSY. (b) Comparison of the σw vs. J relationships for different configurations showing the combined
effects of absolute size, constraint loss and the error introduced by the simple thickness scaling model. The material
flow properties are n = 10, E/σ0 = 500 and v = 0.3, and the Weibull modulus is m = 20.

normalized by bσ0 , where b denotes the remaining ligament length of the SE(B) specimen.
This normalization permits application of the same curve for all geometrically similar SE(B)
specimens with B = W, S = 4W and a/W = 0.5 regardless of absolute size. As shown in
the figure, the normalized curve generated for the CVN model exactly matches the 1T SE(B)
curve. Early in the loading, the J0 vs. J curves almost coincide with the 1:1 line (actually
the curves rise slightly above the 1:1 line due to the small positive T -stress in the SE(B)). As
loading continues, the J0 vs. Javg curves fall below the 1:1 line indicating a loss of constraint
(relative to the reference configuration). The toughness scaling model as illustrated in Figure 6
includes the effects of constraint loss and differences in absolute size. This scaling model
provides the correct (nonlinear) relationship between all geometrically similar SE(B)s and
the SSY condition. For example, given a material toughness in the SSY condition (J0 ), the
corresponding Javg for a small size SE(B) specimen and a large size SE(B) specimen are found
as shown on Figure 6. This resolves the problem described in Figure 5b to scale absolute size
and constraint loss effects between the 1T SE(B) and CVN specimens.
The non-dimensional constraint function g(M) captures the nonlinearity of the toughness
scaling model described above. To show this, apply the scaling model to two geometrically
similar specimens with different sizes (denoted 1 and 2) and loaded to different levels (M1 and
56 X. Gao and R.H. Dodds

Figure 6. The normalized toughness scaling curve for SE(B)s with B = W, S = 4W and a/W = 0.5, where
J0 represents toughness value for plane strain SSY. The material flow properties are n = 10, E/σ0 = 500 and
v = 0.3, and the Weibull modulus is m = 20.

M2 ). Now equate the Weibull stresses for each specimen, using (12) to define each Weibull
stress, to yield
C·B1 ·g(M1 )·J12 = C·B2 ·g(M2 )·J22 , (14)
and solving for J2 to find
   1/2
g(M1 ) 1/2 B1
J2 = · · J1 . (15)
g(M2 ) B2
Because M2 is a function of J2 , a simple iterative method is needed to obtain J2 from (15) for
a known value of J1 . Under plane strain and SSY conditions g(M1 ) = g(M2 ) ≡ 1 then
 1/2
B1
J2 = · J1 , (16)
B2
which defines the widely used relation to scale specimen thickness effects on cleavage fracture
toughness; however, it is strictly applicable only under plane strain, SSY conditions.
Figure 7a illustrates the magnitude of the term to ‘correct’ the simple scaling model of
(16), [g(MCVN )/g(M1T )]1/2 with deformation levels in the CVN specimen. Here the subscripts
‘CVN’ and ‘1T’ denote the CVN and 1T SE(B) specimens, respectively, where MCVN and
M1T correspond to the same fracture probability (Weibull stress). The difference between
[g(MCVN )/g(M1T )]1/2 and unity exceeds 20 percent when MCVN . 100. Figure 7b casts these
results into the form of a scaling model where the reference configuration now becomes the
actual 1T SE(B) specimen. The dotted line indicates the scaling based on (16) which remains
applicable for MCVN . 100, but with growing error as the deformation increases.

5. A three-parameter Weibull stress model

The Weibull stress model defined by (1–2) and discussed in previous sections represents
a pure weakest link description of the fracture event. This two-parameter model describes
 Constraint effects 57

Figure 7. (a) The correction factor for simple thickness scaling formula needed to scale JCVN to 1T size SE(B)
value, where MCVN and M1T correspond to same fracture probability. (b) The error of simple scaling grows larger
with increasing deformation. The material flow properties are n = 10, E/σ0 = 500 and v = 0.3, and the Weibull
modulus is m = 20.

the unconditional cleavage probability that assumes that no microcracks arrest (macroscopic
cleavage fracture occurs once the critical microcrack experiences propagation). However, the
unconditional probability has significant shortcomings to predict cleavage fracture (Anderson
et al., 1994; Gao et al. 1998b). First, it implies that a very small KI leads to a finite failure
probability, which is not true in reality. Cracks cannot propagate in polycrystalline metals
unless sufficient energy exists to break bonds, to drive the crack across grain boundaries and to
perform plastic work. Consequently, there must exist a minimum toughness value below which
cracks arrest. Second, the unconditional probability often over estimates the measured scatter
of fracture toughness (see Figure 8b for an example). Finally, when applied to the analysis of
deep notch SE(B) specimens as described in the previous sections, the two-parameter Weibull
stress model predicts constraint loss at much lower deformation levels (i.e. M ≈ 100) than
the M = 30 − 50 level observed in large-scale testing programs (Wallin, 1993).
Recognition of the first shortcoming led previous researchers (Bakker and Koers, 1991; Xia
and Shih, 1996) to introduce a threshold stress σth directly into the Weibull stress computation
as in (11), where σth is usually taken as a small multiple of the uniaxial yield stress (e.g. σth =
2−3×σ0 ). One such proposal for the integrand of (1) has the form {(σ1 −σth )/(σu −σth )}m . Ra-
tional calibration procedures for σth remain an open issue. To address the shortcomings listed
above, we propose an alternative expression for the cleavage fracture probability in terms
of the Weibull stress that parallels currently adopted expressions for macroscopic toughness
values. A simplified, three-parameter Weibull distribution (Anderson et al., 1994) that approx-
imates the complex modeling of conditional probabilities for cleavage arrest/propagation has
the form

"   #
KI − Kmin 4
Pf (KI ) = 1 − exp − , (17)
K0 − Kmin
58 X. Gao and R.H. Dodds

Figure 8. (a) Comparison of different forms of cumulative probability function. (b) Comparison of predicted
cleavage probabilities using different forms of Weibull stress models with experimental rank probabilities for A36
steel SE(B) specimens (taken from Gao et al., 1998).

where Kmin denotes the minimum toughness value for cleavage fracture. Only when KI >
Kmin does macroscopic cleavage
√ fracture become possible. Kmin has an experimentally es-
timated value of 20 MPa m for common ferritic steels under SSY conditions, independent
of the crack front length. ASTM E1921 (1998) adopts (17) for the toughness distribution
in the ductile-to-brittle transition range when the deformation levels at fracture maintain
approximate SSY conditions.
Similar to (17), a minimum value of Weibull stress σw−min can be introduced to modify the
unconditional probability defined by (1). Equations (18) and (19) define two of the possible
forms to include σw−min in Pf (σw ), i.e.,
 !4 
m/4
σw − σw−min
m/4
Pf (σw ) = 1 − exp − m/4 m/4
, (18)
σu − σw−min

or
   
σw − σw−min m
Pf (σw ) = 1 − exp − , (19)
σu − σw−min

where σw−min represents the minimum σw -value at which macroscopic cleavage fracture be-
comes possible. Consistent with the definition of Kmin , σw−min is defined as the value of σw
calculated at KI = Kmin in the (plane-strain) SSY model, where the SSY model must have a
thickness equal to the configuration of interest for which (18) or (19) is applied.
Figure 8a compares different forms of cumulative failure probability, with m = 10 and
σw−min = 0.5σu used for illustration. The introduction of σw−min reduces the scatter and
(19) provides the least scatter. Although we cannot offer a theoretical justification for (19)
at present, experimental studies on several ferritic steels provide strong support for its use.
Figure 8b (taken from Gao et al., 1998b) compares predictions of cumulative failure proba-
bilities using (1) and (19) with experimental data for an A36 ferritic steel tested by Sorem
et al. (1991). The Weibull modulus has a calibrated value of m = 7.8 for both (1) and
 Constraint effects 59

Figure 9. Effect of σw−min on toughness scaling between CVN and 1T SE(B) specimens.

(19); the introduction of σw−min does not alter the calibration of m for the two models. The
symbols indicate the measured Jc -values and the corresponding rank probabilities. The two-
parameter model over predicts the scatter of the measured fracture toughness data while the
three-parameter model provides a very good fit to this experimental data. More importantly,
the three-parameter model of (19) accurately predicts the practically important tail of the
toughness distribution (the low probability region). Bass et al. (1998) report similar improve-
ments in modeling the cumulative failure probability using (19) for an A533B pressure vessel
steel.
According to (19), the toughness scaling model between different specimen sizes, types
and loading conditions should be constructed at identical σw∗ values, where σw∗ = σw − σw−min.
Figure 9 re-examines the toughness scaling between CVN and 1T SE(B) specimens. The
material properties remain unchanged from those used in Section 4, i.e., E/σ0 = 500, ν =
0.3, n = 10 and m = 20. For the three-parameter Weibull formulation (19), σw−min is obtained
as the value of σw at KI = Kmin in the plane-strain, SSY model having the thickness of each
specimen. The different thicknesses of the CVN and 1T SE(B) lead to different values of
σw−min . The scaling model including σw−min (> 0) increases the predicted amount of defor-
mation prior to constraint loss. The simple thickness scaling model for toughness now breaks
down strictly at M ≈ 70 but at M ≈ 40 when a 20 percent deviation is allowed (i.e., a 10
percent difference in KJ c ). The value of M = 40 is consistent with experimental results and
approaches the M=30 value adopted in ASTM E1921 for use with all materials independent
of strength (E/σ0 ) and strain hardening levels.

6. Specimen size and constraint loss effects on T 0

Recent efforts to develop a standardized testing and data analysis procedure for fracture
toughness characterization in the ductile-to-brittle transition region have produced ASTM
E1921 (ASTM, 1998). Applications of this testing standard determine the temperature, de-
noted T0 , at which the median fracture toughness of the material for a 1T size, deeply-notched
60 X. Gao and R.H. Dodds
√
SE(B) specimen equals 100 MPa m. The testing protocol and data analysis rely upon key
assumptions that:
(1) conditions sufficiently near SSY prevail at fracture such that the parameter KJ c uniquely
characterizes the crack front fields, and
(2) the simple size scaling model for toughness described by
(3) remains applicable (the standard expresses (3) in terms of KJ rather than J ).
Violations of these assumptions, which most likely occur in testing small specimens such as
the CVN, generally decrease the T0 -value, thereby overestimating the real transition region
toughness. The toughness scaling model based on the Weibull stress developed in Sections 4
and 5 provides an analytical tool to estimate quantitatively the effects of such violations on
the T0 -value. The following subsections outline key features of the data analysis procedure to
determine T0 , describe a Monte Carlo procedure that simulates experimental testing to study
the effects of constraint violations on T0 , and lastly present and discuss the simulation results.

6.1. S UMMARY OF THE T0 TEST METHOD

The experimental protocol requires static, elastic-plastic fracture tests performed on standard
SE(B) or C(T) specimens having deep notches (a/W = 0.5) to measure the J -integral values
at cleavage fracture (denoted Jc ). The test temperature (T ) and configuration of all specimens
must be identical (0.5T, 1T, 2T, etc.); future versions of the standard will remove this lim-
itation. T should be selected to lie in the lower part of the DBT region as near as possible
to the eventual T0 . The standard requires a minimum of six replicate tests that meet (fatigue)
crack front straightness tolerances, limits on ductile tearing prior to cleavage, size/deformation
limits, etc. Measured J -integral values at fracture are first converted to their equivalent units
of stress intensity factor (KJ c ). Then KJ c values exceeding
p
KJ c(limit) = Ebσ0 /[(1 − v 2 )M] (20)

are set to the limiting value and marked for subsequent censoring. E1921 sets M = 30, where
M = bσ0 /J is the same non-dimensional deformation limit used previously. For test programs
conducted on other than 1T specimens, measured toughness values are adjusted to their 1T
equivalent values using the simple scaling model of (3), modified for the threshold toughness,
 1/4
Bx √
KJ c(1T) = 20 + [KJ c(x) − 20] MPa m, (21)
B1T

where B1T denotes the 1T specimen size (25 mm) √ and Bx denotes the corresponding dimen-
sion of the test specimens. In the above,√20 MPa m represents the minimum (threshold)
fracture toughness (i.e., Kmin = 20 MPa m) adopted for ferritic steels addressed by the
standard.
Equation (17) describes the toughness distribution having a three parameter Weibull form.
A maximum likelihood estimate for K0 , the scale factor corresponding to the 63.2 percent
cumulative failure probability, is given by
" #1/4
X
N
(KJ c(i) − 20)4 √
K0 = + 20 MPa m, (22)
i=1
(r − 0.3068)
 Constraint effects 61

where N denotes the total number of specimens tested (censored and uncensored) with r
the number of uncensored tests (6 minimum). The median toughness at the test temperature
follows simply as
√
KJ c(med) = 0.9124(K0 − 20) + 20 MPa m. (23)
The ‘master curve’ of median toughness for 1T specimens over the transition range for the
material has the final form (Wallin, 1991; ASTM, 1998)
√
KJ c(med) = 30 + 70 exp[0.019(T − T0 )] ◦ C, MPa m. (24)
This expression for the median toughness applies throughout the lower part of the DBT range
prior to the appearance of significant ductile tearing. The E1921 document outlines standard
procedures to construct various tolerance bounds based on the above representation for the
toughness distribution.
Finally, the difference between the test temperature (T ) and the reference temperature (T0 )
follows from inversion of the master curve (24) and is given by
 
1 KJ c(med) − 30 ◦ √
T − T0 = ln C, MPa m. (25)
0.019 70

6.2. S IMULATION PROCEDURE

The E1921 procedure to determine T0 requires (theoretically) that plane strain, SSY condi-
tions prevail at fracture in each of the replicate test specimens used to compute the statistical
estimate for K0 , and to adjust KJ c -values obtained with other than 1T specimens to their 1T
equivalent values. When constraint loss occurs in the measured fracture toughness values, the
derived T0 has a value lower than would be determined for a data set of toughness values
measured under higher constraint conditions. Within the minimum set of six (6) measured
values of fracture toughness, specimens invariably fail over a potentially large range of crack
front constraint levels, i.e., from large M values (> 200) to those at or near the deformation
M = 30. Each specimen influences the aggregate estimate for T0 ; the effects of lower con-
straint specimens is examined here. We employ a modified version of the approach proposed
by Ruggieri et al. (1998) to evaluate the shift of T0 caused by the nonlinear interaction of
constraint loss and specimen size. A Monte Carlo procedure generates trial sets of SSY
fracture toughness (KJ c ) data that follow the three-parameter Weibull distribution with the
Weibull slope of 4, (17),
gen √
KJ c = (K0 − 20)[−ln(1 − U )]1/4 + 20 MPa m, (26)
where U denotes a random variate uniformly distributed between [0, 1]. The numerical pro-
cedure utilizes a random number generator that reflects a linear congruential procedure (Press
et al., 1992) to produce the uniformly distributed random variate U in the unit interval. Each
SSY toughness value in a trial set is corrected in a reverse manner using the toughness scaling
procedure discussed in Section 5 (with σw−min > 0) to define the equivalent LSY value, that
would presumably be measured in an actual test. Using (20–25), the two sets of SSY and LSY
KJ c -values determine the corresponding relative temperatures T − T0 . To ensure sufficient
precision in the simulation process, the process generates 10,000 trial sets of SSY fracture
toughness data. The simulation proceeds as follows (see Ruggieri et al., 1998 for additional
details):
62 X. Gao and R.H. Dodds

1. Specify the material flow properties; the characteristic toughness (K0 ) for a 1T SSY
configuration, denoted K0SSY(1T) ; and a deformation limit (M) for censoring (typically,
30–50).
2. Execute Monte Carlo simulations to generate a single trial set of toughness values:
(2.1) Generate a random number, Pf , over [0, 1].
(2.2) Compute the toughness value KJSSY(1T) c(i) from (17).
SSY(1T) LSY
(2.3) Reverse correct KJ c(i) to KJ c(i) for the specimen size of interest, using the
toughness scaling model in Section 5 (with σw−min > 0).
(2.4) Check validity of KJLSY c(i) against the SSY deformation limit using (20).
(2.5) If validity limit exceeded, set KJLSY c(i) to KJ c(limit) and mark for censoring.
(2.6) Repeat steps 2.1–2.6 until six uncensored (‘valid’) KJLSY c(i) -values are obtained.
LSY LSY(1T)
3. Adjust each KJ c(i) -value to 1(T) equivalence, KJ c(i) using (21).
4. Compute a maximum likelihood estimate of K0 for the KJSSY(1T) c and KJLSY(1T)
c data sets
using (22) and taking into account the number of censored toughness values from step
2.5.
5. Compute the median value, KJ c(med) , of the KJSSY(1T)
c and KJLSY(1T)
c toughness values using
(23).
6. Compute relative reference temperature for both SSY and LSY toughness distributions,
(T − T0 )SSY and (T − T0 )LSY using (25).
7. Repeat steps 2 through 6 for the specified number of trials, e.g., 10,000.
8. Using results for the trials, compute the average and standard deviation for the shift in T0
due to constraint loss, 1T0 = (T − T0 )SSY − (T − T0 )LSY . The value of 1T0 approaches
zero when the deformation limit, relative to specimen size and material flow properties,
prohibits the use of test results that violate the SSY requirement and thus the simple
thickness scaling model.
9. Repeat steps 1–8 for a range of material flow properties, characteristic toughness val-
ues, SSY deformation limit (M), and Weibull modulus m to generate the SSY→LSY
correction.

6.3. N UMERICAL RESULTS

The parameter study described here considers three sets of material flow properties, n = 5, 10
and 20 with strengths E/σ0 = 800, 500 and 300 respectively, which represent a wide-range of
ferritic steels. Effects of scatter in the local toughness values enter the simulations by varying
the Weibull modulus with m = 16 and m = 20. To study size effects, the parameter study uses
plane-sided SE(B) specimens with B = W, S = 4W and a/W = 0.5 and varying absolute
sizes. As discussed in Section 4.1, a single finite element analysis of this SE(B) configuration
generates the constraint function g(M) for each different set of material flow properties. From
the g(M) function, the σw vs. J relationship (needed to construct the reverse correction in step
2.3) for each specimen size of interest follows by applying (12). A nonzero threshold Weibull
stress, σw−min , has a significant effect on failure probabilities (and thus the toughness scaling
model); the σw−min is included through use of (19) as detailed in Section 5.
The single finite element analysis for each of the three sets of material flow properties is
performed on a 1T SE(B) specimen. Each analysis uses two models that differ in the size
of the initial root radius, 0.25 µm and 2.5 µm. The smaller radius model provides Weibull
stress values early in the loading. At higher loading levels, the larger radius model becomes
 Constraint effects 63

Figure 10. The g(M) function for SE(B) specimens with B = W, S = 4W and a/W = 0.5. Two dif-
ferent values of Weibull modulus are considered: m = 16 and m = 20. The material flow properties are
(a) E/σ0 = 500, v = 0.3 and n = 10, (a) E/σ0 = 800, v = 0.3 and n = 5, and (a) E/σ0 = 300, v = 0.3
and n = 20.

necessary to avoid excessive mesh distortions at the crack front. In both cases, numerical
results are considered valid only when the deformed root radius exceeds 2× the initial radius.
Figure 10 shows the computed g(M) function for each of the three material property sets. The
n = 10 and n = 20 material show significant constraint loss at higher deformation levels
while the n = 5 material displays much less constraint loss. The Weibull modulus m has the
most (relative) effect on constraint loss for the n = 10 material and least effect for the n = 5
material.
Figures 11–13 summarize results of the parameter study to estimate the decreased value
of T0 , i.e., 1T0 as a function of K0 , taken here as the characteristic toughness of the material.
These 1T0 -values represent the mean defined over the 10,000 sets of 6 (uncensored) toughness
values for each configuration and material. Negligible differences in the 1T0 -values occur
when the simulations require, for example, 10 rather than 6 uncensored values in each trial.
The reference condition to assess constraint loss is plane strain, SSY (with zero T -stress)
having the same thickness as the specimen. The use of K SSY(1T) for the abscissa enables direct
comparison across varying absolute sizes, where K0SSY(1T ) denotes the specified K0 for the 1T,
plane strain SSY reference condition used in the Monte Carlo simulations.
The effects of material properties, Weibull modulus, deformation limit and specimen size
on 1T0 have been examined. For all cases 1T0 increases with increasing levels of K0 and
saturates at higher toughness values to nearly equal levels as censoring caps an increasingly
64 X. Gao and R.H. Dodds

Figure 11. 1T0 vs. K0 curves for materials having E/σ0 = 500, v = 0.3 and n = 10, where K0 is the
characteristic toughness value of the 1T SSY configuration. (a) m = 20, (b) m = 16.

Figure 12. 1T0 vs. K0 curves for materials having E/σ0 = 300, v = 0.3 and n = 20, where K0 is the
characteristic toughness value of the 1T SSY configuration. (a) m = 20, (b) m = 16.

larger number of fracture toughness values. Specimen size has a marked effect on 1T0 at
low K0 levels i.e., 1T0 increases as specimen size decreases. Even at relatively low K0SSY(1T)
values, the 1T and 2T size specimens show some apparent effects of constraint loss. At these
low K0SSY(1T) values, the reduction in Weibull stress density near the free surfaces of the
specimens leads to this small decrease of T0 , relative to the reference condition of plane strain,
SSY.
These results also demonstrate the difficulty in assessing the relative effects of constraint
loss using only experimental data. For example, consider Figure 11(a) where tests on CVN
and 1T size SE(B)s at different temperatures could lead to different K0SSY(1T) values, and thus
different values for 1T0CVN and 1T01T . However, the experimental data actually provides just
the difference 1T0CVN − 1T01T = T01T − T0CVN , i.e., the experimental data cannot determine
1T0CVN or 1T01T separately (see also step 8 of the Monte Carlo simulation procedure). Exper-
imental studies often argue that when T01T −T0CVN → 0, constraint loss has not occurred in the
 Constraint effects 65

Figure 13. 1T0 vs. K0 curves for materials having E/σ0 = 800, v = 0.3 and n = 5, where K0 is the characteristic
toughness value of the 1T SSY configuration. (a) m = 20, (b) m = 16.

fracture toughness values. Clearly, there could remain an absolute 1T0 effect present in each
experimental T0 value as indicated in Figures 11–13, such that T01T − T0CVN → 0. Moreover,
in experimental data there exists uncertainty in the measured K0SSY(1T) values, especially for
smaller data sets, which adds uncertainty in the measured T0CVN and T01T values. When the
CVNs and 1T SE(B)s are tested at the same temperature, the same T0 values are obtained
only if the measured K0 values, scaled to 1T size, are identical. Due to uncertainty in the
measured K0 values, exact agreement between CVN and 1T data would not be expected and
the difference could obscure the presence of constraint loss effects. This last set of experi-
ments becomes even more difficult to perform as the size difference between CVNs and 1Ts
generally requires testing the CVNs at a different (lower) temperature to meet the ASTM
E1921 deformation limit. In summary, experimental values of T0 could thus be expressed in
the form
experiment uncertainty
T0 = T0actual − 1T0constraint loss ± 1T0 , (27)

where the constraint loss effect always lowers the measured T0 value compared to the actual
value but uncertainty in the measured K0SSY(1T) can lower or raise T0 (the last term above).
Figures 11–13 show clearly that a high yield stress delays the onset of constraint loss and
strong strain hardening maintains high levels of crack-tip constraint. These 1T0 values reflect
the combined effects of material strength and hardening levels. Due to the increased strength,
even though the strain hardening is lower, the 1T0 vs. K0 curves for the n = 20 material
extend to higher K0 levels compared to the n = 10 material. The n = 10 material has the
critical balance between strength and strain hardening to generate the largest T0 shift of the
three materials examined. The low strength, strongly hardening material n = 5 shows the
least effect of constraint loss. The 1T0 -values obtained for m = 16 exceed those obtained for
m = 20, reflecting the effect of Weibull modulus on constraint loss prediction. The choice
of m-value between 16 and 20 causes a 2 ∼ 5◦ C difference in the saturated 1T0 -values for
the cases. The use of deformation limits M = 30 or M = 50 affects 1T0 predictions only at
higher K0 levels for the n = 10 and n = 20 materials. The deformation limit has no effect on
the n = 5 material. Although not shown on the figures, simulations for the n = 10 material
66 X. Gao and R.H. Dodds

Figure 14. Standard deviation of 1T0 vs. K0 curves for materials having E/σ0 = 500, v = 0.3 and n = 10.
(a) m = 20, (b) m = 16.

Figure 15. Standard deviation of 1T0 vs. K0 curves for materials having E/σ0 = 300, v = 0.3 and n = 20.
(a) m = 20, (b) m = 16.

with a deformation limit of M = 100 reduced the maximum 1T0 -values to less than 2◦ C. For
completeness, Figures 14–16 show standard deviations of the 1T0 -values computed from the
Monte Carlo simulation of SSY toughness values.

7. Summary and concluding remarks

This paper describes a systematic study of specimen size and constraint loss effects on cleav-
age fracture toughness of ferritic steels in the transition region. The investigation adopts a
probabilistic, local approach for cleavage fracture based on the Weibull stress to construct
a macroscopic, toughness scaling model. For fracture specimens, the constraint level varies
along the crack front, and the Weibull stress density (σ̄w ) reflects this nonuniformity. A non-
dimensional function introduced in this work, g(M), defines constraint loss for a specimen
type and material combination, where M = bσ0 /Javg . Larger values of g(M) imply higher
 Constraint effects 67

Figure 16. Standard deviation of 1T0 vs. K0 curves for materials having E/σ0 = 800, v = 0.3 and n = 5.
(a) m = 20, (b) m = 16.

levels of constraint; g(M) ≡ 1 for plane strain, SSY (with zero T -stress), while for SE(B)
specimens, g(M) decreases as deformation levels increase. For a specified set of material flow
properties and Weibull modulus (m), the g(M) function remains unchanged for all geomet-
rically similar specimens regardless of the absolute size. This proves advantageous in studies
of specimen size/constraint effects. Given the g(M) found from the analysis of one specimen
in a group of geometrically similar specimens, the σw vs. J relationships for all others can be
easily obtained from (12).
The expression for cumulative failure probability in terms of the Weibull stress is modified
to include a threshold Weibull stress term σw−min . The proposed form, which reflects an ap-
proximate treatment of the conditional probability of propagation in the DBT region, parallels
the form adopted recently in ASTM E1921 for use with measured KJ c -values. The three-
parameter expression better models the experimental data at low probabilities and reduces the
overall scatter that also better matches experimental data. Calibration of the third parameter,
σw−min , follows directly from known experimental estimates of the threshold toughness for
a wide class of ferritic steels and introduces no new complications in analyses. Use of the
toughness scaling model including σw−min increases the deformation level (M ≈ 40) at which
the CVN size specimen loses constraint compared to a 1T SE(B) specimen.
The coupled effects of specimen size and constraint loss on the cleavage fracture reference
temperature T0 are studied using Monte Carlo simulation together with the toughness scaling
model modified to include the threshold Weibull stress (with σw−min > 0). For three materials
and three specimen sizes studied, the decrease in T0 due to constraint loss 1T0 , increases
with increasing levels of K0 (the characteristic toughness of the material) and saturates at
higher toughness values. Maximum 1T0 -values of 20–25◦ C are predicted at high K0 values
(relative to the specimen size and material flow properties). Because plane strain, SSY (T =
0) is chosen for the reference constraint condition, the predicted 1T0 -values may be viewed
as upper bounds on expected trends in real test data. Even so, SE(B)s having the n = 5
material show almost no constraint loss relative to plane strain, SSY. The parametric results
also prove useful to estimate the relative 1T0 for specimens of different sizes tested at different
temperatures.
68 X. Gao and R.H. Dodds

Acknowledgment

This investigation was supported by grants principally from the U.S. Nuclear Regulatory Com-
mission, Office of Regulatory Research and from the Naval Surface Warfare Center, Carde-
rock Division. This work was also supported by National Computational Science Alliance
under grant MSS970006N that made available computer resources on the NCSA SGI/CRAY
Origin2000. The authors are indebted to Prof. Kim Wallin (VTT, Finland), Dr. Mark Kirk
(Westinghouse Science & Technology Center), and Dr. Ted Anderson (Structural Reliability
Technology) for many enlightening discussions.

References

American Society for Testing and Materials, Philadelphia (1998). Test Method for the Determination of Reference
Temperature T0 for Ferritic Steels in the Transition Range (ASTM E 1921).
Anderson, T.L. and Dodds, R.H. (1991). Specimen size requirements for fracture toughness testing in the ductile-
brittle transition region. Journal of Testing and Evaluation 19, 123–134.
Anderson, T.L., Stienstra, D. and Dodds, R.H. (1994). A theoretical framework for addressing fracture in the
ductile-to-brittle transition region. Fracture Mechanics: 24th Volume, ASTM STP 1207 (Edited by J.D. Landes,
D.E. McCabe and J. A. Boulet), American Society for Testing and Materials, Philadelphia, 186–214.
Bakker, A. and Koers, R.W.J. (1991). Prediction of cleavage fracture events in the brittle-ductile transition region
of a ferritic steel. Defect Assessment in Components–Fundamentals and Applications, ESIS/EG9 (Edited by
Blauel and Schwalbe), Mechanical Engineering Publications, London, 613–632.
Bass, B.R., McAfee, W.J. and Williams, P.T (1998). Evaluation of constraint methodologies applied to a shallow-
flaw cruciform bend specimen tested under biaxial loading conditions. Proceedings, ASME/JSME Pressure
Vessels and Piping Conference, San Diego, CA, July 26–30.
Beremin, F.M. (1983). A local criterion for cleavage fracture of a nuclear pressure vessel steel. Metallurgical
Transactions 14A, 2277–2287.
Dodds, R.H., Anderson, T. L. and Kirk, M. T. (1991). A framework to correlate a/W ratio effects on elastic-plastic
fracture toughness (Jc ). International Journal of Fracture 48, 1–22.
Gao, X., Faleskog, J. and Shih, C.F. (1999). Analysis of ductile to cleavage transition in part-through cracks using
a cell model incorporating statistics. Fatigue & Fracture of Engineering Materials & Structures 22, 239–250.
Gao, X., Ruggieri and Dodds, R.H. (1998). Calibration of Weibull stress parameters using fracture toughness data.
International Journal of Fracture 92, 175–200.
Hahn, G.T. (1984). The influence of microstructure on brittle fracture toughness. Metallurgical Transactions A
15A, 947–959.
Hughes, T.J. (1980). Generalization of selective integration procedures to anisotropic and nonlinear media.
International Journal for Numerical Methods in Engineering 15, 1413–1418.
Koppenhoefer, K., Gullerud, A., Ruggieri, C., Dodds, R. and Healy, B. (1998). ‘WARP3D: Dynamic nonlinear
analysis of solids using a preconditioned conjugate gradient software architecture’, Structural Research Series
(SRS) 596, UILU-ENG-94-2017, University of Illinois at Urbana–Champaign.
Koppenhoefer, K., and Dodds, R.H. (1997). A three-dimensional numerical investigation of loading rate effects
on cleavage fracture of pre-cracked CVN specimens. Engineering Fracture Mechanics 58, 249–270.
Larsson, S.G. and Carlsson, A.J. (1973). Influence of non-singular stress terms and specimen geometry on small
scale yielding at crack-tips in elastic-plastic materials. Journal of Mechanics and Physics of Solids 21, 447–
473.
Lin, T., Evans, A.G. and Ritchie, R.O. (1986). A statistical model of brittle fracture by transgranular cleavage.
Journal of Mechanics and Physics of Solids 21, 263–277.
McMahon, C.L. and Cohen, M. (1965). Initiation of cleavage in polycrystalline iron. Acta Metallurgica 13, 591–
604.
Minami, F., Brückner-Foit, A., Munz, D. and Trolldenier, B. (1992). Estimation procedure for the Weibull
parameters used in the local approach. International Journal of Fracture 54, 197–210.
Moinereau, D. (1996). Local approach to fracture applied to reactor pressure vessel: synthesis of a cooperative
programme between EF, CEA, Framatome and AEA. Journal de Physique IV 6, 243–257.
 Constraint effects 69

Moran, B., and Shih, C.F. (1987). A general treatment of crack tip contour integrals. International Journal of
Fracture 35, 295–310.
Mudry, F. (1987). A local approach to cleavage fracture. Nuclear Engineering and Design 105, 65–76.
Mudry, F., Rienzo, F.D. and Pineau, A. (1989). Numerical Comparison of Global and Local Fracture Criteria
in Compact Tension and Center-Crack Panel Specimens. Non-linear Fracture Mechanics: II–Elastic-Plastic
Fracture, ASTM STP 995 (Edited by Landes et al.), American Society for Testing and Materials, Philadelphia,
24–39.
Nevalainen, M., and Dodds, R.H. (1995). Numerical investigation of three-dimensional constraint effects on brittle
fracture in SE(B) and C(T) specimens. International Journal of Fracture 74, 131–161.
Press, W.H., Teukolsky, S.A., Vetterling, W.T. and Flannery, B.P. (1992). Numerical Recipes in FORTRAN: The
Art of Scientific Computing, 2nd ed., Cambridge University Press.
Rice, J.R. (1974). Limitations to the small scale yielding approximation for crack tip plasticity. Journal of
Mechanics and Physics of Solids 22, 17–26.
Ruggieri, C., Minami, F. and Toyoda, M. (1995). A statistical approach for fracture of brittle materials based on
the chain-of-bundles model. Journal of Applied Mechanics 62, 320–328.
Ruggieri, C. and Dodds, R.H. (1996). A transferability model for brittle fracture including constraint and ductile
tearing effects: a probabilistic approach. International Journal of Fracture 79, 309–340.
Ruggieri, C., and Dodds, R.H. (1998). Numerical evaluation of probabilistic fracture parameters with WSTRESS.
Engineering Computations 15, 49–73.
Ruggieri, C., Dodds, R.H. and Wallin, K. (1998). Constraint effects on reference temperature T0 for ferritic steels
in the transition region. Engineering Fracture Mechanics 60, 19–36.
Sorem, W.A., Dodds, R.H. and Rolfe, S.T. (1991). Effects of crack depth on elastic-plastic fracture toughness.
International Journal of Fracture 47, 105–126.
Wallin, K. (1984). The scatter in KIc results. Engineering Fracture Mechanics 19, 1085–1093.
Wallin, K., Saario, T., and Torronen, K. (1984). Statistical model for carbide induced brittle fracture in steel. Metal
Science 18, 13–16.
Wallin, K. (1991). Fracture toughness transition curve shape for ferritic structural steels. Fracture of Engineering
Materials and Structures (Edited by S.T. Teoh and K.H. Lee), Elsevier Applied Science, 83–88.
Wallin, K. (1993). Statistical aspects of constraint with emphasis on testing and analysis of laboratory specimens
in the transition region. Constraint Effects in Fracture, ASTM STP 1171 (Edited by Hackett, et al.), American
Society for Testing and Materials, Philadelphia, 264–288.
Wang, Y.Y. (1991). A two-parameter characterization of elastic-plastic crack-tip and applications to cleavage
fracture. PhD thesis, Department of Mechanical Engineering, MIT.
Wiesner, C.S. and Goldthorpe, M.R. (1996). The effect of temperature and specimen geometry on the parameters
of the local approach to cleavage fracture. Journal de Physique IV 6, 295–304.
Williams, M.L. (1957). On the stress distribution at the base of a stationary crack. Journal of Applied Mechanics
24, 109–114.
Xia, L. and Shih, C.F. (1996). Ductile crack growth–III. transition to cleavage fracture incorporating statistics.
Journal of Mechanics and Physics of Solids 44, 603–639.
Xia, L. and Cheng, L. (1997). Transition from ductile tearing to cleavage fracture: A cell model approach.
International Journal of Fracture 87, 289–305.