Anda di halaman 1dari 17

Engineering Fracture Mechanics 67 (2000) 101±117

Transferability of elastic±plastic fracture toughness using the

Weibull stress approach: signi®cance of parameter calibration
Claudio Ruggieri a, Xaosheng Gao b, Robert H. Dodds Jr. b,*
Department of Naval Architecture and Ocean Engineering, University of S~ ao Paulo, S~
ao Paulo, SP 05508-900, Brazil
Department of Civil Engineering, University of Illinois, 2129 Newmark Laboratory, 205 N. Mathews Avenue, Urbana, IL 61801, USA

This study focuses on the Weibull stress approach to assess the e€ects of constraint loss on cleavage fracture
toughness (Jc ). The investigation addresses the signi®cance of the Weibull modulus m (which bears direct connection
with parameter calibration schemes) on the correlation of macroscopic fracture toughness for varying crack con®gu-
rations. In particular, we question the ability of current calibration procedures to provide reliable values for the Weibull
parameters (m, ru ) and present arguments that a non-uniqueness arises in the calibrated values, i.e., many pairs of
(m, ru ) provide equally good correlation of critical Weibull stress values with the distribution of measured fracture
toughness values. Our analyses also show a strong sensitivity of corrected Jc -values (LSY ! SSY constraint correction)
on parameter m for several specimen geometries and di€erent material ¯ow properties; such results provide compelling
support to introduce an alternative, improved calibration procedure based on a toughness scaling model. Results of a
parameter study suggest typical values of m for a wide range of material ¯ow properties and toughness values. An
example illustrates application of the toughness scaling model to construct Weibull stress based, constraint corrections
for measured Jc -values to provide the Weibull modulus, m, for a structural C±Mn steel (BS 4360 Gr 50D). Ó 2000
Elsevier Science Ltd. All rights reserved.

Keywords: Cleavage fracture; Statistical e€ects; Weakest link; Local approach; Weibull stress

1. Introduction

The fundamental importance of cleavage fracture behavior in material failure has stimulated a rapidly
increasing amount of research on micromechanics methodologies to assess the integrity of structural
components subjected to various loading and environmental conditions. Such methodologies, collectively
termed local approaches, describe the cleavage process uncoupled from the macroscopic fracture toughness
( Jc , KJc or CTOD) to quantify the impact of defects in load-bearing materials and in-service structures (e.g.,
o€shore and nuclear structures). Among these research e€orts, the seminal work of Beremin [1] provides the
basis for establishing a relationship between the microregime of fracture and macroscopic crack driving

Corresponding author. Tel.: +1-217-333-3276; fax: +1-217-333-9464.
E-mail address: (R.H. Dodds Jr.).

0013-7944/00/$ - see front matter Ó 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 1 3 - 7 9 4 4 ( 0 0 ) 0 0 0 5 2 - 7
102 C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117

forces (such as the J-integral) by introducing the Weibull stress (rw ) as a probabilistic fracture parameter. A
key feature of the Beremin approach is that rw follows a two-parameter Weibull distribution [2] in terms of
the Weibull modulus, m, and the scale parameter, ru . When implemented in a ®nite element code, the
Beremin model predicts the evolution of the Weibull stress with applied (macroscopic) load to de®ne
conditions leading to (local) material failure.
Recent developments for transferability models of elastic±plastic fracture toughness values rely on the
notion of the Weibull stress as a crack-tip driving force [3±7]. The central feature in these methodologies
adopts the simple axiom that unstable crack propagation (cleavage) occurs at a critical value of the Weibull
stress; under increased remote loading (as measured by J), di€erences in evolution of the Weibull stress
re¯ect the potentially strong variations of near-tip stress ®elds. In this context, the Weibull modulus, m,
plays a major role in the process to correlate e€ects of constraint loss for varying crack con®gurations and
loading modes (tension vs. bending). Consequently, robust schemes to calibrate the Weibull parameters
(m, ru ) become a key element in fracture assessment procedures based upon rw .
A number of studies have demonstrated the potential capability of Weibull stress based approaches to
predict constraint and ductile tearing e€ects on measured distributions of Jc -values and CTOD-values for
structural steels [3,8±10]. The apparent success of these research e€orts has prevented, until recently, a more
active pursuit of improved schemes to calibrate the micromechanical parameters (m, ru ). In spite of the
promise evident in those works, diculties still persist in the calibration of Weibull stress parameters under
speci®c testing conditions (small-scale vs. large-scale yielding conditions). In particular, details of the
analysis procedures (e.g. ®nite strain plasticity, 2-D vs. 3-D, etc.) and the mesh re®nement adopted to
compute the stress ®elds become critical in the calibration process using fracture specimens. Large m-values
accentuate the small di€erences in computed stresses ahead of a blunting crack tip with corresponding
e€ects on the calibrated value. The collective evidence at present from a wide range of researchers reveals an
uncertain picture of the calibration process; some studies obtain similar m-values for di€erent specimen
types while others show surprisingly large di€erences. Reported values of m for common structural and
pressure vessel steels range from 10 to 50.
This study addresses the signi®cance of parameter calibration for the transferability model of elastic±
plastic fracture toughness using the Weibull stress approach. The plan of the paper is as follows: Section 2
reviews brie¯y the probabilistic models (macroscopic and microscopic) for cleavage fracture. Section 3
outlines the conventional procedure to calibrate the Weibull stress parameters, followed by discussions of
the non-uniqueness issue that arises of this widely adopted procedure. Section 4 summarizes the compu-
tational procedures and the ®nite element models for the crack con®gurations considered in this study.
Section 5 describes the Weibull stress based, constraint correction method for fracture toughness values,
which leads to the proposal for a new scheme to calibrate the Weibull parameters (m, ru ). The results of a
parameter study reveal values of m for a wide range of material ¯ow properties and toughness values.
Section 6 provides a speci®c example illustrating the application of the new calibration scheme for a
structural C±Mn steel (BS 4360 Gr 50D) using deep-notch SE(B) specimens with di€erent sizes.

2. Probabilistic treatment of brittle fracture

2.1. Limiting distribution of toughness values ± the Weibull model

Experimental studies consistently reveal large scatter in the measured values of cleavage fracture
toughness for ferritic steels tested in the DBT region (see Refs. [8,9,11±14] for illustrative data and the
experimental results presented in Section 6). Connections between the local (cleavage) fracture process and
extreme value statistics play the key role in describing the scatter of fracture toughness values. A continuous
C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117 103

probability function derived from weakest link statistics conveniently characterizes the distribution of
toughness values in the form [2,13,15]:
Jc ÿ Jmin
F …Jc † ˆ 1 ÿ exp ÿ ; …1a†
J0 ÿ Jmin

which is a three-parameter Weibull distribution with parameters (a; J0 ; Jmin ). Here, a denotes the Weibull
modulus (shape parameter), J0 de®nes the characteristic toughness (scale parameter) and Jmin is the
threshold fracture toughness. Often, the threshold fracture toughness is set equal to zero so that the Weibull
function given by Eq. (1a) assumes its more familiar two-parameter form
F …Jc † ˆ 1 ÿ exp ÿ : …1b†

The above limiting distribution remains applicable for other measures of fracture toughness, such as KJc
or CTOD. A central feature emerging from the probabilistic treatment of brittle fracture based upon the
weakest link model is that, under small scale yielding conditions, the scatter of cleavage fracture toughness
data is characterized by a ˆ 2 for Jc -distributions or a ˆ 4 for KJc -distributions [9,12,13].

2.2. Probabilistic fracture parameter ± the Weibull stress

To extend the previous methodology to multiaxially stressed, 3-D crack con®gurations, research e€orts
have focused on probabilistic models which couple the micromechanical features of the fracture process
(such as the inherent random nature of cleavage fracture) with the inhomogeneous character of the near-tip
stress ®elds. Motivated by the speci®c micromechanism of transgranular cleavage, a number of such models
(most often referred to as local approaches) employ weakest link arguments to describe the failure event.
The overall fracture resistance is thus controlled by the largest fracture-triggering particle that is sampled in
the fracture process zone ahead of crack front. Introduction of the Weibull stress (rw ), a term coined by the
Beremin group [1], provides the basis for generalizing the concept of a probabilistic fracture parameter and
supports the development of procedures that unify toughness measures across di€erent crack con®gura-
tions/loading modes. In the local approach to cleavage fracture, the probability distribution for the fracture
stress of a cracked solid follows a two-parameter Weibull distribution [1,3,5,9],
 Z  m    m 
1 r1 rw
F …rw † ˆ 1 ÿ exp ÿ dV ˆ 1 ÿ exp ÿ ; …2†
V0 V ru ru

where V denotes the volume of the (near-tip) fracture process zone, V0 is a reference volume and r1 is the
maximum principal stress acting on material points inside the fracture process zone de®ned by the loci
r1 P kr0 , with k  2 [6,9]. Parameters m and ru appearing in Eq. (2) denote the Weibull modulus and the
scale parameter of the Weibull distribution. Following Beremin [1], the Weibull stress is de®ned as the stress
 Z 1=m
rw ˆ rm1 dV : …3†
V0 V

In the context of probabilistic fracture mechanics, the Weibull stress, rw , thus emerges as a near-tip
fracture parameter to describe the coupling of remote loading with a micromechanics model which in-
corporates the statistics of microcracks (weakest link philosophy).
104 C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117

3. Calibration of Weibull stress parameters

3.1. Signi®cance of current procedures to calibrate the Weibull stress parameter

Previously developed procedures to calibrate the parameters (m, ru ) (see Refs. [1,3±5,9] for additional
details) employ measured toughness data for cleavage fracture (such as Jc -values) to de®ne corresponding
values of the Weibull stress at fracture, rw;c ; these values form the basis to estimate the Weibull parameters
for the material without making recourse to detailed metallographic measurements. This widely adopted
methodology builds upon an iterative procedure incorporating a ®nite element description of the crack-tip
stress ®elds and measured values of fracture toughness.
The ``standard'' process [1,3±5,9] begins by ®nding the material dependent value for the Weibull mod-
ulus, m. The method seeks to determine the parameters {m, ru } of the probability distribution given by Eq.
(2) that satis®es the identity F …Jc † ˆ F …rw;c †, where F …Jc † is the probability distribution for fracture
toughness given by Eq. (1b) and F …rw;c † is the probability distribution for the Weibull stress given by Eq.
(2). Now, let Ffem …rw † and Fexp …rw † denote the distributions of rw corresponding to the stress state obtained
through a ®nite element analysis and the one obtained through fracture toughness testing, respectively. By
postulating that Ffem …rw † and Fexp …rw † have identical distributions, the calibration process becomes one of
determining a set of parameters {m, ru } which satis®es this condition. The algorithm starts by determining
…rw †fem ˆ H …J ; m† for an initial estimate of m, denoted as m0 , where H …J ; m† denotes the computed
functional relationship between J in the ®nite element model and the Weibull stress for the speci®ed value of
m. The experimental Weibull stress, …rw †exp , corresponding to each experimental toughness value, Jc , is
found by substituting Jc and H …J ; m†. By applying a standard statistical analysis (such as the maximum
likelihood method [15]) to these …rw †exp values, the estimates fm ^ 1 ; …^
ru †1 g are found for the distribution
^ 1 , the process starts a new with the distribution Ffem …rw † computed for m ˆ m
Fexp …rw †. If m0 6ˆ m ^1.
A fundamental assumption underlying the above calibration procedure is that each toughness value (Jc )
of the data set de®nes a corresponding rw;c -value for a ®xed m as illustrated in Fig. 1. Consequently, the
associated identity F …Jc † ˆ F …rw;c † coupled with Eq. (1b) and Eq. (2) permits generalizing a relationship
between J and rw in the form
 m  a
rw J
ˆ : …4†
ru J0

Fig. 1. Schematic rw vs. J trajectories for two arbitrary m-values.

C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117 105

However, a major point of criticism of the calibration process described above is that the identity
F …Jc † ˆ F …rw;c † and Eq. (4) are always satis®ed for any value of the parameter m. The non-uniqueness
results from the fact that two free parameters (m and ru ) are involved in only one equation. Consider the
data set for toughness values (J) shown as solid symbols in Fig. 1 and its equivalent data set of rw -values for
two arbitrary m-values, m ˆ m1 and m ˆ m2 . Now, consider the two-parameter Weibull distribution given
by Eq. (1b) which describes the J-values with ®xed parameters (a, J0 ). Each J-value uniquely de®nes a
failure probability for the cracked con®guration, F …Jk †. Let again H …J ; m† denote the computed functional
relationship between J in the ®nite element model and the Weibull stress for the speci®ed value of m.
Because each J-value is transformed into corresponding Weibull stress values, rmw;k 1
or rmw;k
, through H …J ; m†,
m1 m2
it becomes clear that F …Jk † ˆ F …rw;k † ˆ F …rw;k †. Here, there exists no strict requirements to assume SSY or
LSY conditions so long as Eq. (4) remains valid. In other words, because F …J † and F …rw † are coupled
distributions, a non-uniqueness of the parameter m holds true so long as there exists a Weibull distribution
describing the measured toughness values and the associated Weibull distribution for rw (see Fig. 1).
Consequently, it is not possible to calibrate parameters (m, ru ) using the procedure previously outlined.
Further examination reveals that this iteration scheme to determine parameter m merely re¯ects, at best, the
variations in crack-tip ®elds that emerge in numerical analyses of fracture specimens due to meshing re-
®nement, element behavior, integration schemes, etc.
The non-uniqueness of the Weibull parameters arising from the calibration procedure using only a single
set of measured toughness values remains for other (macroscopic) parameters to describe crack front de-
formation, such as the crack mouth opening displacement (CMOD). Here, both J and CMOD increase
strongly with continued deformation well after attainment of a plastic hinge (which invalidates the mea-
sured load as a parameter). Eqs. (1) and (4) remain valid, for example, if we substitute J ˆ J …CMOD† into

3.2. Non-uniqueness of parameter calibration under SSY conditions

Following the work of Gao, Ruggieri and Dodds (GRD) [16], we now examine the non-uniqueness
calibration of the Weibull stress parameters (m, ru ) using the conventional procedure and crack con®gu-
rations (stationary) under mode I, plane-strain conditions. The Weibull stress for plane-strain conditions is
derived from Eq. (3) by using polar coordinates centered at the (blunted) crack tip as
 Z 2p Z q0 1=m
BJ 2
rw ˆ rm1 q dq dh ; …5†
V0 r20 0 0

where integration extends over the area of the cleavage fracture process zone (plastic zone), B denotes a
reference thickness, rp is the radius of the plastic zone and r1 is the maximum principle stress. Here, q, is the
non-dimensional radius given by r ˆ q…J =r0 † and q0 de®nes a non-dimensional radius for the plastic zone
expressed as rp ˆ q0 …J =r0 †.
At load levels suciently small so that the size of the zone where inelastic e€ects prevail remains small
compared to other geometrical dimensions (e.g., crack length, crack ligament and thickness), the elastic±
plastic near-tip ®elds can be expressed in separable form analagous to the HRR ®eld as

r1 ˆ r0 f …q; T =r0 ; n; m; E=r0 †g…n; h†; …6†

where f and g are non-dimensional functions, E is the Young's modulus of the material, m is the Poisson's
ratio, n is the strain hardening exponent and T is the non-singular elastic stress acting parallel to the crack
106 C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117

After substituting Eq. (6) into Eq. (5) and working out the resulting terms, the Weibull stress may be
written as
BXr0mÿ2 2
rmw ˆ J …7†
Z 2p Z q0
Xˆ …fg† q dq dh; …8†
0 0

which is a function of material ¯ow properties, speci®ed constraint level …T =r0 † and m; but X is independent
of the applied loading, J.
Now, substituting Eq. (7) into Eq. (4) with a ˆ 2, leads to
 2  m  
J rw BXr0mÿ2
ˆ ˆ J 2: …9†
J0 ru V0 rmu
Consequently, J0 is related to the Weibull stress parameters by
V0 rmu
J02 ˆ : …10†
Eq. (10) indicates that for ®xed values m of and ru , the e€ect of material ¯ow properties and/or con-
straint level on toughness levels (T =r0 ) on toughness levels (J0 ) arises only through the corresponding X-
values which are independent of J.
Alternatively, the ru ±m relationship can be obtained by solving Eq. (10) for ru ,
ru ˆ r0 ; …11†
V0 r20
which indicates that is ru an exponential function of 1=m. In this expression, r0 , B and V0 are ®xed con-
stants, X is given by Eq. (8) and J0 is a speci®ed level of material toughness. Unfortunately, Eq. (11) does
not permit solution for m and ru since there is only one equation for two unknowns. For a speci®ed Weibull
exponent a ˆ 2 on J and known J0 from test data in Eq. (1b), many (m, ru ) pairs can be found such that Eq.
(9) holds. For a given value of m, Eq. (11) yields the corresponding ru and vice versa. For all the (m, ru )
pairs which satisfy Eq. (11), the predicted failure probability from Eq. (2) will be the same as the predicted
failure probability from Eq. (1b). Therefore, it is not possible to calibrate a unique (m, ru ) pair using only a
single data set of measured fracture toughness data under SSY (T =r0  constant).
To overcome this inherent limitation of previously developed procedures, the present work proposes a
new scheme to calibrate parameters (m, ru ). Section 5 illustrates the process which uses a scaling meth-
odology to correct measured toughness distributions for di€erent crack con®gurations.

4. Computational procedures and ®nite element models

4.1. Finite element procedures

The 3-D computations reported here are generated using the research code WARP3D [19]: (1) imple-
ments a Mises constitutive models in a ®nite-strain framework, (2) solves the equilibrium equations at each
iteration using a linear pre-conditioned conjugate gradient (LPCG) method implemented within an ele-
ment-by-element (EBE) software architecture, (3) evaluates the J-integral using a convenient domain in-
C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117 107

tegral procedure and (4) analyzes fracture models constructed with 3-D, 8-node tri-linear hexahedral ele-
The ®nite element computations employ a domain integral procedure [20] for numerical evaluation of
the J-integral. A thickness average value for J is computed over domains de®ned outside material having
the highly non-proportional histories of the near-tip ®elds and thus retains a strong domain (path) inde-
pendence. Such J-values agree with estimation schemes based upon eta-factors for deformation plasticity.
They provide a convenient parameter to characterize the average intensity of far ®eld loading on the crack

4.2. Constitutive models

The elastic±plastic material employed in the analyses follows a J2 ¯ow theory with conventional Mises
plasticity. The uniaxial true stress-logarithmic strain curve obeys a simple power-law hardening model,
 r  r
ˆ ;  6 0 ; ˆ ;  > 0 ; …12†
0 r0 0 r0
where r0 and 0 are the reference (yield) stress and strain, and n is the strain hardening exponent.
Section 5 describes numerical solutions for the SSY boundary-layer model with T ˆ 0 and standard
0:5…T †, 1…T † and 2…T † SE(B) specimens with a=W ˆ 0:5. These ®nite element analyses consider material
¯ow properties covering most structural and pressure vessel steels: n ˆ 5 …E=r0 ˆ 800†, 10 …E=r0 ˆ 500†
and 20 …E=r0 ˆ 300† with E ˆ 206 GPa and m ˆ 0:3; these ranges of properties also re¯ect the upward
trend in yield stress with the decrease in strain hardening exponent characteristic of ferritic steels. For the
SE(B) specimens used in the fracture testing described in Section 6, we generate numerical solutions that
utilize a piecewise linear approximation to the measured tensile response for the material at the test tem-
perature, T ˆ ÿ120°C given in Ref. [10].

4.3. Finite element models of SE(B) specimens and SSY model

3-D ®nite element analyses are described for plane-side deep notch …a=W ˆ 0:5† SE(B) specimens. Here,
a denotes the crack length and W the specimen width. The analysis matrix includes a conventional geometry
…W =B ˆ 2† with varying thickness B ˆ 12:5 mm ‰0:5…T †Š, B ˆ 25 mm ‰1…T †Š and B ˆ 50 mm ‰2…T †Š.
Fracture toughness tests at di€erent lower transition temperatures for a structural steel (BS 4360 Grade
50D) [10] were performed on the 3-point SE(B) specimens with size 0:5…T † and 2…T †. Fig. 2(a) shows the
geometry and dimensions of the SE(B) specimens employed in the analyses.
Fig. 2(b) shows a typical ®nite element model constructed for analyses of the 1…T † SE(B) specimen. All
other crack models have very similar features. A conventional mesh con®guration having a focused ring of
elements surrounding the crack front is used with a small key-hole at the crack tip; the radius of the key-
hole, q0 , is 10 lm (0.01 mm). Symmetry conditions enable analyses using one-quarter of the 3-D models
with appropriate constraints imposed on the symmetry planes. The mesh has 14 variable thickness layers
de®ned over the half-thickness (B=2); the thickest layer is de®ned at Z ˆ 0 with thinner layers de®ned near
the free surface (Z ˆ B=2) to accommodate strong Z variations in the stress distribution. The quarter-
symmetric, 3-D models for the SE(B) specimens typically have 18,500 nodes and 16,000 elements. These
®nite element models are loaded by displacement increments imposed on the center-plane nodes for the
outermost two layers of elements. A typical solution to load the specimen to J ˆ 100 kJ/m2 uses 100 load
increments and requires 10 CPU hours on a CRAY J-90 computer.
Numerical solutions for a stationary crack under well-de®ned SSY conditions (with the T -stress term [17]
set to zero, i.e., T ˆ 0) are generated by imposing displacements of the elastic, mode I singular ®eld on the
outer circular boundary (r ˆ R) which encloses the crack
108 C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117

Fig. 2. SE(B) specimens with a=W ˆ 0:5 employed in the analyses (all units in mm).

1‡m R h
u…R; h† ˆ KI cos …3 ÿ 4m ÿ cos h†; …13†
E 2p 2

1‡m R h
v…R; h† ˆ KI sin …3 ÿ 4m ÿ cos h†: …14†
E 2p 2

Here, r and h are polar coordinates centered at the crack tip with h ˆ 0 corresponding to a line ahead of
the crack, KI ˆ ‰EJ =…1 ÿ m2 †Š1=2 , E is the Young's modulus, m is the Poisson's ratio.
Evaluation of the Weibull stress requires integration over the process zone, including the region as
r ! 0. The SSY and all other crack models previously described have a small initial root radius at the crack
front (blunt tip) which provides two numerical bene®ts: (1) it accelerates convergence of the ®nite-strain
plasticity algorithms during the initial stage of blunting, and (2) it minimizes numerical problems during
computation of the Weibull stress over material incident on the crack tip. To maintain consistency with the
®nite element models for the SE(B) specimens used to construct the JLSY ! JSSY corrections, the SSY model
has the same mesh con®guration at the crack tip as the SE(B) models. The SSY model has one thickness
layer of 2065 8-node, 3-D elements with plane-strain constraints imposed on all nodes.

5. Constraint corrections of toughness values: implications for parameter calibration

5.1. Constraint corrections based on the Weibull stress

This section describes the results of detailed numerical analyses employing the MBL model and 3-D
®nite element analyses for the standard SE(B) specimens with size 0:5…T †, 1…T † and 2…T † to assess e€ects of
constraint loss on individual fracture toughness values for di€erent material properties and specimen
con®gurations. The objective is to demonstrate the strong e€ect of parameter m (Weibull modulus) on
C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117 109

Fig. 3. Toughness scaling model used to construct JLSY ! JSSY corrections, where the SSY model has the same thickness as the SE(B)
specimen (LSY).

JLSY ! JSSY corrections. These results serve to introduce the framework to construct an alternative pa-
rameter calibration scheme presented in the next section.
Fig. 3 illustrates the procedure to assess the e€ects of constraint loss on fracture toughness needed
to construct JLSY vs. JSSY trajectories. Very detailed, non-linear 3-D ®nite element analyses provide the
functional relationship between the Weibull stress (rw ) and the applied loading (J ) for a speci®ed value of
the Weibull modulus, m. Based on the argument of the Weibull stress as the crack driving force, the scaling
model requires the attainment of equal values for rw to trigger cleavage fracture across di€erent specimen
geometries even though J-values may vary widely due to constraint loss. Fig. 3 shows curves of rw vs. J for
a standard fracture specimen (the present work employs only deep notch SE(B) specimens) and for a plane-
strain, SSY reference solution (T =r0 ˆ 0) with the same thickness of the fracture specimen. Such curves are
constructed for a ®xed, representative value of the Weibull modulus, m, for each set of ¯ow properties (the
normalizing volume for the Weibull stress, V0 , is assigned the value of 1 mm3 for convenience). JLSY values
computed from the domain integral procedures in the ®nite element analyses of the SE(B) specimens
represent thickness average values, which are consistent with experimental (average) values de®ned using a
plastic eta-factor and the measured areas under the load vs. load line displacement curve. For the plane-
strain SSY model, JSSY becomes simply the J value at the crack tip. Given the JLSY value for the fracture
specimen, the lines shown on Fig. 3 readily illustrate the technique used to determine JSSY . Toughness
values are often normalized by br0 to provide a set of curves (JSSY =br0 vs. JLSY =br0 † applicable for geo-
metrically scaled specimens, i.e., all SE(B)s with a=W ˆ 0:5, W ˆ B, S ˆ 4W .
Figs. 4±6 provides the constraint corrections (LSY ! SSY) for the 0:5…T †, 1…T † and 2…T † SE(B) speci-
mens with di€erent material properties (n ˆ 5 with E=r0 ˆ 800; n ˆ 10 with E=r0 ˆ 500; n ˆ 20 with
E=r0 ˆ 300) and for varying Weibull moduli, m. The present computations consider values of m ˆ 10, 15,
20, 25 and 30 to assess the sensitivity of constraint corrections on the speci®ed Weibull modulus. These m-
values are consistent with previously reported values for structural steels. Each curve provides pairs of J-
values, JLSY in the SE(B) specimen and JSSY in SSY, that produce the same rw (and thus the same failure
probability). Reference lines are shown which de®ne a constant ratio of ``constraint loss'', e.g., Javg ˆ 1:2J0
which implies that the SE(B) average J must be 20% greater than the SSY value to generate the same
Weibull stress. For each value of the Weibull modulus, the SE(B) and SSY curves agree very well early in
the loading history while the SE(B) specimen maintains near SSY conditions across the crack front (recall
that computation of rw in the SE(B) specimens considers the entire crack front). Once near-front stresses
deviate from the (plane-strain) SSY levels, the rw curves for the SE(B) specimens fail to increase at the
same rate with further loading. These results illustrate clearly the gradual nature of constraint loss in the
110 C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117

Fig. 4. JLSY ! JSSY correction using a scaling methodology based upon the Weibull stress with varying Weibull moduli for plane-sided
0:5…T † SE(B) specimens.

Fig. 5. JLSY ! JSSY correction using a scaling methodology based upon the Weibull stress with varying Weibull moduli for plane-sided
1…T † SE(B) specimens.
C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117 111

Fig. 6. JLSY ! JSSY correction using a scaling methodology based upon the Weibull stress with varying Weibull moduli for plane-sided
2…T † SE(B) specimens.

deep-notch SE(B) specimens, especially for moderate to low hardening materials. The Weibull modulus
does have an appreciable a€ect on predictions of constraint loss; increasing m values indicate a higher load
level at the onset of constraint loss and a reduced rate of constraint loss under further loading. The larger m
values, in e€ect, assign a greater weight factor to stresses at locations very near the crack front. The bending
®eld, which impinges on the crack front, a€ects the smaller m curves more readily.

5.2. Parameter calibration using JLSY ! JSSY trajectories

The previous results exhibit the essential features of Weibull stress based, constraint corrections for some
key specimen geometries. Each SSY toughness value corrected from its corresponding LSY value re¯ects
both the e€ects of stressed near-tip volume and of the strong changes in the character of the near-tip stress
®elds due to constraint loss. Further, in the context of probabilistic fracture mechanics, each pair (JSSY ,
JLSY ) on a given m-curve de®nes equal failure probabilities for cleavage fracture.
Motivated by the above observations, we propose an alternative parameter calibration scheme that uses
the scaling methodology previously outlined to correct measured toughness distributions for di€erent crack
con®gurations. The procedure extends previous work by GRD [16] to calibrate parameter (m, ru ) using
high constraint (SSY) and low constraint (LSY) fracture toughness data measured at the same temperature
and loading rate. Here, we use the quantitative e€ect of constraint loss for di€erent specimen thickness and
crack ligament to calibrate m while GRD provide similar procedure employing same size specimen with
di€erent a=W ratios. Because each measured JLSY value is corrected to its equivalent JSSY value, the sta-
tistical (Weibull) distribution of JLSY values is also corrected to an equivalent statistical (Weibull) distri-
bution of JSSY values. Consequently, our scheme de®nes the calibrated value of m for the material as the
value that corrects the characteristic toughness J0LSY (i.e., the scale parameter of Eq. (1b), see Section 2.1) to
112 C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117

its equivalent J0SSY . Because parameter m is assumed independent of specimen geometry (as long as the
framework upon which the Weibull stress is based remains valid), the scheme remains equally applicable
when two sets of fracture toughness data from di€erent crack con®gurations, but with sucient di€erences
in the evolution of rw vs. J, are used (e.g., 1…T † SE(B) specimens with a=W ˆ 0:5 and a=W ˆ 0:15).
The following steps describe the key procedures in the proposed calibration scheme which slightly
simpli®es the methodology described by GRD [16]. Section 6 illustrates application of the process for a
ferritic structural steel (BS 4360 Gr 50D).
Step 1
Test two sets of specimens with di€erent crack con®gurations (A and B) in the ductile-to-brittle tran-
sition region to generate two distributions of fracture toughness data. Select the specimen geometries and
the common test temperature to insure di€erent evolutions of constraint levels for the two con®gurations.
No ductile tearing should develop prior to cleavage fracture in either sets of tests. Several alternatives to
obtain the two sets of toughness values at the same test temperature include
(1.1) Test high constraint deep-notch SE(B)s or C…T †s as con®guration A. To insure SSY conditions at
fracture, set the specimen size so that Jc 6 brys =M for each specimen, with M conservatively set to 75±100
[21]. For con®guration B, use similar size SE(B) specimens but with shallow-notches, i.e., a=W ˆ 0:15.
These will undergo signi®cant constraint loss when the deep-notch values just satisfy the suggested de-
formation limit.
(1.2) Test large (A) and small (B) deep-notch SE(B) specimens having the same a=W P 0:5 ratio. Set the
large specimen size to provide high constraint conditions. Set the small specimen size to insure sucient loss
of constraint at fracture for most specimens.
Step 2
Perform detailed, 3-D ®nite element analyses for the tested specimen geometries. The mesh re®nements
must be sucient to insure converged rw vs. J histories for the expected range of m-values and loading
Step 3
(3.1) Assume an m-value. Compute the rw vs. J history for con®gurations A and B to construct the
toughness scaling model relative to both con®gurations.
(3.2) Constraint correct J0B to its equivalent J0;m
(i.e., the corrected value of the scale parameter for the
assumed m-value). De®ne the error of toughness scaling as R…m† ˆ …J0;m ÿ J0A †=J0A . If R…m† 6ˆ 0, repeat 3.1±
3.2 for additional m-values.
(3.3) Plot R…m† vs. m. The calibrated Weibull modulus makes R…m† ˆ 0 within a small tolerance.
Step 4
Compute the ru -value. After m is determined, ru is obtained easily from the rw vs. J history for the
calibrated m-value. ru equals the Weibull stress value at J ˆ J0A or J ˆ J0B in the corresponding con®gu-

5.3. E€ects of strain hardening and toughness on the Weibull modulus, m

This section examines the e€ect of strain hardening on the calibrated Weibull modulus, m, using the
procedure outlined in Section 5.2. We consider a parameter study adopting a 0:5…T † SE(B) specimen with
material properties: n ˆ 5 with E=r0 ˆ 800; n ˆ 10 with E=r0 ˆ 500 and n ˆ 20 with E=r0 ˆ 300. These
are typical for ferritic steels. Moreover, because measured toughness distributions are not available, we
adopt J0SSY ˆ 60 kJ/m2 for materials with n ˆ 5 and 10, and J0SSY ˆ 90 kJ/m2 for the material with n ˆ 20.
C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117 113

Fig. 7. Dependence of Weibull modulus, m, on materialÕs strain hardening exponent and characteristic toughness.

In the present study, we set the toughness change for the material due to constraint loss as J0LSY =J0SSY ˆ 1:5,
1.75 and 2 (i.e., J0LSY ˆ 90, 105 and 120 kJ/m2 for materials with n ˆ 5 and 10, and J0LSY ˆ 135, 158 and 180
kJ/m2 for the material with n ˆ 20). The calibrated value of m for the material becomes the value that
corrects each characteristic toughness J0LSY to its equivalent J0SSY (speci®ed).
Fig. 7 shows the dependence of parameter m on strain hardening exponent, n. The upward trend of
increased m-values with increased strain hardening exponents (decreased hardening) is evident. For a given
material, the Weibull modulus m decreases with increased toughness changes as measured by the ratio
J0LSY =J0SSY ; such e€ect arises from the more severe constraint loss in the specimen for higher levels of
toughness changes J0LSY =J0SSY which requires smaller m-values to predict the more pronounced J0 shift
between LSY and SSY conditions. This brief parameter study illustrates the expected value for m for a
typical range of material ¯ow properties and toughness levels. For the material properties considered here,
the calibrated m-values lie in the range of 10±25.

6. Calibration of Weibull stress parameters for a structural steel

6.1. Fracture toughness testing using SE(B) specimens

Wiesner and Goldthorpe [10] provide cleavage fracture toughness values at di€erent lower transition
temperatures obtained from testing of 3-point SE(B) specimens (plane-sided) with ®xed crack length to
width ratio, a=W ˆ 0:5, and varying specimen thickness, B. The specimens have B ˆ 12:5 and 50 mm with
width W ˆ 2B and span S ˆ 8B (refer to Fig. 2). The material is a ferritic C±Mn steel (BS 4360 Grade 50D)
with high strain hardening (rut =rys ˆ 1:53). Mechanical tensile test data for this material is presented in
Ref. [10].
Fig. 8 provides a Weibull diagram of the measured toughness values for the test temperature
T ˆ ÿ120°C. The open symbols in the plots indicate the experimental fracture toughness data for the SE(B)
specimens. Values of cumulative probability, F , are obtained by ordering the Jc -values and using F ˆ
…i ÿ 0:5†=N , where i denotes the rank number and N de®nes the total number of experimental toughness
values. The straight lines indicate the two-parameter Weibull distribution obtained by a maximum like-
lihood analysis of the data set. The maximum likelihood estimates (^ a, J^0 ) for each data set are: (1.5, 44.3)
for the 0:5…T † specimen and (1.4, 29.3) for the 2…T † specimen.
114 C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117

Fig. 8. Weibull plots of toughness values for two sets of SE(B) specimens at T ˆ ÿ120°C.

6.2. Calibration of Weibull stress parameters

The procedure used here to calibrate the Weibull stress parameters for the C±Mn steel follows the
proposed scheme outlined in Section 5.2. In the present application, we calibrate parameters (m, ru ) by
scaling the measured toughness distribution for the 0:5…T † SE(B) specimen to an equivalent toughness
distribution for the 2…T † SE(B) specimen. Very detailed ®nite element computations of these specimens
enable construction of the J0:5…T † ! J2…T † correction shown in Fig. 9. The calibration process simply becomes
0:5…T † 2…T †
one of determining an m-value that corrects J0 to its equivalent J0 .
A key feature of this methodology also lies in the choice of parameter a describing the scatter of the
toughness distribution. We propose to adopt the ®xed theoretical value a ˆ 2 to describe the distribution of
Jc -values, which makes contact with the probabilistic treatment of fracture under SSY conditions based
on weakest link statistics described in Section 2.1. Such an approach is also consistent with the so-called
``master curve'' approach [22] where it is recognized that measured values of a depend strongly on the
sample size. An important question to resolve with this approach concerns the variation of the charac-
teristic toughness …J0 †, i.e., the scale parameter of the two-parameter Weibull distribution, when parameter
a is ®xed. Fortunately, statistical analyses of the experimentally measured Jc -values based upon the max-

Fig. 9. J0:5…T † ! J2…T † correction using a scaling methodology based upon the Weibull stress with varying Weibull moduli for the tested
structural steel BS4360 Gr50. The lines on the plot indicates the calibration process to determine the Weibull modulus m.
C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117 115

Fig. 10. Predicted Weibull distribution for the 2…T † SE(B) specimen at T ˆ ÿ120°C from the toughness data of the 0:5…T † SE(B)
specimen using the toughness scaling model with m ˆ 15.

imum likelihood method with a ®xed value of a reveal a weak dependence of J0 on a. For the SE(B)
0:5…T †
specimens with B ˆ 12:5 mm at T ˆ ÿ120°C, the maximum likelihood estimate of J0 with a ˆ 2 is 49.3
kJ/m2 and di€ers by 11% from the previous value of 44.3 kJ/m2 (recall the large deviations of the Weibull
modulus from a ˆ 2 for this data set). Similarly, for the SE(B) specimens with B ˆ 50 mm at T ˆ ÿ120°C,
2…T †
the maximum likelihood estimate of J0 with a ˆ 2 is 34.0 kJ/m2 and di€ers by 12% from the previous
value of 29.3 kJ/m2 .
With the introduction of a ®xed value a ˆ 2, our calibration procedure yields m  15 for the steel tested
at T ˆ ÿ120°C. Fig. 9 illustrates the graphical procedure to determine parameter m. Fig. 10 shows the
Weibull distribution for the 2…T † SE(B) specimen as predicted from the toughness data for the 0:5…T † SE(B)
specimen using the toughness scaling methodology with the calibrated m-value. To generate this plot, we (1)
compute the rw vs. J relationships for both 0:5…T † and 2…T † specimens using the calibrated value of Weibull
modulus (m ˆ 15); (2) scale the toughness data for 0:5…T † specimen to its 2…T † equivalent value based on
equal rw -values; (3) plot the corrected toughness data in a Weibull diagram. While the lower and upper tails
of the predicted curve (solid line) lie somewhat higher than the 2…T † experimental data (symbols), generally
good agreement exists between the predicted distribution and the experimental data.
In this study, we adopt the original two-parameter Weibull stress model, (2), to describe the cumulative
failure probability and use experimental data obtained from two sets of deep notch SE(B) specimens with
di€erent absolute sizes to calibrate the Weibull modulus, m. Alternatively, specimens with the same size but
having di€erent a=W ratios or specimens subjected to di€erent loading modes (bending vs. tension) can be
used to calibrate m according to the procedures described in Section 5.2 [16,23]. For some materials, it may
be possible to achieve suciently di€erent constraint conditions to enable calibration by testing the same
type/size of specimens at two di€erent temperatures, i.e., one temperature near the lower-shelf to generate
toughness values under SSY and another temperature well-up in transition to generate toughness values
under LSY conditions. Two issues arise with this approach: (1) the specimens tested under LSY conditions
should not experience ductile tearing prior to fracture and (2) the calibration process as described previ-
ously must assume that the Weibull modulus, m, remains identical over the temperature range. The tem-
perature (and loading rate) dependence of the Weibull modulus, m, remain under active investigation. A
slight variation of this approach employs the ``master curve'' [22] to determine the material's reference
temperature, T0 , using the low temperature, SSY toughness values. The master curve enables estimation of
the characteristic toughness value (K0 or J0 ) at the temperature of the LSY tests. This estimated value of K0
(or J0 ) and the measured LSY toughness distribution provide the toughness data (at high and low
constraint levels) needed for the calibration of Weibull parameters. This modi®cation eliminates the
116 C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117

requirement to assume m remains unchanged with temperature (only the material ¯ow properties at the
higher temperature are need to support the ®nite element analysis).
Recently, Gao et al. [16] modify the expression for the failure probability to introduce a threshold value
…rwÿmin † for cleavage fracture which re¯ects an approximate model for the conditional failure probability.
The value of …rwÿmin † ties directly with Kmin (minimum toughness value for ferritic steels) of ASTM E-1921
[22]. The introduction of …rwÿmin † in the expression for failure probability brings numerical predictions of
the scatter in fracture toughness data into better agreement with experiments (especially the tails of the
toughness distribution) [16,23].

7. Concluding remarks

Our extensive numerical analyses to assess constraint loss due to absolute size e€ects (LSY ! SSY) for
standard, deep notch SE(B) specimens demonstrate the strong in¯uence of the Weibull stress parameter m
(Weibull modulus) on JLSY ! JSSY corrections. These analyses provide support to introduce alternative,
improved procedure based on a toughness scaling model to calibrate the Weibull stress parameters, (m, ru ).
Results of a parameter study reveal values of m for a wide range of material ¯ow properties and toughness
values. The toughness scaling model enables construction of Weibull stress based constraint corrections for
experimentally measured Jc -values to provide the Weibull parameters for a structural C±Mn steel (BS 4360
Grade 50D). Further work is in progress to validate the proposed procedure as a more general and robust
scheme to calibrate the Weibull stress parameters for several crack con®gurations and its implication on
predictions of toughness distributions.


This investigation was supported by grants principally from the Scienti®c Foundation of the State of S~ao
Paulo (FAPESP) under Grant 95/9898-0. XG and RHD are funded by US Nuclear Regulatory Com-
mission, Oce of Regulatory Research and from the Naval Surface Warfare Center, Carderock Division.
Access to the Cray J-90 at the High Performance Computing Center (LCCA) of the University of S~ao
Paulo is gratefully acknowledged. The authors are indebted to Dr. Christoph S. Wiesner and Dr. Martin R.
Goldthorpe (TWI) for the useful discussions.


[1] Beremin FM. A local criterion for cleavage fracture of a nuclear pressure vessel steel. Metallur Trans 1983;14A:2277±87.
[2] Weibull W. The phenomenon of rupture in solids. Ingeniors Vetenskaps Akademien Handlingar 1939;153:55.
[3] Ruggieri C, Dodds RH. A transferability model for brittle fracture including constraint and ductile tearing e€ects: a probabilistic
approach. Int J Fract 1996;79:309±40.
[4] Ruggieri C, Dodds RH. Probabilistic modeling of brittle fracture including 3-D e€ects on constraint loss and ductile tearing.
In: Proceedings of the First European Mechanics of Materials Conference on Local Approach to Fracture (1986±1996),
Fontainbleau, France, September 9±11, 1996.
[5] Ruggieri C, Dodds RH. WSTRESS Release 1.0: Numerical computation of probabilistic fracture parameters for 3-D cracked
solids, BT-PNV-30 (Technical Report), EPUSP, University of S~ ao Paulo, 1997.
[6] Ruggieri C, Dodds RH. Numerical computation of probabilistic fracture parameters using WSTRESS. Engng Computat
[7] Ruggieri C, Dodds RH, Wallin K. Constraints e€ects on reference temperature T0 for ferritic steels in the transition region.
Engng Fract Mech 1998;60:19±36.
C. Ruggieri et al. / Engineering Fracture Mechanics 67 (2000) 101±117 117

[8] Bakker A, Koers RWJ. Prediction of cleavage fracture events in the brittle-ductile transition region of a ferritic steel. In: J.G.
Blauel, K.H. Schwalbe, editors. Defect assessment in components ± fundamentals and applications. ESIS/EG9, Mechanical
Engineering Publications, London, 1991. p. 613±32.
[9] Minami F, Br uckner-Foit A, Munz D, Trolldenier B. Estimation procedure for the Weibull parameters used in the local approach.
Int J Fract 1992;54:197±210.
[10] Wiesner CS, Goldthorpe MR. The e€ect of temperature and specimen geometry on the parameters of the local approach to
cleavage fracture. International Conference on Local Approach to Fracture (MECAMAT 96), Fontainebleau, France, 1996, 9.
[11] Ruggieri C, Minami F, Toyoda M. A statistical approach for fracture of brittle materials based on the chain-of-bundles model.
J Appl Mech 1995;62:320±8.
[12] Wallin K. The scatter in KIc results. Engng Fract Mech 1984;19:1085±93.
[13] Wallin K. Statistical aspects of constraint with emphasis on testing and analysis of laboratory specimens in the transition region.
In: E.M. Hackett, K.H. Schwalbe, R.H. Dodds, editors. Constraint e€ects in fracture, ASTM STP 1171, American Society for
Testing and Materials, Philadelphia, 1993. p. 264±88.
[14] Sorem WA, Dodds RH, Rolfe ST. E€ects of crack depth on elastic plastic fracture toughness. Int J Fract 1991;47:105±26.
[15] Mann NR, Schafer RE, Singpurwalla ND. Methods for statistical analysis of reliability and life data. New York: Wiley; 1974.
[16] Gao X, Ruggieri C, Dodds RH. Calibration of Weibull stress parameters using fracture toughness data. Int J Fract 1998;92:
[17] Williams ML. On the stress distribution at the base of a stationary crack. J Appl Mech 1957;24:109±14.
[18] Larsson SG, Carlsson AJ. In¯uence of non-singular stress terms and specimen geometry on small scale yielding at crack-tips in
elastic±plastic materials. J Mech Phys Solids 1973;21:447±73.
[19] Koppenhoefer K, Gullerud A, Ruggieri C, Dodds R, Healy B. 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, Urbana-Champaign, 1994.
[20] Moran B, Shih CF. A general treatment of crack tip contour integrals. Int J Fract 1987;35:295±310.
[21] Nevalainen M, Dodds RH. Numerical investigation of 3-D constraint e€ects on brittle fracture in SE(B) and C(T) specimens. Int
J Fract 1995;74:131±61.
[22] Test method for the determination of reference temperature, T0 , for ferritic steels in the transition range (ASTM E-1921),
American Society for Testing and Materials, Philadelphia, 1998.
[23] Gao X, Dodds RH, Tregoning RL, Joyce JA, Link RE. A Weibull stress model to predict cleavage fracture in plates containing
surface cracks. Fatig Fract Engng Mater Struct 1999;22:481±94.