Available online at www.sciencedirect.

com

ScienceDirect
Available
Available online
online at at www.sciencedirect.com
www.sciencedirect.com
Structural Integrity Procedia 00 (2016) 000–000

ScienceDirect
www.elsevier.com/locate/procedia
ScienceDirect
Procedia Structural
Structural IntegrityIntegrity
Procedia2 (2016) 3727–3734
00 (2016) 000–000
www.elsevier.com/locate/procedia

21st European Conference on Fracture, ECF21, 20-24 June 2016, Catania, Italy

Evaluation and Identifying the Ductile Fracture Area of X70 Steel

XV Portuguese Conference on Fracture, PCF 2016, 10-12 February 2016, Paço de Arcos, Portugal
from DWTT Broken Specimens.
Thermo-mechanical modeling of a high pressure turbine blade of an
airplane gas turbine engine
Pavel Skalny*
P. Brandãoa, V. Infanteb, A.M. Deusc*
VSB-Technical University Ostrava, 17. listopadu 15/2172, 708 33 Ostrava - Poruba
a
Department of Mechanical Engineering, Instituto Superior Técnico,
Czech RepublicUniversidade de Lisboa, Av. Rovisco Pais, 1, 1049-001 Lisboa,
b
b Portugal
Second affiliation, Address, City and Postcode, Country
b
IDMEC, Department of Mechanical Engineering, Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais, 1, 1049-001 Lisboa,
Portugal
c
CeFEMA, Department of Mechanical Engineering, Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais, 1, 1049-001 Lisboa,
Portugal
Abstract

Fracture
Abstractsurfaces of X70 steel DWTT broken samples are analyzed using statistical methods and fractal concepts.
Besides of fractal geometry concept the new approach based on the normal vectors analysis is presented. The
fracture
Duringsurface is coveredmodern
their operation, by a triangle
aircraft net. Forcomponents
engine every triangle are the normaltovector
subjected is determined.
increasingly demanding Fractal dimension
operating is
conditions,
significantly correlated
especially the to the
high pressure angular
turbine (HPT)deviation of neighboring
blades. Such normal
conditions cause vectors.
these parts to The mean
undergo value types
different of deviation angles
of time-dependent
anddegradation, one of which
fractal dimension is islower
creep. for
A model usingthan
ductile the finite elementfracture
for brittle method (FEM)
areas.was Thedeveloped,
k-meansin cluster
order to be able to predict
analysis and
multivariate probability distribution are applied to determine the local characteristics of the fracture surface.aviation
the creep behaviour of HPT blades. Flight data records (FDR) for a specific aircraft, provided by a commercial The
company,
fracture weremap
surface usedoftoductile
obtain thermal
fractureand mechanical
probability anddata for three
k-means different
clusters flightcorresponds
highly cycles. In order
to theto real
create the 3D model
placement of
needed for the FEM analysis, a HPT blade scrap was scanned, and its chemical composition and material properties were
ductile and brittle fracture area on fracture surface of the broken sample. Newly presented statistical methods
obtained. The data that was gathered was fed into the FEM model and different simulations were run, first with a simplified 3D
applied on theblock
rectangular fracture
shape,surface
in orderoftoDWTT broken the
better establish samples
model,areanduseful tools
then with thefor
realobjective evaluation
3D mesh obtained fromofthe
ductile
bladefracture
scrap. The
percentage.
overall expected behaviour in terms of displacement was observed, in particular at the trailing edge of the blade. Therefore such a
model can
Copyright be useful
© 2016 in the Published
The Authors. goal of predicting
by Elsevierturbine blade
B.V. This is anlife,
opengiven a set
access of FDR
article underdata.
the CC BY-NC-ND license
© 2016 The Authors. Published by Elsevier B.V.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review
© 2016 The
Peer-review under responsibility
Authors.
under Published
responsibility ofby
of the theElsevier
Scientific
Scientific Committee
B.V.
Committee of ECF21.
of ECF21.
Peer-review under responsibility of the Scientific Committee of PCF 2016.
Keywords: Fracture, DWTT specimen, cluster analysis, probability. 596 918 507
Keywords: High Pressure Turbine Blade; Creep; Finite Element Method; 3D Model; Simulation.

* Pavel Skalny. Tel.: +420-597-324-181; fax: +420-596-918-507.

E-mail address: pavel.skalny@vsb.cz

2452-3216 © 2016 The Authors. Published by Elsevier B.V.

Peer-review underauthor.
* Corresponding responsibility
Tel.: +351of218419991.
the Scientific Committee of ECF21.
E-mail address: amd@tecnico.ulisboa.pt

2452-3216 © 2016 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the Scientific Committee of PCF 2016.
Copyright © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer review under responsibility of the Scientific Committee of ECF21.
10.1016/j.prostr.2016.06.463
3728 Pavel Skalny / Procedia Structural Integrity 2 (2016) 3727–3734
2 Pavel Skalny/ Structural Integrity Procedia 00 (2016) 000–000

1. Introduction

The correct evaluation of fracture surfaces is very important in selecting a suitable construction material in
technical practice. For example, the knowledge of fracture toughness is crucial for the construction of pipelines to
transport oil.
Apart from the Charpy V-notch test see Yang et al. (2008) the Drop Weight Tear Test is often used for evaluation
of the fracture surface, see Fang et al. (2014), Strnadel et al. (2013), Seifert (1984). The tested specimen is broken by
the stroke. After realizing the tested surface is evaluated by the expert, who determines the ratio between ductile and
brittle fracture. Although the analysis by specialist has many advantages, it is also loaded with some degree of
human error. In some cases expert opinions may vary significantly, see Horsley (2003). For example the fracture
surface of steel used for the pipeline display higher parameters of abnormal fracture see and Yang et al. (2008)
Hwang et al (2004). This factor complicates the visual evaluation of the fracture surfaces.
The alternative to an expert evaluation is a realization of a 3D scan and usage of the computer analysis. The 3D
scan is elaborated with two methods- the fractal geometry concept and new approach based on the normal vector
analysis.
The fractal geometry is often used to evaluate and describe the fracture, e. g. Mengija et al. (2015), Balakin et al.
(2000), Mandelbrot et al (1984). It was shown that the value of the fractal dimension is closely related to the ductile
and brittle fracture, see Strnadel et al. (2013) Stach et al. (2001), Weiss (2001) Jiang et al. (1994), Furthermore the
fractal dimension is related to the mechanical properties of the material and roughness characteristics, see Strnadel et
al. (2013), Chang et al (2011), Ray and Mandal (1992). One of disadvantages of the fractal geometry concept is the
fact that the fractal dimension is a “global” characteristic. It could be calculated only on a sufficiently large number
of points (or area). So that it does not characterize the surface in the certain place. Furthermore the estimation of
fractal dimension can be inaccurate, see Schmittbuhl et al. (1995).
The aim of this paper is to present alternative methods of fracture surface evaluation. In previous research the
approach based on angular deviations of normal vectors was introduced, see Strnadel at al. 2015, Skalny and
Strnadel 2015. The fracture surface is covered with the net of triangles. For every triangle vector perpendicular to the
triangle - normal vector is calculated. Every triangle was evaluated by the greatest angle of its normal vector with
normal vectors of neighbouring triangles. Angular deviations were in different meaning applied e. g. in Berniera et
al. (2013) and Eckart et al. (1985). Apart from previous work where the usage of normal vector angular deviation
was discussed, in the present paper multivariate characteristics of normal vectors are taken into account. To evaluate
the ductile fracture surface the k-means++ clustering method is used see Arthur and Vassilvitskii (2007) and
conditional probability distribution.

Nomenclature

𝐷𝐷S box counting dimension

𝑥𝑥 𝑥𝑥-th normal vector component
𝑦𝑦 𝑦𝑦-th normal vector component
𝑑𝑑 𝑥𝑥 difference in the 𝑥𝑥-th direction
𝑑𝑑 𝑦𝑦 difference in the 𝑥𝑥-th direction
∇𝑖𝑖 covering triangle
𝜑𝜑 maximal angular deviaton
𝜙𝜙 potential function
𝐶𝐶 cluster
𝑐𝑐 cluster centre
 Pavel Skalny/ Structural Integrity Procedia 00 (2016) 000–000 3
Pavel Skalny / Procedia Structural Integrity 2 (2016) 3727–3734 3729

Fig. 1. (a) first picture; (b) second picture.

2. Drop weight tear test specimens Data

Fracture surfaces of broken DWTT specimens are studied using samples from commercially produced API 5L X-
70 sheet steel with thickness 18.7 mm. The chemical composition of the steel from five different melts and other
materials properties are presented in Strnadel et al. (2013). The steel was austenitized at 1200°C and rolled with an
initial temperature of 985°C and a final rolling temperature of 832°C. Then it was water-cooled from 800°C to 465°
at 9.1 °C/s.
The DWTT specimen fracture surfaces were photographed using a 3D camera produced by Limess Measurement
Technique and Software. The 3D camera projects straight lines onto the DWTT specimen and photographs the
deformed image of the lines. This projection method makes it possible to approximate the fracture surface with a
network of discrete points and to record their 𝑥𝑥 , 𝑦𝑦 and 𝑧𝑧 coordinates. The scan was not realized with high
magnification so every square millimeter is represent with no more than 80 measurements. In Figure 1 there is a top
view on the DWTT specimen. We can see that the structure of the fracture surface is almost unidentifiable
(comparing to Figure 4). In following chapters we will present that despite the sparse data source it is possible to
correctly evaluate the fracture surface. Fracture surface characterization
In this chapter two approaches to the fracture surface evaluation are presented. The fracture surface is
characterized with the box-counting dimension and with the new approach based on normal vector analysis. Results
of Both methods are compared in the chapter 2.3.

2.1. Box-Counting Dimension

The fractal geometry concept is used to describe highly segmented surfaces. A suitable way to describe the
degree of segmentation of fracture surfaces is the usage of the fractal dimension 𝐷𝐷𝐹𝐹 . In this paper the fractal
dimension is estimated with the box counting dimension. The box-counting dimension 𝐷𝐷𝑆𝑆 is probably the most often
used method for estimating the fractal dimension. The box-counting dimension is relatively easy to implement on a
computer, and for a large class of sets gives analogous results as a direct calculation of fractal dimension. However
the box-counting dimension is loaded with significant error in some cases see Schmittbuhl et al. 1995. It can prove
that the box-counting dimension is an upper estimate of the fractal dimension see Falconer (2011). The box-
counting method 𝐷𝐷𝑆𝑆 can be calculated as:
log 𝑁𝑁(𝛿𝛿)
𝐷𝐷S = lim𝛿𝛿→0 (1)
log(1/𝛿𝛿)
3730 Pavel Skalny / Procedia Structural Integrity 2 (2016) 3727–3734
4 Pavel Skalny/ Structural Integrity Procedia 00 (2016) 000–000

Where 𝑁𝑁(𝛿𝛿) is the number of covering sets with the diameter not greater then δ. In our implementation the 𝑁𝑁(𝛿𝛿)
denotes the number of squares (for other possibilities see Falconer 2014) in the square mesh grid with the edge
length equal to 𝛿𝛿. In practical calculation the 𝐷𝐷S is calculated using the least square method approximating the linear
dependency between log(1/𝛿𝛿) and log 𝑁𝑁(𝛿𝛿).
𝑦𝑦
As 𝐷𝐷𝑆𝑆𝑥𝑥 𝐷𝐷𝑆𝑆 we will denote the box counting dimension in the 𝑥𝑥, 𝑦𝑦 direction. The 𝑥𝑥 direction corresponds to the
direction of crack propagation whether the 𝑦𝑦 direction is perpendicular to the direction of the crack propagation.

2.2. Mormal vector characteristics

Besides methods of the fractal geometry an alternative approach to identify fracture surface was applied. The
fracture surface was covered with the net of triangles. The vertices of triangle correspond to real measurements of
the fracture surface. To create the triangle net the Delaunay triangulation was used, see Ceong and Kreveld (2008)
The Delaunay triangulation maximizes the lower angle of each triangle, so the final triangulation contents as regular
triangles as possible. For every triangle the unit vector perpendicular to the triangle -normal vector was calculated.
In a previous research every triangle was evaluated by the greatest angle 𝜑𝜑 of its normal vector with normal
vectors of neighbouring triangles. Although the angular deviation describes the fracture surface quite well see
Skalny and Strnadel (2015), we lose the information about the direction where normal vector change the most. Due
to the fact we will add other characteristics to the angular deviation. For every (unit) vector we will use the length of
its x and y component. The components describe how much is the surface tilted in the 𝑥𝑥, 𝑦𝑦 direction. Furthermore
we will consider changes of 𝑥𝑥 and 𝑦𝑦 components of two vectors with the greatest angle deviation. The changes of
two vectors 𝑛𝑛1 , 𝑛𝑛2 was computed in following way:
After heuristic analysis of roughness based on angular deviations of normal vectors we concern our attention to
other characteristics. For given normal vector 𝑛𝑛𝑖𝑖 = ( 𝑥𝑥𝑖𝑖 , 𝑦𝑦𝑖𝑖 , 𝑧𝑧𝑖𝑖 ) related to the centre of gravity 𝑇𝑇𝑖𝑖 = ( 𝑥𝑥𝑖𝑖𝑇𝑇 , 𝑦𝑦𝑖𝑖𝑇𝑇 , 𝑧𝑧𝑖𝑖𝑇𝑇 )
of the triangle ∇𝑖𝑖 we take into account the pair 𝑥𝑥𝑖𝑖 , 𝑦𝑦𝑖𝑖 (the 𝑧𝑧-th component 𝑧𝑧𝑖𝑖 does not provide valuable information).
Values 𝑥𝑥𝑖𝑖 , 𝑦𝑦𝑖𝑖 partially describe inclination of the triangle. ∇𝑖𝑖 (see Figure ...). For every fixed 𝑖𝑖 chose a neighbour ∇𝑗𝑗
of the triangle ∇𝑖𝑖 with maximal angular deviation of related normal vectors. Now, as additional input data for cluster
the further analysis we compute differences
𝑥𝑥𝑗𝑗 − 𝑥𝑥𝑖𝑖 𝑦𝑦 𝑦𝑦𝑗𝑗 − 𝑦𝑦𝑖𝑖
𝑑𝑑𝑖𝑖𝑥𝑥 = 𝑇𝑇 𝑇𝑇 , 𝑑𝑑𝑖𝑖 = 𝑇𝑇 𝑇𝑇. (2)
𝑥𝑥𝑖𝑖 – 𝑥𝑥𝑖𝑖 𝑦𝑦𝑖𝑖 – 𝑦𝑦𝑖𝑖

Note, the choice of the proper triangle can be realized with respect to the different criterion, e.g. 𝛻𝛻𝑗𝑗 can be done as
a neighbour of 𝛻𝛻𝑖𝑖 with minimal distance of the centres of gravity. This approach gives similar results as method
described above.
Let us summarize, that every triangle with the centre of gravity 𝑇𝑇𝑖𝑖 is evaluated with five values maximal
𝑦𝑦
deviation 𝜑𝜑𝑖𝑖 , normal vector components 𝑥𝑥𝑖𝑖 , 𝑦𝑦𝑖𝑖 and differences 𝑑𝑑𝑖𝑖𝑥𝑥 , 𝑑𝑑𝑖𝑖 .

2.3. Comparison

In this section we will present mean values of box counting dimension and normal vector characteristics. All the
values were calculated from 30 different DWTT specimens. In table 1 there are presented mean values of studied
characteristics calculated for the area with brittle and ductile fracture. The simple paired t-test was applied to test
whether the values on the area of brittle and ductile fracture are significantly different. In table 2 there is presented
the correlation matrix. The correlation coefficients are presented in the upper triangular matrix. The results of the
𝑛𝑛−2
test of the statistical significance (using the testing statistics 𝑡𝑡 = √ distributed with the student distribution with
1−𝑟𝑟 2
𝑛𝑛 − 2 degrees of freedom) of correlation coefficients are presented in the lower triangular matrix. Both statistical
tests are realized on the significance level 0.05. The alternative hypotheses are formulated in the one tail form.
Generally we can conclude, that the box counting dimensions are correlated with the angular deviation 𝜑𝜑 and
with differences 𝑑𝑑 𝑥𝑥 , 𝑑𝑑 𝑦𝑦 . Except for vector components all the values are significantly different in the area of the
brittle and ductile fracture.
 Pavel Skalny / Procedia Structural Integrity 2 (2016) 3727–3734 3731
Pavel Skalny/ Structural Integrity Procedia 00 (2016) 000–000 5

Table 1. mean values of box counting dimensions and normal vector characteristics
y (n)-statistical significant (insignificant) difference.
Area Brittle fracture Ductile fracture Stat. significance
differencee
𝐷𝐷𝑆𝑆𝑥𝑥 1.18 1.08 y
𝑦𝑦
𝐷𝐷𝑆𝑆 1.15 1.02 y
𝜑𝜑 17.51 15.32 y
𝑥𝑥-component (abs) 0.32 0.28 n
𝑦𝑦 -componnent (abs) 0.27 0.28 n
𝑑𝑑 𝑥𝑥 (abs) 3.1 2.3 y
𝑦𝑦
𝑑𝑑 (abs) 2.9 2.2 y

Table 1. Correlation between proposed characteristics y (n)-statistical significant (insignificant) correlation.

𝑦𝑦
𝐷𝐷𝑠𝑠𝑥𝑥 𝐷𝐷𝑠𝑠 𝜑𝜑 𝑥𝑥 𝑦𝑦 𝑑𝑑 𝑥𝑥 𝑑𝑑 𝑦𝑦
𝐷𝐷𝑠𝑠𝑥𝑥 1 0.6 0.71 0.22 0.15 0.49 0.38
𝑦𝑦
𝐷𝐷𝑠𝑠 y 1 0.75 0.27 0.31 0.41 0.35
𝜑𝜑 y y 1 0.25 0.28 0.35 0.38
𝑥𝑥-component (abs) n n n 1 0.15 0.8 0.22
𝑦𝑦 -componnent (abs) n y n n 1 0.23 0.75
𝑑𝑑 𝑥𝑥 (abs) y y y y n 1 0.35
𝑦𝑦
𝑑𝑑 (abs) y y y y y y 1

3. Proposed methods

In this chapter there are presented two methods for analyzing normal vectors characteristics. The usage of
Bayesian conditional distribution (section 4.2) was already used in Skalny and Strnadel (2015) and Strnadel et al.
(2015). In section 4.1 the cluster k-means algorithm is presented.

3.1. K-means

The k-means is well known clustering algorithm. It divides the data set 𝑋𝑋 ⊂ ℝ𝑑𝑑 to 𝑘𝑘 clusters 𝐶𝐶1 , … , 𝐶𝐶𝑘𝑘 , so every
𝑥𝑥 ∈ 𝑋𝑋 belongs to the cluster 𝐶𝐶𝑖𝑖 with the nearest center 𝑐𝑐𝑖𝑖 . K-means solves the problem of minimizing the potential
function 𝜙𝜙
2
ϕ = ∑𝑥𝑥∈𝑋𝑋 ∑𝑐𝑐𝑖𝑖∈𝐶𝐶‖𝑥𝑥 − 𝑐𝑐𝑖𝑖 ‖ (3)

with respect to 𝐶𝐶.

Principally the problem of minimizing the potential function can hardly be solved by finding the best solution from
all possible realizations-optimal solution. K-means solve the form 1 in finding suboptimal solution with respect to
the choice of initial centers (or clusters). The basic procedure of the algorithm can be described in the following
way:

 Choose 𝑘𝑘 initial centers {𝑐𝑐1 , 𝑐𝑐2 , … , 𝑐𝑐𝑘𝑘 }.

 Assign each observation 𝑥𝑥 to the cluster (1, … , 𝑘𝑘) with the “nearest” center (often used Euclidean distance).
 Set new centre as a mean of every cluster.
 Repeat previous two steps until no cluster 𝐶𝐶𝑖𝑖 changes.

Every change of the k-means algorithm decreases the potential function until the local minimum of the function
is achieved. The general k-means algorithm has many modifications, e.g. instead of mean the median or the mode
3732 Pavel Skalny / Procedia Structural Integrity 2 (2016) 3727–3734
6 Pavel Skalny/ Structural Integrity Procedia 00 (2016) 000–000

can be used. The results of the k-means are also influenced by the chosen distance for quantifying the nearest
distance. Euclidean distance is used in this work. In the presented paper we will use the k-means++ variant, where
the initial centres are achieved in following way:

 The first center 𝑐𝑐1 is choosen uniformly from the data set 𝑋𝑋.
𝐷𝐷(𝑥𝑥)2
 All other centers (up to 𝑘𝑘) are one after other chosen from 𝑋𝑋 with the probability ∑ 2 , where the 𝐷𝐷(𝑥𝑥)
𝑥𝑥∈𝑋𝑋 𝐷𝐷(𝑥𝑥)
denotes the shortest distance from the data point 𝑥𝑥 to the nearest center.

So that the highest probability to choose other centre has the data point with the highest (nearest) distance from
already chosen centers. K-means++ compared to k-means is faster and guarantees the lower bound to optimal
solution, see Arthur and Vassilvitskii (2007).

3.2. Probability identification

Another approach to the fracture identification is based on the use of conditional probability distribution.
Characteristics of normal vectors are observed separately in the area with purely brittle and ductile fracture. For
every type of the fracture the probability distribution is estimated. In previous work the distribution was estimated as
a Gaussian mixture. In the presented paper simple multivariate normal distribution is used. For every observation 𝑥𝑥𝑖𝑖
(five elements vector) the probability 𝑝𝑝𝑑𝑑 (𝑥𝑥𝑖𝑖 ) that it belongs to the ductile fracture is calculated.
𝑓𝑓 (𝑥𝑥 )
𝑝𝑝𝑑𝑑 (𝑥𝑥𝑖𝑖 ) = (𝑥𝑥 𝑑𝑑)+𝑓𝑓𝑖𝑖 (4)
𝑓𝑓𝑏𝑏 𝑖𝑖 𝑑𝑑 (𝑥𝑥𝑖𝑖 )

where 𝑓𝑓𝑑𝑑 (𝑥𝑥𝑖𝑖 ), 𝑓𝑓𝑏𝑏 (𝑥𝑥𝑖𝑖 ) are distributions of the ductile and brittle fracture area.

4. Results

Best result in k-means clustering can be achieved by using four clusters. Fewer clusters have proved
unsatisfactory and the result with a larger number of clusters is very confusing. In Figure 2 there is presented the
result of k-means clustering with four clusters. In this case the first and the second cluster (blue and light blue)
represents the brittle fracture area, the third cluster (yellow) represents the ductile fracture area. The fourth cluster
(red) represents mainly the notch in the DWTT specimen. Some inaccuracies are seen at the borders of the fracture
surface and at the high plastic deformation area (down on the left side).
The probability that the vector belongs to the ductile fracture is presented in Figure 3. As in the two cluster
analysis the probability identification is loaded with an error in the area of notch and in the area of high plastic
deformation.

Fig. 2. (a) first picture; (b) second picture.

 PavelPavel
Skalny/ Structural
Skalny Integrity
/ Procedia Procedia
Structural 00 (2016)
Integrity 000–000
2 (2016) 3727–3734 7
3733

Fig. 3. (a) first picture; (b) second picture.

Fig. 4 analyzed DWTT specimen

5. Conclusions

A detailed quantitative fractographic analysis of fracture surfaces of X70 steel DWTT specimens was performed
in order to investigate all possible ways of evaluating its character, especially the ductile fracture percentage,
independently of individual observation.
The roughness of the ductile fracture is lower than the ductile fracture of tested X70 steel at -20°C. Thus the
ductile fracture area is represented by lower fractal dimension and normal vector characteristics. Presented vectors
characteristics seems to be useful alternative to the fractal geometry concept and can be used for determining the
ductile fracture area. It can be expected that the normal vector characteristics are related to the mechanical properties
of the material as well as the fractal dimension.
Both k-means and conditional probability distribution are useful tools for the fracture identification. The
advantage of the k-means method is the computational effectiveness and the complete elimination (except the
determining an amount of clusters) of human factor in the fracture analysis. Conversely the probability identification
is partially dependent on the right choice of the area where the probability distributions are estimated. In the other
way the output of probability identification is better for further analyzing.
3734 Pavel Skalny / Procedia Structural Integrity 2 (2016) 3727–3734
8 Pavel Skalny/ Structural Integrity Procedia 00 (2016) 000–000

References

Arthur, D., Vassilvitskii, S., 2007. K-means++: the advantages of careful seeding, Proceedings of the eighteenth annual ACM-SIAM symposium
on Discrete algorithms. Philadelphia, United States, 1027-1035.
Balankin, A., Morales, D., Mancilla, G., Susarrey, O., Campos, I., Sandoval, F., Bravo, A., García, A., Galicia, M., 2000. Fractal Properties of
Fracture Surfaces in Steel 1045. International Journal of Fracture 106, 21-26.
Berniera, N., Leunisa, E., Carlos Furtadoa, C., Puttea, T., Bana, G., 2013. EBSD study of angular deviations from the Goss component in grain-
oriented electrical steels. Micron 54, 43-51.
Chang, Q., Chen, D., Ru., H., Yue, L., Yu, C., 2011. Three-dimensional fractal analysis of fracture surface sin titanium–iron particulate
reinforced hydroxyapatitecomposites: relationship between fracture toughness and fractal dimension, Journal of Material Science, 46, 6118-
6123.
CHEONG, O., KREVELD, M., OVERMARS, M., Computational Geometry: Algorithms and Applications. Springer, London, 2008, pp. 380.
Falconer, K., 2003. Fractal Geometry: Mathematical Foundations and Applications. John Wiley & Sons, New York, pp. 368.
Fang, J., Zhang, J., 2014. Evaluation of cracking behavior and critical CTOA values of pipeline steel from DWTT specimens. Engineering
Fracture Mechanics 124, 18-29.
Horsley, D., 2003 Background to the use of CTOA for prediction of dynamic ductile fracture arrest in pipelines. Engineering Fracture Mechanics
70, 547–552.
Hwang, B., Lee, S. Kim, YM., Kim, NJ., Yoo, JY., Woo, ChS., 2004. Analysis of abnormal fracture occurring during drop-weight tear test of
high-toughness line-pipesteel. Material Science Engineering 368, 18-27.
Jiang, X., Chu, W., Hsiao, C., 1994. Relationship between JIC and fractal value of fracture surface of ductile materials. Acta Metallurgica et
Materialia 42, 105–8.
Mandelbrot, B., Passoja, D., Paullay, AJ., 1984, Fractal character of fracture surfaces of metals. Nature 308, 721–722.
Mengjia, Xu., Jijin, Xu, Hao, Lu., 2015. Fractal and probability analysis of creep crack growth behavior in 2.25Cr-1.6W steel incorporating
residual stresses. Applied Surface Science 359, 73-81.
Ray, K., Mandal, G., 1992. Study of correlation between fractal dimension and impact energy in a high strength low alloy steel. Acta
Metallurgica et Materialia 40, 463–469.
Stach, S., Cybo, J., Chmiela, J., 2000. Fracture surface - fractal or multifractal?. Stereology and Image Analysis in Materials Science 46, 163–167.
Seifert, K., 1984. Abnormal fracture appearances of DWTT specimen from high-toughness line pipe steels. Materialprüfung 26, 277–80.
Schmittbuhl, J., Vilotte, J., Roux, S., 1995. Reliability of self-affine measurements. Physical Review 51, 131-137.
Skalny, P., Strnadel, B., 2015. Application of Gaussian mixtures in the Fracture Identiffication. 8th International Conference on Information and
Digital Technologies, Zilina, Slovakia, 330-334.
Strnadel, B., Hasegawa, K., Skalný, P., 2015. Evaluation of Cracking Behavior of Pipeline X70 Steel From DWTT Broken Specimens. 24th
Metal Conference, Brno, Czech Republic, 128-133.
Strnadel, B.; Ferfecki, P.; Zidlik, P., 2013. Statistical characteristics of fracture surfaces in high-strength steel drop weight tear test specimens.
Engineering Fracture Mechanics 112, 1-13.
Yang, XB., Zhuang, Z., You, XC., Feng, YR., Huo, CY., Zhuang, CJ.,2008. Dynamic fracture study by an experiment/simulation method for rich
gas transmission X80steel pipelines. Engineering Fracture Mechanics 75, 5018–5028.
Yang, Z., Kim, ChB., Feng, Y., Cho, Ch., 2008. Abnormal fracture appearance in drop-weight tear test specimens of pipeline steeel. Mater
Science Engineering 483, 239–41.
Weiss, J., 2001. Self-affinity of fracture surfaces and implications on a possible size effect on fracture energy. International Journal of Fracture
109, 365-381.