Full Chapter Ekc 2019 Conference Proceedings Science Technology and Humanity Advancement and Sustainability Jong Mun Park PDF
Full Chapter Ekc 2019 Conference Proceedings Science Technology and Humanity Advancement and Sustainability Jong Mun Park PDF
Full Chapter Ekc 2019 Conference Proceedings Science Technology and Humanity Advancement and Sustainability Jong Mun Park PDF
https://textbookfull.com/product/sustainability-in-environmental-
engineering-and-science-select-proceedings-of-sees-2019-sunil-
kumar/
https://textbookfull.com/product/advanced-multimedia-and-
ubiquitous-engineering-future-information-technology-
volume-2-1st-edition-james-j-jong-hyuk-park/
https://textbookfull.com/product/proceedings-of-the-
international-petroleum-and-petrochemical-technology-
conference-2019-jiaen-lin/
https://textbookfull.com/product/proceedings-of-international-
conference-on-data-science-and-applications-icdsa-2019-kanad-ray/
https://textbookfull.com/product/smart-sensors-and-systems-
technology-advancement-and-application-demonstrations-yongpan-
liu/
https://textbookfull.com/product/energy-materials-and-devices-
proceedings-of-e-mad-2022-advances-in-sustainability-science-and-
technology-1st-edition-ambesh-dixit/
Jong Mun Park
Dong Ryeol Whang Editors
EKC 2019
Conference
Proceedings
Science, Technology, and Humanity:
Advancement and Sustainability
EKC 2019 Conference Proceedings
Jong Mun Park • Dong Ryeol Whang
Editors
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface
EKC 2019, the 12th EKC, was held in Vienna, Austria, from July 15 to 18, 2019. It
was hosted by the Korean Scientists and Engineers Association in Austria
(KOSEAA) together with the Korean Federation of Science and Technology Soci-
eties (KOFST) and eight other Korean Scientists and Engineers Associations in
Europe which are in Germany (VeKNI), the UK (KSEAUK), France (ASCoF),
Austria (KOSEAA), Finland (KOSES), Scandinavia (KSSEA), the Netherlands
(KOSEANL), Switzerland (KSEAS), and Belgium (KOSEAbe). Since its first
successful start in 2008 in Heidelberg, Germany, EKC has been held annually in
different European countries and has become the most important scientific and social
event, bringing scientists and engineers from Europe and Korea together.
Under the theme of “Science, Technology and Humanity: Advancement and
Sustainability,” EKC 2019 was successfully held with more than 680 registered
participants. Three hundred twenty nine papers were presented in 50 technical
sessions. Among them, high impact-research results from EKC 2019 are selected,
peer-reviewed, and stapled in this volume. Fourteen research papers were submitted
and 11 were accepted after rigorous peer-review process from the program commit-
tee members and external experts.
v
Organization
CONFERENCE CHAIR
PARK, Jong Mun (ams AG/KOSEAA President)
CONFERENCE CO-CHAIRS
KIM, Myung-Ja (KOFST President)
PARK, Wonsun (GEOMAR Helmholtz Centre for Ocean Research Kiel/VeKNI
President)
LIM, Sungwoo (The Open University/KSEAUK President)
KIM, Junbeum (Université de Technologie de Troyes/ASCoF President)
CHO, Hyong Sil (Microsoft/SiLnD/KOSEANL President)
CHOE, Young Han (International Telecommunication Union/KSEAS President)
LEE, Jae Wung (VTT Technical Research Centre of Finland/KOSES President)
YOO, YoonSeon (BlackBerry/KSSEA President)
OH, Kun Sang (KOSEAbe President)
ADVISORY BOARD
LEE, Eun-Woo (KOFST)
YOO, Martin S. D. (CRUSE Offshore GmbH/VeKNI)
SEOK, Joon-Weon (GORI Engineering/VeKNI)
JUN, Chang Hoon (ITER/ASCoF)
JEUNG, Gwang-Hi (Institut des sciences moléculaires de Marseille/ASCoF)
HAN, Man Wook (Technische Universität Wien/KOSEAA)
KIM, Keunjae (SSPA/KSSEA)
PARK, Migeun (University of Strathclyde/KSEAUK)
SECRETARY GENERAL
LEE, Hana (Technische Universität Graz/KOSEAA)
vii
viii Organization
REGISTRATION
GWON, Jihee (Muthesius University of Fine Arts and Design/VeKNI)
KWON, Jaedeok (University of Glasgow/KSEAUK)
KANG, Myung-Ah ( Universté Clermont Auvergne/ASCoF)
LEE, Sun Mi (KOSEANL)
LEE, Juneseung (ETH Zurich/KSEAS)
MUN, Gwan-gyeong (Intel/KOSES)
KIM, Jaeoh (KSSEA)
HEO, Changhoon (imec/KOSEAbe)
PROGRAMME CHAIR
WHANG, Dong Ryeol (Johannes Kepler Universität Linz/KOSEAA)
DIVISION CHAIRS
KANG, Kab Seok (Max Planck Institute for Plasma Physics/VeKNI)
KIM, Chan (European XFEL/VeKNI)
YOON, Songhak (Fraunhofer IWKS/VeKNI)
KIM, Wonjae (VTT Technical Research Center of Finland/KOSES)
MOK, K. Hun (Trinity College Dublin/KSEAUK)
NAM, Kiwoong (Institut National de la Recherche Agronomique/ASCoF)
KIM, Junbeum (University of Technology of Troyes, France/ASCoF)
LEE, Hyunjung (City of Stuttgart, Office for Environmental Protection/VeKNI)
LEE, Pyoung-Jik (University of Liverpool/KSEAUK)
SEO, Hyewon (CNRS-Univ. Strasbourg/ASCoF)
JUNG, Sung Kyo (NXP Software/ KOSEAbe)
CHOI, Jung Han (Fraunhofer Heinrich Hertz Institute/VeKNI)
JEONG, Cheol-Ho (DTU (Denmark Technical University)/KSSEA)
HA, Kwangtae (Fraunhofer IWES/VeKNI)
CHO, Hyong Sil (SiLnD; Microsoft/KOSEANL)
LOC CHAIR
HAN, Man Wook (Technische Universität Wien/KOSEAA)
FINANCE DIRECTOR
LEE, Seung-Hun (KOSEAA)
LOC MEMBERS
PARK, Young-Saeng (University of Warwick/KSEAUK)
KOCH, Kyungran (KOSEAA)
MIN, Jihoon (IIASA/KOSEAA)
Contents
ix
Novel Adamantane Asymmetrically
Substituted Diketopyrrolopyrroles
Abstract The effect of the alkylation and solubilization ethyladamantyl side chains
to diketopyrrolopyrroles (DPPs) on the thermal and optical properties was system-
atically investigated. Nucleophilic substitution of various alkyl chains into the DPP
molecule is a powerful tool for increasing solubility and processability of these
pigments, leading to their broader applications. Moreover, alkylation can often
contribute to solid state packing and structural ordering of DPPs, depending mainly
on a character of alkyl chain. In this work, we have synthesized the series containing
three derivatives (N,N0 -, N,O0 - and O,O0 -substituted) of alkylated thiophene-DPP
derivatives by ethyladamantyl substituents. It was focused on precise separation of
all formed DPP derivatives in order to their deeper study. It has been found that the
O-substitution leads to worse thermal stability of materials, based on
thermogravimetric measurements. On the other hand, ethyladamantyl side chain as
rigid alicyclic substituent very effectively contributed to increase of melting point
and thermal stability. The new findings provide valuable information about the
previously overlooked regioisomers formed during alkylation of the DPP molecule,
in particular in terms of thermal stability and optical properties in solution.
1 Introduction
characterized by high absorption coefficients [2], high thermal and chemical stability
[2] and good charge transfer capability [3]. One of the most recently examined and
most attractive groups of organic pigments are derivatives of 2,5-dihydropyrrolo
[4,3-c]pyrrolo-1,4-dione (DPP) [4], having a considerable application potential for
easily processed low-cost sustainable electronics devices [5]. The DPP backbone
provides the great potential for a wide range of chemical derivatizations to prepare
target derivatives exhibiting the desired chemical and physical properties [6]. As a
result, there are currently many significant scientific reports dealing with the appli-
cations of DPP derivatives as high performance organic pigments used in functional
devices such as organic field-effect transistors [7–11], organic light-emitting diodes
[12, 13], dye-sensitized [14] and bulk heterojunction solar cells [15], sensors [16]
and biosensors [17], fluorescence probes [18] etc.
The DPP derivative substituted with 2-thiophene at positions 3 and 6 is most
often found in the previously mentioned fields of application [19]. An important
approach to increase efficiency of this material is to extend the length of
π-conjugation in the molecule, which can be achieved by polymerization via direct
arylation [20, 21] or by the cross-coupling reactions on the side aromatic rings of the
DPP core [22, 23]. Another very important derivatization is N,N0 -alkylation of the
DPP molecule, providing considerably higher solubility compared to basic N,N-
0
-unsubstituted DPP derivatives. These derivatives possess good thermo and
photostability [2, 24], however strong intermolecular hydrogen bonding between
neighbouring oxygen and nitrogen atoms of lactam groups leads to a relatively low
solubility in most of common organic solvents [2]. The solubility improvement of N,
N0 -alkylated DPPs with linear or branched alkyl chains is caused by interruption of
intermolecular hydrogen bonds and leads to broader applications, especially in
previously mentioned areas of electronics [25]. On this basis, alkylation is one of
the most important and most common derivatization of DPPs. Incorporation of alkyl
chains into DPP core is performed through base-catalysed nucleophilic substitution,
usually carried out as a one-pot reaction [26]. Nevertheless, two neighbouring O-
and N-atoms of the lactam group lead to delocalization of the negative charge in the
formation of the DPP anion, resulting to decreasing of regioselectivity of N,N-
0
-substitutions. The formation of O-alkylated DPP by-products was already proved
[27] but there are very few reports dealing with their properties [28, 29]. Nevertheless,
O,O0 - and asymmetrical N,O0 - adamantane-substituted derivatives can be a point of
interest not only from fundamental but also from practical point of view.
Recently the synthesis and study of a novel DPP derivative N,N0 -substituted by
bulky ethyladamantyl chains was reported by our group [30]. Substitution has led to
extraordinary high ordered adamantane induced molecular packing what resulted to
exceptional ambipolar behaviors of this new pigment. During the synthesis of the
aforesaid DPP derivative, very low regioselectivity of N-alkylation was encountered
and the formation of by-products was largely observed.
Therefore, the main aim of this work has been to focus on systematically study of
the influence of bulky adamantyl chains on either regioselectivity of the DPP
alkylation and properties of the resulting derivatives. The target was to synthesize
and isolate all formed products in weighable quantities and subjecting these
Novel Adamantane Asymmetrically Substituted Diketopyrrolopyrroles 3
derivatives to the study of the effect of adamantyl substituents on optical and thermal
properties in case of symmetrical (N,N0 -; O,O0 -) and asymmetrical (N,O0 -) products.
2 Experimental Section
2.1 Materials
2.2 Characterization
1
H NMR spectra were recorded on a FT-NMR spectrometer Bruker Avance III
300 MHz or 500 MHz in CDCl3. Chemical shifts (δ) are given in parts per million
(ppm) relative to TMS as an internal reference. The melting point was determined on
a Kofler apparatus and the temperature was not calibrated. Mass spectra were
recorded on a GC–MS spectrometer Thermo Fisher Scientific ITQ 700 (DEP).
Elemental analysis was measured with an elemental analyser Flash 2000 CHNS
Thermo Fisher Scientific.
Thermogravimetry (TG) was conducted on TA Instruments Q5000IR (New
Castle, Delaware, USA) to analyze the stability of the derivatives and changes in
mass before degradation. A sample was placed on the Pt crucible sample holder and
heated at 10 C/min from room temperature to 600 C under a stream of nitrogen
(5.0 ultra high purity) flow rate 40 mL/min. TG was used to analyze the thermal
stability of investigated materials. Briefly, the TG record (dependence of mass on
temperature) was derived and onset of the derivate was determined. TG records were
evaluated using TRIOS and Universal analysis software provided by TA
Instruments.
Solutions were prepared by diluting materials in anhydrous chloroform. Concen-
tration of materials was 105 to 106 mol dm3. Solutions were characterized in a
quartz cuvette (Herasil®, Heraeus Quarzglas Co.). Absorption spectra of samples
were measured by employing Varian Cary Probe 50 UV-VIS spectrometer (Agilent
4 M. Cigánek et al.
Technologies Inc.). The fluorescence spectra were recorded with a Horiba Jobin
Yvon Fluorolog. This apparatus equipped by integration sphere was also used to
determination of fluorescence quantum yield by absolute method. Fluorescence
lifetime was measured by Horiba Jobin Yvon Fluorocube.
2.3 Synthesis
H S
O O N
S CN O (i) (ii)
+
O
N O
O S H
Th-DPP: 59%
O N S O N S O N S
+ +
S N O S N O S N O
Fig. 1 Synthetic pathway for the synthesis of the basic thiophene DPP molecule and subsequently
N,N0 -; N,O0 - and O,O0 -alkylated DPP derivatives
(i)
Na, FeCl3 (cat.)/t-amyl alcohol, 102 C, 24 h
(ii)
1. K2CO3; 2. 1-(2-bromoethyl)adamantane/DMF, 105 C, 2 h
Novel Adamantane Asymmetrically Substituted Diketopyrrolopyrroles 5
ppm): δ ¼ 8.05 (d, J ¼ 4.5 Hz, 2H), 7.55 (d, J ¼ 5.5 Hz, 2H), 7.19 (dd, J ¼ 8.1,
1.1 Hz, 2H), 4.69–4.65 (m, 4H), 1.98 (m, 4H), 1.70–1.64 (m, 24H), 1.54 (s, 4H),
1.34–1.26 (m, 2H), Anal. calcd. for C38H44N2O2S2: C 73.04%, H 7.10%, N 4.48%,
Found: C 72.96%, H 6.99%, N 4.51%.
The optical properties of DPP derivatives were studied in solution to investigate the
effects given by the position of solubilization group substitution.
The Fig. 3 summarizes absorption and fluorescence emission spectra. The origin
of fluorescence spectra is confirmed by excitation spectra, which are in good
agreement with absorption spectra. There is significant difference between spectra
of N,N0 -; N,O0 - and O,O0 -substituted DPPs. Quantitative parameters describing
optical properties of solutions are summarized in Table 2. They are confirming
that the position of alkyl chain substitution was leading to significant differences
in optical properties.
Absorption maximum is shifted with change of substitution position, while N,
N0 -EtAd-Th-DPP had maximum at 549 nm, N,O0 -EtAd-Th-DPP had maximum at
533 nm and O,O0 -EtAd-Th-DPP at 505 nm (see spectra on Fig. 3 and Table 2).
These shifts seem to be mainly consequence of suppression of electron transition to
lower vibration modes of excitation states in case of N,O0 - and O,O0 -derivatives
(zero phonon transitions are suppressed).
Novel Adamantane Asymmetrically Substituted Diketopyrrolopyrroles 7
100 1.2
404.91°C
1.0
N,N’-EtAd-Th-DPP
Weight (%)
60 0.6
0.4
40
0.2
20
372.51°C
0 0.0
0 100 200 300 400 500 600
Temperature (°C)
120 1.0 120 0.6
378.65°C
344.17°C
100 0.8 100
N,O’-EtAd-Th-DPP O,O’-EtAd-Th-DPP
0.4
Weight (%)
80 0.6 80
Weight (%)
60 0.4 60 0.2
296.50°C
40 0.2 40
0.0
20 0.0 20
0 –0.2 0 –0.2
0 200 400 600 800 0 200 400 600 800
Temperature (°C) Temperature (°C)
Table 2 Optical parameters of DPPs determined from absorption and fluorescence measurements
Derivative N,N0 -EtAd-Th-DPP N,O0 -EtAd-Th-DPP O,O0 -EtAd-Th-DPP
λABS max [nm] 549 533 505
ε [dm3mol1 cm1] 25,000 1000 12,000 200 26,000 1000
λABS edge [nm] 572 602 598
λPL[nm] 562 598 731
Stokes shift [eV] 0.05 0.26 0.76
τ [ns] 6.07 0.01 0.65 0.02 0.32 0.05
Eg opt [eV] 2.23 2.16 2.05
4 Conclusion
Acknowledgments This work was supported by the Czech Ministry of Industry and Trade Grant
TRIO FV20022 and MC thanks project No. FCH-S-19-5834.
References
1. Bagher AM (2014) Comparison of organic solar cells and inorganic solar cells. Int J Renew
Sustain Energy 3(3):53–58
2. Hao Z, Iqbal A (1997) Some aspects of organic pigments. Chem Soc Rev 26(3):203–213
3. Gsänger M, Huang DBL, Stolte M, Würthner F (2016) Organic semiconductors based on dyes
and color pigments. Adv Mater 28(19):3615–3645
4. Farnum DG, Mehta G, Moore GGI, Siegal FP (1974) Attempted reformatskii reaction of
benzonitrile, 1,4-diketo-3,6-diphenylpyrrolo[3,4-c]pyrrole. A lactam analogue of pentalene.
Tetrahedron Lett 15(29):2549–2552
5. Nielsen CB, Turbiez M, McCulloch I (2013) Recent advances in the development of semicon-
ducting DPP-containing polymers for transistor applications. Adv Mater 25(13):1859–1880
6. Grzybowski M, Gryko DT (2015) Diketopyrrolopyrroles: synthesis, reactivity, and optical
properties. Adv Opt Mater 3(3):280–320
7. Bronstein H, Chen Z, Ashraf RS et al (2011) Thieno[3,2-b]thiophene-diketo-pyrrolopyrrole-
containing polymers for high-performance organic field-effect transistors and organic photo-
voltaic devices. J Am Chem Soc 133(10):3272–3275
8. Chen TL, Zhang Y, Smith P, Tamayo A, Liu Y, Ma B (2011) Diketo-pyrrolopyrrole-containing
oligothiophene-fullerene triads and their use in organic solar cells. ACS Appl Mater Interfaces 3
(7):2275–2280
9. Li Y, Sonar P, Singh SP, Soh MS, Van Meurs M, Tan J (2011) Annealing-free high-mobility
diketopyrrolopyrrole-quaterthiophene copolymer for solution-processed organic thin film tran-
sistors. J Am Chem Soc 133(7):2198–2204
10. Yumusak C, Abbas M, Sariciftci NS (2013) Optical and electrical properties of electrochem-
ically doped organic field effect transistors. J Lumin 134:107–112
11. Glowacki ED, Romanazzi G, Yumusak C et al (2015) Epindolidiones-versatile and stable
hydrogen-bonded pigments for organic field-effect transistors and light-emitting diodes. Adv
Funct Mater 25(5):776–787
12. Vala M, Weiter M, Vyňuchal J, Toman P, Luňák S (2008) Comparative studies of diphenyl-
diketo-pyrrolopyrrole derivatives for electroluminescence applications. J Fluoresc 18
(6):1181–1186
13. Kalyani TN, Dhoble SJ (2012) Organic light emitting diodes: energy saving lighting
technology – a review. Renew Sust Energ Rev 16(5):2696–2723
14. Qu S, Wu W, Hua J, Kong C, Long Y, Tian H (2010) New diketopyrrolopyrrole (DPP) dyes for
efficient dye-sensitized solar cells. J Phys Chem C 114(2):1343–1349
15. Huo L, Hou J, Chen H-Y, Zhang S, Jiang Y, Chen TL, Yang Y (2009) Bandgap and molecular
level control of the low-bandgap polymers based on 3,6-dithiophen-2-yl-2,5-dihydropyrrolo
[3,4-c]pyrrole-1,4-dione toward highly efficient polymer solar cells. Macromolecules 42
(17):6564–6571
16. Qu Y, Hua J, Tian H (2010) Colorimetric and ratiometric red fluorescent chemosensor for
fluoride ion based on diketopyrrolopyrrole. Org Lett 12(15):3320–3323
17. Sokolov AN, Roberts ME, Bao Z (2009) Fabrication of low-cost electronic biosensors. Mater
Today 12(9):12–20
18. Fischer GM, Jüngst C, Isomäki-Krondahl M, Gauss D, Möller HM, Daltrozzo E, Zumbusch A
(2010) Asymmetric PPCys: strongly fluorescing NIR labels. Chem Commun 46
(29):5289–5291
Novel Adamantane Asymmetrically Substituted Diketopyrrolopyrroles 11
19. Bürgi L, Turbiez M, Pfeiffer R, Bienewald F, Kirner H-J, Winnewisser C (2008) High-mobility
ambipolar near-infrared light-emitting polymer field-effect transistors. Adv Mater 20
(11):2217–2224
20. Zhang J, Kang D-Y, Barlow S, Marder SR (2012) Transition metal-catalyzed C–H activation as
a route to structurally diverse di(arylthiophenyl)-diketopyrrolopyrroles. J Mater Chem 22
(40):21392–21394
21. Liu S-Y, Shi M-M, Huang J-C et al (2013) C–H activation: making diketopyrrolopyrrole
derivatives easily accessible. J Mater Chem A 1(8):2795–2805
22. Chan WK, Chen Y, Peng Z, Yu L (1993) Rational designs of multifunctional polymers. J Am
Chem Soc 115(25):11735–11743
23. Beyerlein T, Tieke B (2000) New photoluminescent conjugated polymers with 1,4-dioxo-3,6-
diphenylpyrrolo[3,4-c]pyrrole (DPP) and 1,4-phenylene units in the main chain. Macromol
Rapid Commun 21(4):182–189
24. Iqbal A, Jost M, Kirchmayr R, Pfenniger J, Rochart A, Wallquist O (1998) Bull Soc Chim Belg
97:615
25. Tieke B, Rabindranath AR, Zhang K, Zhu Y (2010) Conjugated polymers containing
diketopyrrolopyrrole units in the main chain. Beilstein J Org Chem 6:830–845
26. Naik MA, Venkatramaiah N, Kanimozhi C, Patil S (2012) Influence of side-chain on structural
order and photophysical properties in thiophene based diketopyrrolopyrroles: a systematic
study. J Phys Chem C 116(50):26128–26137
27. Frebort Š, Eliáš Z, Lyčka A, Luňák S, Vyňuchal J, Kubáč L, Hrdina R, Burget L (2011) O- and
N-alkylated diketopyrrolopyrrole derivatives. Tetrahedron Lett 52(44):5769–5773
28. Zhao B, Sun K, Xue F, Ouyang J (2012) Isomers of dialkyl diketo-pyrrolo-pyrrole: electron-
deficient units for organic semiconductors. Org Electron 13(11):2516–2524
29. Zhang L, Shen W, He R, Liu X, Tang X, Yang Y, Li M (2016) Fine structural tuning of
diketopyrrolopyrrole-cored donor materials for small molecule-fullerene organic solar cells: a
theoretical study. Org Electron 32:134–144
30. Kovalenko A, Yumusak C, Heinrichová P et al (2017) Adamantane substitutions: a path to
high-performing, soluble, versatile and sustainable organic semiconducting materials. J Mater
Chem C 5(19):4716–4723
31. Grzybowski M, Glodkowska-Mrowka E, Hugues V, Brutkowski W, Blanchard-Desce M,
Gryko DT (2015) Polar diketopyrrolopyrrole-imidazolium salts as selective probes for staining
mitochondria in two-photon fluorescence microscopy. Chem Eur J 21(25):9101–9110
32. Qian G, Qi J, Davey JA, Wright JS, Wang ZY (2012) Family of diazapentalene chromophores
and narrow-band-gap polymers: synthesis, halochromism, halofluorism, and visible–near infra-
red photodetectivity. Chem Mater 24(12):2364–2372
33. Zhou N, Vegiraju S, Yu X et al (2015) Diketopyrrolopyrrole (DPP) functionalized
tetrathienothiophene (TTA) small molecules for organic thin film transistors and photovoltaic
cells. J Mater Chem C 3(34):8932–8941
34. Govindan V, Wu CG (2017) Facile synthesis of low band-gap DPP–EDOT containing small
molecules for solar cell applications. RSC Adv 7(46):28788–28796
Machine Learning Approach on Steel
Microstructure Classification
H. Park (*)
Oasis International School, Ankara, Turkey
A. Öztürk
Middle East Technical University, Ankara, Turkey
1 Introduction
Steel is still one of the most important and extensively used alloy materials because
of its excellent mechanical properties while keeping costs low which gives a huge
variety of applications [1]. The principal elements in most steel, in addition to iron
and carbon, are boron, chromium, manganese, nickel, tungsten, and vanadium
[2]. The mechanical properties of steel are mainly determined by its microstructure
which are closely dependent on its chemical compositions.
Traditionally, the microstructures of steel are characterized by using standard
metallographic procedures based on chemical etching and optical microscopy and
they are compared with reference series [3]. Moreover, the steel’s microstructural
comparison with reference series is dependent on the expert’s subjective opinion.
The microstructure of the steel is usually controlled by heat treatment which
accompanies eutectoid transformation from austenite to pearlite. During the eutec-
toid phase transformation, FCC austenite is transformed into two different solid
phases of BCC ferrite and cementite (Fe3C) below 727 C. Here the two phases are
placed as alternative layers and it is named pearlite since its appearance is like a pearl
under the microscope as seen in Fig. 1. If the microstructure consists of more than
one phase, the properties of the steel strongly depend on the type and distribution of
the different phases in order to access the understanding between structure and
property relationship.
2 Experiment
2.1 Procedure
Steel samples of 2 cm diameter and 2 cm height were heat treated at 900 C for 1 h
and then cooled in a salt bath to form a mixture structure of pearlite and ferrite. The
samples were polished with sandpapers of #600, #1000, and #1200 successively,
and with diamond suspension to create a mirror surface. Chemical etching was
applied to selectively corrode microstructural features (grain boundaries, morpho-
logical features, pores, cracks, etc.) through putting a sample into a Nital solution
(2 ml HNO3 + 98 ml Ethanol) for 6 s. Microhardness was tested on the steel surface
with Vickers indenter and calculated by measuring the diagonal crack lengths.
Scanning Electron Microscope was operated to capture images of the steel micro-
structures. All the images were cropped to equal pixel size and converted it to JPG.
The images were separated into training data (used for training the model) and
testing data (used to test the effectiveness and accuracy of the model). The training
data images were manually color segmented into pearlite and ferrite. Data augmen-
tation was performed on the training images to generate 792 images from 18. AWS
EC2 p2.8large instance was set up and connected via terminal on a computer. The
EC2 GPU services were connected to an agent and the images were trained on a
U-NET model based on FCNN. The testing data images were tested on the new
model and the results were stored. The accuracy of the model was measured and the
properties of the microstructure were determined through a ratio test.
Physical The surface of the prepared steel sample was carefully grinded and
polished with a polisher. Polycrystalline diamond suspensions of 6- and 1-μm
particle size were used to make a mirror surface.
Figure 3 shows the microscope image of ferrite and pearlite phases which was taken
by SEM. While the dark area is the ferrite phase, the alternating layer areas with
bright strips are pearlite phase. The pearlite phases are well developed during the
cooling history and grain boundaries are observed clearly because of the proper
chemical etching.
Figure 4 shows the optical microscope image and the indentation after the Vickers
hardness test. By measuring the edge lengths of the diamond shape indentation, the
microhardness was calculated with the following equation,
HV ¼ 1:854 F=D2 ð1Þ
Machine Learning Approach on Steel Microstructure Classification 17
Fig. 4 Optical microscope image of ferrite and pearlite phases with the indention
18 H. Park and A. Öztürk
Fig. 5 Flow chart process of the training and testing of the model
where HV is the Hardness Value, F is the force (0.1 kgf), and D is the area of
indentation. The calculated microhardness was 122.6 2.4 MPa.
Hafiz [9] investigated the effect of matrix structure on the mechanical properties
in iron and suggested that hardness and pearlite volume fraction has the following
relationship,
3 Results
Table 1 summarizes the comparison between large and small images analyses. The
accuracy of the large training images, which were resized as 256 256 pixel and
with 12 epochs, is 98,689%, while that of the small training images, which were
resized as 128 128 pixel and with 5 epochs, is 98.095%. It is observed that both
results convey noticeably high accuracy with the use of U-NET architecture. Figure 6
shows the accuracy graph with large training images, and Fig. 7 shows the loss graph
with large training images according to epoch times (generations). Figure 8 shows
the accuracy graph with small training images. Figure 9 shows the loss graph with
small training images. The rapid progress the algorithm makes and learns each
generation can be observed. Small training took about 25 min while large training
took about 40 min.
20 H. Park and A. Öztürk
Fig. 10 Examples of successful cases in SEM segmentation using the U-NET architecture
All the machine learning training and testing were carried out through Amazon
Cloud Computing Service (AWS). Figure 10 shows successful examples of SEM
segmentation using FCNN networks. It can be said that most of the objects in each
microstructure image are classified correctly. From the data in Table 1 and the
accuracy graphs, the Machine Learning algorithm equipped with U-NET architec-
ture model based on the FCNN successfully distinguished the pearlite phases from
the ferrite phases.
Machine Learning Approach on Steel Microstructure Classification 21
3.3 Analysis
Machine Learning techniques were capable of separating ferrite and pearlite regions
into different colors. Collecting only the pearlite region would be considered as the
fraction of pearlite phase. Three SEM images were selected as seen in Fig. 11. After
applying the phase classification algorithm, the amounts of pearlite phase were
calculated as 14% for the first row SEM image, 23% for the second row, and 9%
for the third row. Those pearlite phase fractions were inserted to Eq. (2), and then the
estimated hardness was able to be acquired as 132.9, 138.8, and 130.3, respectively.
Table 2 summarizes the difference between measured and estimated hardness values.
The average difference between the measured and estimated hardness values is
8.7%, so it can be said that estimating the mechanical properties with machine
learning image segmentation is a practical and feasible technique.
Some of the ways this experiment could be improved or increased in accuracy is
by having more standardized images. Some of the images were taken at slightly
different magnifications, which may result in larger or smaller pearlite and ferrite
ratio compared to the actual. Another straightforward method to increase the accu-
racy is to have more training data. Training size of perhaps 50 or 100 images will
significantly improve the accuracy, but it will also increase the time it takes to create
and train the ground truth and require more computing power. One of the problems
that was faced during the experiment was in the manual colorization of the ground
truth. As labeling each 18 training images of every pearlite and ferrite can be time-
consuming and tedious, mistakes were made such as labeling pearlite and ferrite
incorrectly or going out of the boundaries of the phases. The flaws were directly used
for training, and some of the mistakes may have appeared in the inferring process,
thus resulting in a slight decrease in accuracy.
4 Conclusion
THE SCANDAL.
THE CLAIMS.